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         int apie_h;
 292         int key_msk = test_kvm_facility(vcpu->kvm, 76);
 293         int fmt_o = crycbd_o & CRYCB_FORMAT_MASK;
 294         int fmt_h = vcpu->arch.sie_block->crycbd & CRYCB_FORMAT_MASK;
 295         int ret = 0;
 296
 297         scb_s->crycbd = 0;
 298
 299         apie_h = vcpu->arch.sie_block->eca & ECA_APIE;
 300         if (!apie_h && !key_msk)
 301                 return 0;
 302
 303         if (!crycb_addr)
 304                 return set_validity_icpt(scb_s, 0x0039U);
 305
 306         if (fmt_o == CRYCB_FORMAT1)
 307                 if ((crycb_addr & PAGE_MASK) !=
 308                     ((crycb_addr + 128) & PAGE_MASK))
 309                         return set_validity_icpt(scb_s, 0x003CU);
 310
 311         if (apie_h && (scb_o->eca & ECA_APIE)) {
 312                 ret = setup_apcb(vcpu, &vsie_page->crycb, crycb_addr,
 313                                  vcpu->kvm->arch.crypto.crycb,
 314                                  fmt_o, fmt_h);
 315                 if (ret)
 316                         goto end;
 317                 scb_s->eca |= scb_o->eca & ECA_APIE;
 318         }
 319
 320         /* we may only allow it if enabled for guest 2 */
 321         ecb3_flags = scb_o->ecb3 & vcpu->arch.sie_block->ecb3 &
 322                      (ECB3_AES | ECB3_DEA);
 323         if (!ecb3_flags)
 324                 goto end;
 325
 326         /* copy only the wrapping keys */
 327         if (read_guest_real(vcpu, crycb_addr + 72,
 328                             vsie_page->crycb.dea_wrapping_key_mask, 56))
 329                 return set_validity_icpt(scb_s, 0x0035U);
 330
 331         scb_s->ecb3 |= ecb3_flags;
 332
 333         /* xor both blocks in one run */
 334         b1 = (unsigned long *) vsie_page->crycb.dea_wrapping_key_mask;
 335         b2 = (unsigned long *)
 336                             vcpu->kvm->arch.crypto.crycb->dea_wrapping_key_mask;
 337         /* as 56%8 == 0, bitmap_xor won't overwrite any data */
 338         bitmap_xor(b1, b1, b2, BITS_PER_BYTE * 56);
 339 end:
 340         switch (ret) {
 341         case -EINVAL:
 342                 return set_validity_icpt(scb_s, 0x0020U);
 343         case -EFAULT:
 344                 return set_validity_icpt(scb_s, 0x0035U);
 345         case -EACCES:
 346                 return set_validity_icpt(scb_s, 0x003CU);
 347         }
 348         scb_s->crycbd = ((__u32)(__u64) &vsie_page->crycb) | CRYCB_FORMAT2;
 349         return 0;
 350 }
 351
 352 /* shadow (round up/down) the ibc to avoid validity icpt */
 353 static void prepare_ibc(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page)
 354 {
 355         struct kvm_s390_sie_block *scb_s = &vsie_page->scb_s;
 356         struct kvm_s390_sie_block *scb_o = vsie_page->scb_o;
 357         /* READ_ONCE does not work on bitfields - use a temporary variable */
 358         const uint32_t __new_ibc = scb_o->ibc;
 359         const uint32_t new_ibc = READ_ONCE(__new_ibc) & 0x0fffU;
 360         __u64 min_ibc = (sclp.ibc >> 16) & 0x0fffU;
 361
 362         scb_s->ibc = 0;
 363         /* ibc installed in g2 and requested for g3 */
 364         if (vcpu->kvm->arch.model.ibc && new_ibc) {
 365                 scb_s->ibc = new_ibc;
 366                 /* takte care of the minimum ibc level of the machine */
 367                 if (scb_s->ibc < min_ibc)
 368                         scb_s->ibc = min_ibc;
 369                 /* take care of the maximum ibc level set for the guest */
 370                 if (scb_s->ibc > vcpu->kvm->arch.model.ibc)
 371                         scb_s->ibc = vcpu->kvm->arch.model.ibc;
 372         }
 373 }
 374
 375 /* unshadow the scb, copying parameters back to the real scb */
 376 static void unshadow_scb(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page)
 377 {
 378         struct kvm_s390_sie_block *scb_s = &vsie_page->scb_s;
 379         struct kvm_s390_sie_block *scb_o = vsie_page->scb_o;
 380
 381         /* interception */
 382         scb_o->icptcode = scb_s->icptcode;
 383         scb_o->icptstatus = scb_s->icptstatus;
 384         scb_o->ipa = scb_s->ipa;
 385         scb_o->ipb = scb_s->ipb;
 386         scb_o->gbea = scb_s->gbea;
 387
 388         /* timer */
 389         scb_o->cputm = scb_s->cputm;
 390         scb_o->ckc = scb_s->ckc;
 391         scb_o->todpr = scb_s->todpr;
 392
 393         /* guest state */
 394         scb_o->gpsw = scb_s->gpsw;
 395         scb_o->gg14 = scb_s->gg14;
 396         scb_o->gg15 = scb_s->gg15;
 397         memcpy(scb_o->gcr, scb_s->gcr, 128);
 398         scb_o->pp = scb_s->pp;
 399
 400         /* branch prediction */
 401         if (test_kvm_facility(vcpu->kvm, 82)) {
 402                 scb_o->fpf &= ~FPF_BPBC;
 403                 scb_o->fpf |= scb_s->fpf & FPF_BPBC;
 404         }
 405
 406         /* interrupt intercept */
 407         switch (scb_s->icptcode) {
 408         case ICPT_PROGI:
 409         case ICPT_INSTPROGI:
 410         case ICPT_EXTINT:
 411                 memcpy((void *)((u64)scb_o + 0xc0),
 412                        (void *)((u64)scb_s + 0xc0), 0xf0 - 0xc0);
 413                 break;
 414         case ICPT_PARTEXEC:
 415                 /* MVPG only */
 416                 memcpy((void *)((u64)scb_o + 0xc0),
 417                        (void *)((u64)scb_s + 0xc0), 0xd0 - 0xc0);
 418                 break;
 419         }
 420
 421         if (scb_s->ihcpu != 0xffffU)
 422                 scb_o->ihcpu = scb_s->ihcpu;
 423 }
 424
 425 /*
 426  * Setup the shadow scb by copying and checking the relevant parts of the g2
 427  * provided scb.
 428  *
 429  * Returns: - 0 if the scb has been shadowed
 430  *          - > 0 if control has to be given to guest 2
 431  */
 432 static int shadow_scb(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page)
 433 {
 434         struct kvm_s390_sie_block *scb_o = vsie_page->scb_o;
 435         struct kvm_s390_sie_block *scb_s = &vsie_page->scb_s;
 436         /* READ_ONCE does not work on bitfields - use a temporary variable */
 437         const uint32_t __new_prefix = scb_o->prefix;
 438         const uint32_t new_prefix = READ_ONCE(__new_prefix);
 439         const bool wants_tx = READ_ONCE(scb_o->ecb) & ECB_TE;
 440         bool had_tx = scb_s->ecb & ECB_TE;
 441         unsigned long new_mso = 0;
 442         int rc;
 443
 444         /* make sure we don't have any leftovers when reusing the scb */
 445         scb_s->icptcode = 0;
 446         scb_s->eca = 0;
 447         scb_s->ecb = 0;
 448         scb_s->ecb2 = 0;
 449         scb_s->ecb3 = 0;
 450         scb_s->ecd = 0;
 451         scb_s->fac = 0;
 452         scb_s->fpf = 0;
 453
 454         rc = prepare_cpuflags(vcpu, vsie_page);
 455         if (rc)
 456                 goto out;
 457
 458         /* timer */
 459         scb_s->cputm = scb_o->cputm;
 460         scb_s->ckc = scb_o->ckc;
 461         scb_s->todpr = scb_o->todpr;
 462         scb_s->epoch = scb_o->epoch;
 463
 464         /* guest state */
 465         scb_s->gpsw = scb_o->gpsw;
 466         scb_s->gg14 = scb_o->gg14;
 467         scb_s->gg15 = scb_o->gg15;
 468         memcpy(scb_s->gcr, scb_o->gcr, 128);
 469         scb_s->pp = scb_o->pp;
 470
 471         /* interception / execution handling */
 472         scb_s->gbea = scb_o->gbea;
 473         scb_s->lctl = scb_o->lctl;
 474         scb_s->svcc = scb_o->svcc;
 475         scb_s->ictl = scb_o->ictl;
 476         /*
 477          * SKEY handling functions can't deal with false setting of PTE invalid
 478          * bits. Therefore we cannot provide interpretation and would later
 479          * have to provide own emulation handlers.
 480          */
 481         if (!(atomic_read(&scb_s->cpuflags) & CPUSTAT_KSS))
 482                 scb_s->ictl |= ICTL_ISKE | ICTL_SSKE | ICTL_RRBE;
 483
 484         scb_s->icpua = scb_o->icpua;
 485
 486         if (!(atomic_read(&scb_s->cpuflags) & CPUSTAT_SM))
 487                 new_mso = READ_ONCE(scb_o->mso) & 0xfffffffffff00000UL;
 488         /* if the hva of the prefix changes, we have to remap the prefix */
 489         if (scb_s->mso != new_mso || scb_s->prefix != new_prefix)
 490                 prefix_unmapped(vsie_page);
 491          /* SIE will do mso/msl validity and exception checks for us */
 492         scb_s->msl = scb_o->msl & 0xfffffffffff00000UL;
 493         scb_s->mso = new_mso;
 494         scb_s->prefix = new_prefix;
 495
 496         /* We have to definetly flush the tlb if this scb never ran */
 497         if (scb_s->ihcpu != 0xffffU)
 498                 scb_s->ihcpu = scb_o->ihcpu;
 499
 500         /* MVPG and Protection Exception Interpretation are always available */
 501         scb_s->eca |= scb_o->eca & (ECA_MVPGI | ECA_PROTEXCI);
 502         /* Host-protection-interruption introduced with ESOP */
 503         if (test_kvm_cpu_feat(vcpu->kvm, KVM_S390_VM_CPU_FEAT_ESOP))
 504                 scb_s->ecb |= scb_o->ecb & ECB_HOSTPROTINT;
 505         /* transactional execution */
 506         if (test_kvm_facility(vcpu->kvm, 73) && wants_tx) {
 507                 /* remap the prefix is tx is toggled on */
 508                 if (!had_tx)
 509                         prefix_unmapped(vsie_page);
 510                 scb_s->ecb |= ECB_TE;
 511         }
 512         /* branch prediction */
 513         if (test_kvm_facility(vcpu->kvm, 82))
 514                 scb_s->fpf |= scb_o->fpf & FPF_BPBC;
 515         /* SIMD */
 516         if (test_kvm_facility(vcpu->kvm, 129)) {
 517                 scb_s->eca |= scb_o->eca & ECA_VX;
 518                 scb_s->ecd |= scb_o->ecd & ECD_HOSTREGMGMT;
 519         }
 520         /* Run-time-Instrumentation */
 521         if (test_kvm_facility(vcpu->kvm, 64))
 522                 scb_s->ecb3 |= scb_o->ecb3 & ECB3_RI;
 523         /* Instruction Execution Prevention */
 524         if (test_kvm_facility(vcpu->kvm, 130))
 525                 scb_s->ecb2 |= scb_o->ecb2 & ECB2_IEP;
 526         /* Guarded Storage */
 527         if (test_kvm_facility(vcpu->kvm, 133)) {
 528                 scb_s->ecb |= scb_o->ecb & ECB_GS;
 529                 scb_s->ecd |= scb_o->ecd & ECD_HOSTREGMGMT;
 530         }
 531         if (test_kvm_cpu_feat(vcpu->kvm, KVM_S390_VM_CPU_FEAT_SIIF))
 532                 scb_s->eca |= scb_o->eca & ECA_SII;
 533         if (test_kvm_cpu_feat(vcpu->kvm, KVM_S390_VM_CPU_FEAT_IB))
 534                 scb_s->eca |= scb_o->eca & ECA_IB;
 535         if (test_kvm_cpu_feat(vcpu->kvm, KVM_S390_VM_CPU_FEAT_CEI))
 536                 scb_s->eca |= scb_o->eca & ECA_CEI;
 537         /* Epoch Extension */
 538         if (test_kvm_facility(vcpu->kvm, 139))
 539                 scb_s->ecd |= scb_o->ecd & ECD_MEF;
 540
 541         /* etoken */
 542         if (test_kvm_facility(vcpu->kvm, 156))
 543                 scb_s->ecd |= scb_o->ecd & ECD_ETOKENF;
 544
 545         scb_s->hpid = HPID_VSIE;
 546
 547         prepare_ibc(vcpu, vsie_page);
 548         rc = shadow_crycb(vcpu, vsie_page);
 549 out:
 550         if (rc)
 551                 unshadow_scb(vcpu, vsie_page);
 552         return rc;
 553 }
 554
 555 void kvm_s390_vsie_gmap_notifier(struct gmap *gmap, unsigned long start,
 556                                  unsigned long end)
 557 {
 558         struct kvm *kvm = gmap->private;
 559         struct vsie_page *cur;
 560         unsigned long prefix;
 561         struct page *page;
 562         int i;
 563
 564         if (!gmap_is_shadow(gmap))
 565                 return;
 566         if (start >= 1UL << 31)
 567                 /* We are only interested in prefix pages */
 568                 return;
 569
 570         /*
 571          * Only new shadow blocks are added to the list during runtime,
 572          * therefore we can safely reference them all the time.
 573          */
 574         for (i = 0; i < kvm->arch.vsie.page_count; i++) {
 575                 page = READ_ONCE(kvm->arch.vsie.pages[i]);
 576                 if (!page)
 577                         continue;
 578                 cur = page_to_virt(page);
 579                 if (READ_ONCE(cur->gmap) != gmap)
 580                         continue;
 581                 prefix = cur->scb_s.prefix << GUEST_PREFIX_SHIFT;
 582                 /* with mso/msl, the prefix lies at an offset */
 583                 prefix += cur->scb_s.mso;
 584                 if (prefix <= end && start <= prefix + 2 * PAGE_SIZE - 1)
 585                         prefix_unmapped_sync(cur);
 586         }
 587 }
 588
 589 /*
 590  * Map the first prefix page and if tx is enabled also the second prefix page.
 591  *
 592  * The prefix will be protected, a gmap notifier will inform about unmaps.
 593  * The shadow scb must not be executed until the prefix is remapped, this is
 594  * guaranteed by properly handling PROG_REQUEST.
 595  *
 596  * Returns: - 0 on if successfully mapped or already mapped
 597  *          - > 0 if control has to be given to guest 2
 598  *          - -EAGAIN if the caller can retry immediately
 599  *          - -ENOMEM if out of memory
 600  */
 601 static int map_prefix(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page)
 602 {
 603         struct kvm_s390_sie_block *scb_s = &vsie_page->scb_s;
 604         u64 prefix = scb_s->prefix << GUEST_PREFIX_SHIFT;
 605         int rc;
 606
 607         if (prefix_is_mapped(vsie_page))
 608                 return 0;
 609
 610         /* mark it as mapped so we can catch any concurrent unmappers */
 611         prefix_mapped(vsie_page);
 612
 613         /* with mso/msl, the prefix lies at offset *mso* */
 614         prefix += scb_s->mso;
 615
 616         rc = kvm_s390_shadow_fault(vcpu, vsie_page->gmap, prefix);
 617         if (!rc && (scb_s->ecb & ECB_TE))
 618                 rc = kvm_s390_shadow_fault(vcpu, vsie_page->gmap,
 619                                            prefix + PAGE_SIZE);
 620         /*
 621          * We don't have to mprotect, we will be called for all unshadows.
 622          * SIE will detect if protection applies and trigger a validity.
 623          */
 624         if (rc)
 625                 prefix_unmapped(vsie_page);
 626         if (rc > 0 || rc == -EFAULT)
 627                 rc = set_validity_icpt(scb_s, 0x0037U);
 628         return rc;
 629 }
 630
 631 /*
 632  * Pin the guest page given by gpa and set hpa to the pinned host address.
 633  * Will always be pinned writable.
 634  *
 635  * Returns: - 0 on success
 636  *          - -EINVAL if the gpa is not valid guest storage
 637  */
 638 static int pin_guest_page(struct kvm *kvm, gpa_t gpa, hpa_t *hpa)
 639 {
 640         struct page *page;
 641
 642         page = gfn_to_page(kvm, gpa_to_gfn(gpa));
 643         if (is_error_page(page))
 644                 return -EINVAL;
 645         *hpa = (hpa_t) page_to_virt(page) + (gpa & ~PAGE_MASK);
 646         return 0;
 647 }
 648
 649 /* Unpins a page previously pinned via pin_guest_page, marking it as dirty. */
 650 static void unpin_guest_page(struct kvm *kvm, gpa_t gpa, hpa_t hpa)
 651 {
 652         kvm_release_pfn_dirty(hpa >> PAGE_SHIFT);
 653         /* mark the page always as dirty for migration */
 654         mark_page_dirty(kvm, gpa_to_gfn(gpa));
 655 }
 656
 657 /* unpin all blocks previously pinned by pin_blocks(), marking them dirty */
 658 static void unpin_blocks(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page)
 659 {
 660         struct kvm_s390_sie_block *scb_s = &vsie_page->scb_s;
 661         hpa_t hpa;
 662
 663         hpa = (u64) scb_s->scaoh << 32 | scb_s->scaol;
 664         if (hpa) {
 665                 unpin_guest_page(vcpu->kvm, vsie_page->sca_gpa, hpa);
 666                 vsie_page->sca_gpa = 0;
 667                 scb_s->scaol = 0;
 668                 scb_s->scaoh = 0;
 669         }
 670
 671         hpa = scb_s->itdba;
 672         if (hpa) {
 673                 unpin_guest_page(vcpu->kvm, vsie_page->itdba_gpa, hpa);
 674                 vsie_page->itdba_gpa = 0;
 675                 scb_s->itdba = 0;
 676         }
 677
 678         hpa = scb_s->gvrd;
 679         if (hpa) {
 680                 unpin_guest_page(vcpu->kvm, vsie_page->gvrd_gpa, hpa);
 681                 vsie_page->gvrd_gpa = 0;
 682                 scb_s->gvrd = 0;
 683         }
 684
 685         hpa = scb_s->riccbd;
 686         if (hpa) {
 687                 unpin_guest_page(vcpu->kvm, vsie_page->riccbd_gpa, hpa);
 688                 vsie_page->riccbd_gpa = 0;
 689                 scb_s->riccbd = 0;
 690         }
 691
 692         hpa = scb_s->sdnxo;
 693         if (hpa) {
 694                 unpin_guest_page(vcpu->kvm, vsie_page->sdnx_gpa, hpa);
 695                 vsie_page->sdnx_gpa = 0;
 696                 scb_s->sdnxo = 0;
 697         }
 698 }
 699
 700 /*
 701  * Instead of shadowing some blocks, we can simply forward them because the
 702  * addresses in the scb are 64 bit long.
 703  *
 704  * This works as long as the data lies in one page. If blocks ever exceed one
 705  * page, we have to fall back to shadowing.
 706  *
 707  * As we reuse the sca, the vcpu pointers contained in it are invalid. We must
 708  * therefore not enable any facilities that access these pointers (e.g. SIGPIF).
 709  *
 710  * Returns: - 0 if all blocks were pinned.
 711  *          - > 0 if control has to be given to guest 2
 712  *          - -ENOMEM if out of memory
 713  */
 714 static int pin_blocks(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page)
 715 {
 716         struct kvm_s390_sie_block *scb_o = vsie_page->scb_o;
 717         struct kvm_s390_sie_block *scb_s = &vsie_page->scb_s;
 718         hpa_t hpa;
 719         gpa_t gpa;
 720         int rc = 0;
 721
 722         gpa = READ_ONCE(scb_o->scaol) & ~0xfUL;
 723         if (test_kvm_cpu_feat(vcpu->kvm, KVM_S390_VM_CPU_FEAT_64BSCAO))
 724                 gpa |= (u64) READ_ONCE(scb_o->scaoh) << 32;
 725         if (gpa) {
 726                 if (gpa < 2 * PAGE_SIZE)
 727                         rc = set_validity_icpt(scb_s, 0x0038U);
 728                 else if ((gpa & ~0x1fffUL) == kvm_s390_get_prefix(vcpu))
 729                         rc = set_validity_icpt(scb_s, 0x0011U);
 730                 else if ((gpa & PAGE_MASK) !=
 731                          ((gpa + sizeof(struct bsca_block) - 1) & PAGE_MASK))
 732                         rc = set_validity_icpt(scb_s, 0x003bU);
 733                 if (!rc) {
 734                         rc = pin_guest_page(vcpu->kvm, gpa, &hpa);
 735                         if (rc)
 736                                 rc = set_validity_icpt(scb_s, 0x0034U);
 737                 }
 738                 if (rc)
 739                         goto unpin;
 740                 vsie_page->sca_gpa = gpa;
 741                 scb_s->scaoh = (u32)((u64)hpa >> 32);
 742                 scb_s->scaol = (u32)(u64)hpa;
 743         }
 744
 745         gpa = READ_ONCE(scb_o->itdba) & ~0xffUL;
 746         if (gpa && (scb_s->ecb & ECB_TE)) {
 747                 if (gpa < 2 * PAGE_SIZE) {
 748                         rc = set_validity_icpt(scb_s, 0x0080U);
 749                         goto unpin;
 750                 }
 751                 /* 256 bytes cannot cross page boundaries */
 752                 rc = pin_guest_page(vcpu->kvm, gpa, &hpa);
 753                 if (rc) {
 754                         rc = set_validity_icpt(scb_s, 0x0080U);
 755                         goto unpin;
 756                 }
 757                 vsie_page->itdba_gpa = gpa;
 758                 scb_s->itdba = hpa;
 759         }
 760
 761         gpa = READ_ONCE(scb_o->gvrd) & ~0x1ffUL;
 762         if (gpa && (scb_s->eca & ECA_VX) && !(scb_s->ecd & ECD_HOSTREGMGMT)) {
 763                 if (gpa < 2 * PAGE_SIZE) {
 764                         rc = set_validity_icpt(scb_s, 0x1310U);
 765                         goto unpin;
 766                 }
 767                 /*
 768                  * 512 bytes vector registers cannot cross page boundaries
 769                  * if this block gets bigger, we have to shadow it.
 770                  */
 771                 rc = pin_guest_page(vcpu->kvm, gpa, &hpa);
 772                 if (rc) {
 773                         rc = set_validity_icpt(scb_s, 0x1310U);
 774                         goto unpin;
 775                 }
 776                 vsie_page->gvrd_gpa = gpa;
 777                 scb_s->gvrd = hpa;
 778         }
 779
 780         gpa = READ_ONCE(scb_o->riccbd) & ~0x3fUL;
 781         if (gpa && (scb_s->ecb3 & ECB3_RI)) {
 782                 if (gpa < 2 * PAGE_SIZE) {
 783                         rc = set_validity_icpt(scb_s, 0x0043U);
 784                         goto unpin;
 785                 }
 786                 /* 64 bytes cannot cross page boundaries */
 787                 rc = pin_guest_page(vcpu->kvm, gpa, &hpa);
 788                 if (rc) {
 789                         rc = set_validity_icpt(scb_s, 0x0043U);
 790                         goto unpin;
 791                 }
 792                 /* Validity 0x0044 will be checked by SIE */
 793                 vsie_page->riccbd_gpa = gpa;
 794                 scb_s->riccbd = hpa;
 795         }
 796         if (((scb_s->ecb & ECB_GS) && !(scb_s->ecd & ECD_HOSTREGMGMT)) ||
 797             (scb_s->ecd & ECD_ETOKENF)) {
 798                 unsigned long sdnxc;
 799
 800                 gpa = READ_ONCE(scb_o->sdnxo) & ~0xfUL;
 801                 sdnxc = READ_ONCE(scb_o->sdnxo) & 0xfUL;
 802                 if (!gpa || gpa < 2 * PAGE_SIZE) {
 803                         rc = set_validity_icpt(scb_s, 0x10b0U);
 804                         goto unpin;
 805                 }
 806                 if (sdnxc < 6 || sdnxc > 12) {
 807                         rc = set_validity_icpt(scb_s, 0x10b1U);
 808                         goto unpin;
 809                 }
 810                 if (gpa & ((1 << sdnxc) - 1)) {
 811                         rc = set_validity_icpt(scb_s, 0x10b2U);
 812                         goto unpin;
 813                 }
 814                 /* Due to alignment rules (checked above) this cannot
 815                  * cross page boundaries
 816                  */
 817                 rc = pin_guest_page(vcpu->kvm, gpa, &hpa);
 818                 if (rc) {
 819                         rc = set_validity_icpt(scb_s, 0x10b0U);
 820                         goto unpin;
 821                 }
 822                 vsie_page->sdnx_gpa = gpa;
 823                 scb_s->sdnxo = hpa | sdnxc;
 824         }
 825         return 0;
 826 unpin:
 827         unpin_blocks(vcpu, vsie_page);
 828         return rc;
 829 }
 830
 831 /* unpin the scb provided by guest 2, marking it as dirty */
 832 static void unpin_scb(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page,
 833                       gpa_t gpa)
 834 {
 835         hpa_t hpa = (hpa_t) vsie_page->scb_o;
 836
 837         if (hpa)
 838                 unpin_guest_page(vcpu->kvm, gpa, hpa);
 839         vsie_page->scb_o = NULL;
 840 }
 841
 842 /*
 843  * Pin the scb at gpa provided by guest 2 at vsie_page->scb_o.
 844  *
 845  * Returns: - 0 if the scb was pinned.
 846  *          - > 0 if control has to be given to guest 2
 847  */
 848 static int pin_scb(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page,
 849                    gpa_t gpa)
 850 {
 851         hpa_t hpa;
 852         int rc;
 853
 854         rc = pin_guest_page(vcpu->kvm, gpa, &hpa);
 855         if (rc) {
 856                 rc = kvm_s390_inject_program_int(vcpu, PGM_ADDRESSING);
 857                 WARN_ON_ONCE(rc);
 858                 return 1;
 859         }
 860         vsie_page->scb_o = (struct kvm_s390_sie_block *) hpa;
 861         return 0;
 862 }
 863
 864 /*
 865  * Inject a fault into guest 2.
 866  *
 867  * Returns: - > 0 if control has to be given to guest 2
 868  *            < 0 if an error occurred during injection.
 869  */
 870 static int inject_fault(struct kvm_vcpu *vcpu, __u16 code, __u64 vaddr,
 871                         bool write_flag)
 872 {
 873         struct kvm_s390_pgm_info pgm = {
 874                 .code = code,
 875                 .trans_exc_code =
 876                         /* 0-51: virtual address */
 877                         (vaddr & 0xfffffffffffff000UL) |
 878                         /* 52-53: store / fetch */
 879                         (((unsigned int) !write_flag) + 1) << 10,
 880                         /* 62-63: asce id (alway primary == 0) */
 881                 .exc_access_id = 0, /* always primary */
 882                 .op_access_id = 0, /* not MVPG */
 883         };
 884         int rc;
 885
 886         if (code == PGM_PROTECTION)
 887                 pgm.trans_exc_code |= 0x4UL;
 888
 889         rc = kvm_s390_inject_prog_irq(vcpu, &pgm);
 890         return rc ? rc : 1;
 891 }
 892
 893 /*
 894  * Handle a fault during vsie execution on a gmap shadow.
 895  *
 896  * Returns: - 0 if the fault was resolved
 897  *          - > 0 if control has to be given to guest 2
 898  *          - < 0 if an error occurred
 899  */
 900 static int handle_fault(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page)
 901 {
 902         int rc;
 903
 904         if (current->thread.gmap_int_code == PGM_PROTECTION)
 905                 /* we can directly forward all protection exceptions */
 906                 return inject_fault(vcpu, PGM_PROTECTION,
 907                                     current->thread.gmap_addr, 1);
 908
 909         rc = kvm_s390_shadow_fault(vcpu, vsie_page->gmap,
 910                                    current->thread.gmap_addr);
 911         if (rc > 0) {
 912                 rc = inject_fault(vcpu, rc,
 913                                   current->thread.gmap_addr,
 914                                   current->thread.gmap_write_flag);
 915                 if (rc >= 0)
 916                         vsie_page->fault_addr = current->thread.gmap_addr;
 917         }
 918         return rc;
 919 }
 920
 921 /*
 922  * Retry the previous fault that required guest 2 intervention. This avoids
 923  * one superfluous SIE re-entry and direct exit.
 924  *
 925  * Will ignore any errors. The next SIE fault will do proper fault handling.
 926  */
 927 static void handle_last_fault(struct kvm_vcpu *vcpu,
 928                               struct vsie_page *vsie_page)
 929 {
 930         if (vsie_page->fault_addr)
 931                 kvm_s390_shadow_fault(vcpu, vsie_page->gmap,
 932                                       vsie_page->fault_addr);
 933         vsie_page->fault_addr = 0;
 934 }
 935
 936 static inline void clear_vsie_icpt(struct vsie_page *vsie_page)
 937 {
 938         vsie_page->scb_s.icptcode = 0;
 939 }
 940
 941 /* rewind the psw and clear the vsie icpt, so we can retry execution */
 942 static void retry_vsie_icpt(struct vsie_page *vsie_page)
 943 {
 944         struct kvm_s390_sie_block *scb_s = &vsie_page->scb_s;
 945         int ilen = insn_length(scb_s->ipa >> 8);
 946
 947         /* take care of EXECUTE instructions */
 948         if (scb_s->icptstatus & 1) {
 949                 ilen = (scb_s->icptstatus >> 4) & 0x6;
 950                 if (!ilen)
 951                         ilen = 4;
 952         }
 953         scb_s->gpsw.addr = __rewind_psw(scb_s->gpsw, ilen);
 954         clear_vsie_icpt(vsie_page);
 955 }
 956
 957 /*
 958  * Try to shadow + enable the guest 2 provided facility list.
 959  * Retry instruction execution if enabled for and provided by guest 2.
 960  *
 961  * Returns: - 0 if handled (retry or guest 2 icpt)
 962  *          - > 0 if control has to be given to guest 2
 963  */
 964 static int handle_stfle(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page)
 965 {
 966         struct kvm_s390_sie_block *scb_s = &vsie_page->scb_s;
 967         __u32 fac = READ_ONCE(vsie_page->scb_o->fac) & 0x7ffffff8U;
 968
 969         if (fac && test_kvm_facility(vcpu->kvm, 7)) {
 970                 retry_vsie_icpt(vsie_page);
 971                 if (read_guest_real(vcpu, fac, &vsie_page->fac,
 972                                     sizeof(vsie_page->fac)))
 973                         return set_validity_icpt(scb_s, 0x1090U);
 974                 scb_s->fac = (__u32)(__u64) &vsie_page->fac;
 975         }
 976         return 0;
 977 }
 978
 979 /*
 980  * Run the vsie on a shadow scb and a shadow gmap, without any further
 981  * sanity checks, handling SIE faults.
 982  *
 983  * Returns: - 0 everything went fine
 984  *          - > 0 if control has to be given to guest 2
 985  *          - < 0 if an error occurred
 986  */
 987 static int do_vsie_run(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page)
 988         __releases(vcpu->kvm->srcu)
 989         __acquires(vcpu->kvm->srcu)
 990 {
 991         struct kvm_s390_sie_block *scb_s = &vsie_page->scb_s;
 992         struct kvm_s390_sie_block *scb_o = vsie_page->scb_o;
 993         int guest_bp_isolation;
 994         int rc = 0;
 995
 996         handle_last_fault(vcpu, vsie_page);
 997
 998         if (need_resched())
 999                 schedule();
1000         if (test_cpu_flag(CIF_MCCK_PENDING))
1001                 s390_handle_mcck();
1002
1003         srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
1004
1005         /* save current guest state of bp isolation override */
1006         guest_bp_isolation = test_thread_flag(TIF_ISOLATE_BP_GUEST);
1007
1008         /*
1009          * The guest is running with BPBC, so we have to force it on for our
1010          * nested guest. This is done by enabling BPBC globally, so the BPBC
1011          * control in the SCB (which the nested guest can modify) is simply
1012          * ignored.
1013          */
1014         if (test_kvm_facility(vcpu->kvm, 82) &&
1015             vcpu->arch.sie_block->fpf & FPF_BPBC)
1016                 set_thread_flag(TIF_ISOLATE_BP_GUEST);
1017
1018         local_irq_disable();
1019         guest_enter_irqoff();
1020         local_irq_enable();
1021
1022         /*
1023          * Simulate a SIE entry of the VCPU (see sie64a), so VCPU blocking
1024          * and VCPU requests also hinder the vSIE from running and lead
1025          * to an immediate exit. kvm_s390_vsie_kick() has to be used to
1026          * also kick the vSIE.
1027          */
1028         vcpu->arch.sie_block->prog0c |= PROG_IN_SIE;
1029         barrier();
1030         if (!kvm_s390_vcpu_sie_inhibited(vcpu))
1031                 rc = sie64a(scb_s, vcpu->run->s.regs.gprs);
1032         barrier();
1033         vcpu->arch.sie_block->prog0c &= ~PROG_IN_SIE;
1034
1035         local_irq_disable();
1036         guest_exit_irqoff();
1037         local_irq_enable();
1038
1039         /* restore guest state for bp isolation override */
1040         if (!guest_bp_isolation)
1041                 clear_thread_flag(TIF_ISOLATE_BP_GUEST);
1042
1043         vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
1044
1045         if (rc == -EINTR) {
1046                 VCPU_EVENT(vcpu, 3, "%s", "machine check");
1047                 kvm_s390_reinject_machine_check(vcpu, &vsie_page->mcck_info);
1048                 return 0;
1049         }
1050
1051         if (rc > 0)
1052                 rc = 0; /* we could still have an icpt */
1053         else if (rc == -EFAULT)
1054                 return handle_fault(vcpu, vsie_page);
1055
1056         switch (scb_s->icptcode) {
1057         case ICPT_INST:
1058                 if (scb_s->ipa == 0xb2b0)
1059                         rc = handle_stfle(vcpu, vsie_page);
1060                 break;
1061         case ICPT_STOP:
1062                 /* stop not requested by g2 - must have been a kick */
1063                 if (!(atomic_read(&scb_o->cpuflags) & CPUSTAT_STOP_INT))
1064                         clear_vsie_icpt(vsie_page);
1065                 break;
1066         case ICPT_VALIDITY:
1067                 if ((scb_s->ipa & 0xf000) != 0xf000)
1068                         scb_s->ipa += 0x1000;
1069                 break;
1070         }
1071         return rc;
1072 }
1073
1074 static void release_gmap_shadow(struct vsie_page *vsie_page)
1075 {
1076         if (vsie_page->gmap)
1077                 gmap_put(vsie_page->gmap);
1078         WRITE_ONCE(vsie_page->gmap, NULL);
1079         prefix_unmapped(vsie_page);
1080 }
1081
1082 static int acquire_gmap_shadow(struct kvm_vcpu *vcpu,
1083                                struct vsie_page *vsie_page)
1084 {
1085         unsigned long asce;
1086         union ctlreg0 cr0;
1087         struct gmap *gmap;
1088         int edat;
1089
1090         asce = vcpu->arch.sie_block->gcr[1];
1091         cr0.val = vcpu->arch.sie_block->gcr[0];
1092         edat = cr0.edat && test_kvm_facility(vcpu->kvm, 8);
1093         edat += edat && test_kvm_facility(vcpu->kvm, 78);
1094
1095         /*
1096          * ASCE or EDAT could have changed since last icpt, or the gmap
1097          * we're holding has been unshadowed. If the gmap is still valid,
1098          * we can safely reuse it.
1099          */
1100         if (vsie_page->gmap && gmap_shadow_valid(vsie_page->gmap, asce, edat))
1101                 return 0;
1102
1103         /* release the old shadow - if any, and mark the prefix as unmapped */
1104         release_gmap_shadow(vsie_page);
1105         gmap = gmap_shadow(vcpu->arch.gmap, asce, edat);
1106         if (IS_ERR(gmap))
1107                 return PTR_ERR(gmap);
1108         gmap->private = vcpu->kvm;
1109         WRITE_ONCE(vsie_page->gmap, gmap);
1110         return 0;
1111 }
1112
1113 /*
1114  * Register the shadow scb at the VCPU, e.g. for kicking out of vsie.
1115  */
1116 static void register_shadow_scb(struct kvm_vcpu *vcpu,
1117                                 struct vsie_page *vsie_page)
1118 {
1119         struct kvm_s390_sie_block *scb_s = &vsie_page->scb_s;
1120
1121         WRITE_ONCE(vcpu->arch.vsie_block, &vsie_page->scb_s);
1122         /*
1123          * External calls have to lead to a kick of the vcpu and
1124          * therefore the vsie -> Simulate Wait state.
1125          */
1126         kvm_s390_set_cpuflags(vcpu, CPUSTAT_WAIT);
1127         /*
1128          * We have to adjust the g3 epoch by the g2 epoch. The epoch will
1129          * automatically be adjusted on tod clock changes via kvm_sync_clock.
1130          */
1131         preempt_disable();
1132         scb_s->epoch += vcpu->kvm->arch.epoch;
1133
1134         if (scb_s->ecd & ECD_MEF) {
1135                 scb_s->epdx += vcpu->kvm->arch.epdx;
1136                 if (scb_s->epoch < vcpu->kvm->arch.epoch)
1137                         scb_s->epdx += 1;
1138         }
1139
1140         preempt_enable();
1141 }
1142
1143 /*
1144  * Unregister a shadow scb from a VCPU.
1145  */
1146 static void unregister_shadow_scb(struct kvm_vcpu *vcpu)
1147 {
1148         kvm_s390_clear_cpuflags(vcpu, CPUSTAT_WAIT);
1149         WRITE_ONCE(vcpu->arch.vsie_block, NULL);
1150 }
1151
1152 /*
1153  * Run the vsie on a shadowed scb, managing the gmap shadow, handling
1154  * prefix pages and faults.
1155  *
1156  * Returns: - 0 if no errors occurred
1157  *          - > 0 if control has to be given to guest 2
1158  *          - -ENOMEM if out of memory
1159  */
1160 static int vsie_run(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page)
1161 {
1162         struct kvm_s390_sie_block *scb_s = &vsie_page->scb_s;
1163         int rc = 0;
1164
1165         while (1) {
1166                 rc = acquire_gmap_shadow(vcpu, vsie_page);
1167                 if (!rc)
1168                         rc = map_prefix(vcpu, vsie_page);
1169                 if (!rc) {
1170                         gmap_enable(vsie_page->gmap);
1171                         update_intervention_requests(vsie_page);
1172                         rc = do_vsie_run(vcpu, vsie_page);
1173                         gmap_enable(vcpu->arch.gmap);
1174                 }
1175                 atomic_andnot(PROG_BLOCK_SIE, &scb_s->prog20);
1176
1177                 if (rc == -EAGAIN)
1178                         rc = 0;
1179                 if (rc || scb_s->icptcode || signal_pending(current) ||
1180                     kvm_s390_vcpu_has_irq(vcpu, 0) ||
1181                     kvm_s390_vcpu_sie_inhibited(vcpu))
1182                         break;
1183         }
1184
1185         if (rc == -EFAULT) {
1186                 /*
1187                  * Addressing exceptions are always presentes as intercepts.
1188                  * As addressing exceptions are suppressing and our guest 3 PSW
1189                  * points at the responsible instruction, we have to
1190                  * forward the PSW and set the ilc. If we can't read guest 3
1191                  * instruction, we can use an arbitrary ilc. Let's always use
1192                  * ilen = 4 for now, so we can avoid reading in guest 3 virtual
1193                  * memory. (we could also fake the shadow so the hardware
1194                  * handles it).
1195                  */
1196                 scb_s->icptcode = ICPT_PROGI;
1197                 scb_s->iprcc = PGM_ADDRESSING;
1198                 scb_s->pgmilc = 4;
1199                 scb_s->gpsw.addr = __rewind_psw(scb_s->gpsw, 4);
1200         }
1201         return rc;
1202 }
1203
1204 /*
1205  * Get or create a vsie page for a scb address.
1206  *
1207  * Returns: - address of a vsie page (cached or new one)
1208  *          - NULL if the same scb address is already used by another VCPU
1209  *          - ERR_PTR(-ENOMEM) if out of memory
1210  */
1211 static struct vsie_page *get_vsie_page(struct kvm *kvm, unsigned long addr)
1212 {
1213         struct vsie_page *vsie_page;
1214         struct page *page;
1215         int nr_vcpus;
1216
1217         rcu_read_lock();
1218         page = radix_tree_lookup(&kvm->arch.vsie.addr_to_page, addr >> 9);
1219         rcu_read_unlock();
1220         if (page) {
1221                 if (page_ref_inc_return(page) == 2)
1222                         return page_to_virt(page);
1223                 page_ref_dec(page);
1224         }
1225
1226         /*
1227          * We want at least #online_vcpus shadows, so every VCPU can execute
1228          * the VSIE in parallel.
1229          */
1230         nr_vcpus = atomic_read(&kvm->online_vcpus);
1231
1232         mutex_lock(&kvm->arch.vsie.mutex);
1233         if (kvm->arch.vsie.page_count < nr_vcpus) {
1234                 page = alloc_page(GFP_KERNEL | __GFP_ZERO | GFP_DMA);
1235                 if (!page) {
1236                         mutex_unlock(&kvm->arch.vsie.mutex);
1237                         return ERR_PTR(-ENOMEM);
1238                 }
1239                 page_ref_inc(page);
1240                 kvm->arch.vsie.pages[kvm->arch.vsie.page_count] = page;
1241                 kvm->arch.vsie.page_count++;
1242         } else {
1243                 /* reuse an existing entry that belongs to nobody */
1244                 while (true) {
1245                         page = kvm->arch.vsie.pages[kvm->arch.vsie.next];
1246                         if (page_ref_inc_return(page) == 2)
1247                                 break;
1248                         page_ref_dec(page);
1249                         kvm->arch.vsie.next++;
1250                         kvm->arch.vsie.next %= nr_vcpus;
1251                 }
1252                 radix_tree_delete(&kvm->arch.vsie.addr_to_page, page->index >> 9);
1253         }
1254         page->index = addr;
1255         /* double use of the same address */
1256         if (radix_tree_insert(&kvm->arch.vsie.addr_to_page, addr >> 9, page)) {
1257                 page_ref_dec(page);
1258                 mutex_unlock(&kvm->arch.vsie.mutex);
1259                 return NULL;
1260         }
1261         mutex_unlock(&kvm->arch.vsie.mutex);
1262
1263         vsie_page = page_to_virt(page);
1264         memset(&vsie_page->scb_s, 0, sizeof(struct kvm_s390_sie_block));
1265         release_gmap_shadow(vsie_page);
1266         vsie_page->fault_addr = 0;
1267         vsie_page->scb_s.ihcpu = 0xffffU;
1268         return vsie_page;
1269 }
1270
1271 /* put a vsie page acquired via get_vsie_page */
1272 static void put_vsie_page(struct kvm *kvm, struct vsie_page *vsie_page)
1273 {
1274         struct page *page = pfn_to_page(__pa(vsie_page) >> PAGE_SHIFT);
1275
1276         page_ref_dec(page);
1277 }
1278
1279 int kvm_s390_handle_vsie(struct kvm_vcpu *vcpu)
1280 {
1281         struct vsie_page *vsie_page;
1282         unsigned long scb_addr;
1283         int rc;
1284
1285         vcpu->stat.instruction_sie++;
1286         if (!test_kvm_cpu_feat(vcpu->kvm, KVM_S390_VM_CPU_FEAT_SIEF2))
1287                 return -EOPNOTSUPP;
1288         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
1289                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
1290
1291         BUILD_BUG_ON(sizeof(struct vsie_page) != PAGE_SIZE);
1292         scb_addr = kvm_s390_get_base_disp_s(vcpu, NULL);
1293
1294         /* 512 byte alignment */
1295         if (unlikely(scb_addr & 0x1ffUL))
1296                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
1297
1298         if (signal_pending(current) || kvm_s390_vcpu_has_irq(vcpu, 0) ||
1299             kvm_s390_vcpu_sie_inhibited(vcpu))
1300                 return 0;
1301
1302         vsie_page = get_vsie_page(vcpu->kvm, scb_addr);
1303         if (IS_ERR(vsie_page))
1304                 return PTR_ERR(vsie_page);
1305         else if (!vsie_page)
1306                 /* double use of sie control block - simply do nothing */
1307                 return 0;
1308
1309         rc = pin_scb(vcpu, vsie_page, scb_addr);
1310         if (rc)
1311                 goto out_put;
1312         rc = shadow_scb(vcpu, vsie_page);
1313         if (rc)
1314                 goto out_unpin_scb;
1315         rc = pin_blocks(vcpu, vsie_page);
1316         if (rc)
1317                 goto out_unshadow;
1318         register_shadow_scb(vcpu, vsie_page);
1319         rc = vsie_run(vcpu, vsie_page);
1320         unregister_shadow_scb(vcpu);
1321         unpin_blocks(vcpu, vsie_page);
1322 out_unshadow:
1323         unshadow_scb(vcpu, vsie_page);
1324 out_unpin_scb:
1325         unpin_scb(vcpu, vsie_page, scb_addr);
1326 out_put:
1327         put_vsie_page(vcpu->kvm, vsie_page);
1328
1329         return rc < 0 ? rc : 0;
1330 }
1331
1332 /* Init the vsie data structures. To be called when a vm is initialized. */
1333 void kvm_s390_vsie_init(struct kvm *kvm)
1334 {
1335         mutex_init(&kvm->arch.vsie.mutex);
1336         INIT_RADIX_TREE(&kvm->arch.vsie.addr_to_page, GFP_KERNEL);
1337 }
1338
1339 /* Destroy the vsie data structures. To be called when a vm is destroyed. */
1340 void kvm_s390_vsie_destroy(struct kvm *kvm)
1341 {
1342         struct vsie_page *vsie_page;
1343         struct page *page;
1344         int i;
1345
1346         mutex_lock(&kvm->arch.vsie.mutex);
1347         for (i = 0; i < kvm->arch.vsie.page_count; i++) {
1348                 page = kvm->arch.vsie.pages[i];
1349                 kvm->arch.vsie.pages[i] = NULL;
1350                 vsie_page = page_to_virt(page);
1351                 release_gmap_shadow(vsie_page);
1352                 /* free the radix tree entry */
1353                 radix_tree_delete(&kvm->arch.vsie.addr_to_page, page->index >> 9);
1354                 __free_page(page);
1355         }
1356         kvm->arch.vsie.page_count = 0;
1357         mutex_unlock(&kvm->arch.vsie.mutex);
1358 }
1359
1360 void kvm_s390_vsie_kick(struct kvm_vcpu *vcpu)
1361 {
1362         struct kvm_s390_sie_block *scb = READ_ONCE(vcpu->arch.vsie_block);
1363
1364         /*
1365          * Even if the VCPU lets go of the shadow sie block reference, it is
1366          * still valid in the cache. So we can safely kick it.
1367          */
1368         if (scb) {
1369                 atomic_or(PROG_BLOCK_SIE, &scb->prog20);
1370                 if (scb->prog0c & PROG_IN_SIE)
1371                         atomic_or(CPUSTAT_STOP_INT, &scb->cpuflags);
1372         }
1373 }