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 #include <linux/io.h> 16 #include <linux/mman.h> 17 18 #include <asm/mmu_context.h> 19 #include <asm/sclp.h> 20 #include <asm/nmi.h> 21 #include <asm/dis.h> 22 #include <asm/facility.h> 23 #include "kvm-s390.h" 24 #include "gaccess.h" 25 #include "gmap.h" 26 27 enum vsie_page_flags { 28 VSIE_PAGE_IN_USE = 0, 29 }; 30 31 struct vsie_page { 32 struct kvm_s390_sie_block scb_s; /* 0x0000 */ 33 /* 34 * the backup info for machine check. ensure it's at 35 * the same offset as that in struct sie_page! 36 */ 37 struct mcck_volatile_info mcck_info; /* 0x0200 */ 38 /* 39 * The pinned original scb. Be aware that other VCPUs can modify 40 * it while we read from it. Values that are used for conditions or 41 * are reused conditionally, should be accessed via READ_ONCE. 42 */ 43 struct kvm_s390_sie_block *scb_o; /* 0x0218 */ 44 /* 45 * Flags: must be set/cleared atomically after the vsie page can be 46 * looked up by other CPUs. 47 */ 48 unsigned long flags; /* 0x0220 */ 49 /* address of the last reported fault to guest2 */ 50 unsigned long fault_addr; /* 0x0228 */ 51 /* calculated guest addresses of satellite control blocks */ 52 gpa_t sca_gpa; /* 0x0230 */ 53 gpa_t itdba_gpa; /* 0x0238 */ 54 gpa_t gvrd_gpa; /* 0x0240 */ 55 gpa_t riccbd_gpa; /* 0x0248 */ 56 gpa_t sdnx_gpa; /* 0x0250 */ 57 /* 58 * guest address of the original SCB. Remains set for free vsie 59 * pages, so we can properly look them up in our addr_to_page 60 * radix tree. 61 */ 62 gpa_t scb_gpa; /* 0x0258 */ 63 /* the shadow gmap in use by the vsie_page */ 64 struct gmap_cache gmap_cache; /* 0x0260 */ 65 __u8 reserved[0x0700 - 0x0278]; /* 0x0278 */ 66 struct kvm_s390_crypto_cb crycb; /* 0x0700 */ 67 __u8 fac[S390_ARCH_FAC_LIST_SIZE_BYTE]; /* 0x0800 */ 68 }; 69 70 static_assert(sizeof(struct vsie_page) == PAGE_SIZE); 71 72 /* trigger a validity icpt for the given scb */ 73 static int set_validity_icpt(struct kvm_s390_sie_block *scb, 74 __u16 reason_code) 75 { 76 scb->ipa = 0x1000; 77 scb->ipb = ((__u32) reason_code) << 16; 78 scb->icptcode = ICPT_VALIDITY; 79 return 1; 80 } 81 82 /* mark the prefix as unmapped, this will block the VSIE */ 83 static void prefix_unmapped(struct vsie_page *vsie_page) 84 { 85 atomic_or(PROG_REQUEST, &vsie_page->scb_s.prog20); 86 } 87 88 /* mark the prefix as unmapped and wait until the VSIE has been left */ 89 static void prefix_unmapped_sync(struct vsie_page *vsie_page) 90 { 91 prefix_unmapped(vsie_page); 92 if (vsie_page->scb_s.prog0c & PROG_IN_SIE) 93 atomic_or(CPUSTAT_STOP_INT, &vsie_page->scb_s.cpuflags); 94 while (vsie_page->scb_s.prog0c & PROG_IN_SIE) 95 cpu_relax(); 96 } 97 98 /* mark the prefix as mapped, this will allow the VSIE to run */ 99 static void prefix_mapped(struct vsie_page *vsie_page) 100 { 101 atomic_andnot(PROG_REQUEST, &vsie_page->scb_s.prog20); 102 } 103 104 /* test if the prefix is mapped into the gmap shadow */ 105 static int prefix_is_mapped(struct vsie_page *vsie_page) 106 { 107 return !(atomic_read(&vsie_page->scb_s.prog20) & PROG_REQUEST); 108 } 109 110 /* copy the updated intervention request bits into the shadow scb */ 111 static void update_intervention_requests(struct vsie_page *vsie_page) 112 { 113 const int bits = CPUSTAT_STOP_INT | CPUSTAT_IO_INT | CPUSTAT_EXT_INT; 114 int cpuflags; 115 116 cpuflags = atomic_read(&vsie_page->scb_o->cpuflags); 117 atomic_andnot(bits, &vsie_page->scb_s.cpuflags); 118 atomic_or(cpuflags & bits, &vsie_page->scb_s.cpuflags); 119 } 120 121 /* shadow (filter and validate) the cpuflags */ 122 static int prepare_cpuflags(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page) 123 { 124 struct kvm_s390_sie_block *scb_s = &vsie_page->scb_s; 125 struct kvm_s390_sie_block *scb_o = vsie_page->scb_o; 126 int newflags, cpuflags = atomic_read(&scb_o->cpuflags); 127 128 /* we don't allow ESA/390 guests unless explicitly enabled */ 129 if (!(cpuflags & CPUSTAT_ZARCH) && !vcpu->kvm->arch.allow_vsie_esamode) 130 return set_validity_icpt(scb_s, 0x0001U); 131 132 if (cpuflags & (CPUSTAT_RRF | CPUSTAT_MCDS)) 133 return set_validity_icpt(scb_s, 0x0001U); 134 else if (cpuflags & (CPUSTAT_SLSV | CPUSTAT_SLSR)) 135 return set_validity_icpt(scb_s, 0x0007U); 136 137 /* intervention requests will be set later */ 138 newflags = 0; 139 if (cpuflags & CPUSTAT_ZARCH) 140 newflags = CPUSTAT_ZARCH; 141 if (cpuflags & CPUSTAT_GED && test_kvm_facility(vcpu->kvm, 8)) 142 newflags |= CPUSTAT_GED; 143 if (cpuflags & CPUSTAT_GED2 && test_kvm_facility(vcpu->kvm, 78)) { 144 if (cpuflags & CPUSTAT_GED) 145 return set_validity_icpt(scb_s, 0x0001U); 146 newflags |= CPUSTAT_GED2; 147 } 148 if (test_kvm_cpu_feat(vcpu->kvm, KVM_S390_VM_CPU_FEAT_GPERE)) 149 newflags |= cpuflags & CPUSTAT_P; 150 if (test_kvm_cpu_feat(vcpu->kvm, KVM_S390_VM_CPU_FEAT_GSLS)) 151 newflags |= cpuflags & CPUSTAT_SM; 152 if (test_kvm_cpu_feat(vcpu->kvm, KVM_S390_VM_CPU_FEAT_IBS)) 153 newflags |= cpuflags & CPUSTAT_IBS; 154 if (test_kvm_cpu_feat(vcpu->kvm, KVM_S390_VM_CPU_FEAT_KSS)) 155 newflags |= cpuflags & CPUSTAT_KSS; 156 157 atomic_set(&scb_s->cpuflags, newflags); 158 return 0; 159 } 160 /* Copy to APCB FORMAT1 from APCB FORMAT0 */ 161 static int setup_apcb10(struct kvm_vcpu *vcpu, struct kvm_s390_apcb1 *apcb_s, 162 unsigned long crycb_gpa, struct kvm_s390_apcb1 *apcb_h) 163 { 164 struct kvm_s390_apcb0 tmp; 165 unsigned long apcb_gpa; 166 167 apcb_gpa = crycb_gpa + offsetof(struct kvm_s390_crypto_cb, apcb0); 168 169 if (read_guest_real(vcpu, apcb_gpa, &tmp, 170 sizeof(struct kvm_s390_apcb0))) 171 return -EFAULT; 172 173 apcb_s->apm[0] = apcb_h->apm[0] & tmp.apm[0]; 174 apcb_s->aqm[0] = apcb_h->aqm[0] & tmp.aqm[0] & 0xffff000000000000UL; 175 apcb_s->adm[0] = apcb_h->adm[0] & tmp.adm[0] & 0xffff000000000000UL; 176 177 return 0; 178 179 } 180 181 /** 182 * setup_apcb00 - Copy to APCB FORMAT0 from APCB FORMAT0 183 * @vcpu: pointer to the virtual CPU 184 * @apcb_s: pointer to start of apcb in the shadow crycb 185 * @crycb_gpa: guest physical address to start of original guest crycb 186 * @apcb_h: pointer to start of apcb in the guest1 187 * 188 * Returns 0 and -EFAULT on error reading guest apcb 189 */ 190 static int setup_apcb00(struct kvm_vcpu *vcpu, unsigned long *apcb_s, 191 unsigned long crycb_gpa, unsigned long *apcb_h) 192 { 193 unsigned long apcb_gpa; 194 195 apcb_gpa = crycb_gpa + offsetof(struct kvm_s390_crypto_cb, apcb0); 196 197 if (read_guest_real(vcpu, apcb_gpa, apcb_s, 198 sizeof(struct kvm_s390_apcb0))) 199 return -EFAULT; 200 201 bitmap_and(apcb_s, apcb_s, apcb_h, 202 BITS_PER_BYTE * sizeof(struct kvm_s390_apcb0)); 203 204 return 0; 205 } 206 207 /** 208 * setup_apcb11 - Copy the FORMAT1 APCB from the guest to the shadow CRYCB 209 * @vcpu: pointer to the virtual CPU 210 * @apcb_s: pointer to start of apcb in the shadow crycb 211 * @crycb_gpa: guest physical address to start of original guest crycb 212 * @apcb_h: pointer to start of apcb in the host 213 * 214 * Returns 0 and -EFAULT on error reading guest apcb 215 */ 216 static int setup_apcb11(struct kvm_vcpu *vcpu, unsigned long *apcb_s, 217 unsigned long crycb_gpa, 218 unsigned long *apcb_h) 219 { 220 unsigned long apcb_gpa; 221 222 apcb_gpa = crycb_gpa + offsetof(struct kvm_s390_crypto_cb, apcb1); 223 224 if (read_guest_real(vcpu, apcb_gpa, apcb_s, 225 sizeof(struct kvm_s390_apcb1))) 226 return -EFAULT; 227 228 bitmap_and(apcb_s, apcb_s, apcb_h, 229 BITS_PER_BYTE * sizeof(struct kvm_s390_apcb1)); 230 231 return 0; 232 } 233 234 /** 235 * setup_apcb - Create a shadow copy of the apcb. 236 * @vcpu: pointer to the virtual CPU 237 * @crycb_s: pointer to shadow crycb 238 * @crycb_gpa: guest physical address of original guest crycb 239 * @crycb_h: pointer to the host crycb 240 * @fmt_o: format of the original guest crycb. 241 * @fmt_h: format of the host crycb. 242 * 243 * Checks the compatibility between the guest and host crycb and calls the 244 * appropriate copy function. 245 * 246 * Return 0 or an error number if the guest and host crycb are incompatible. 247 */ 248 static int setup_apcb(struct kvm_vcpu *vcpu, struct kvm_s390_crypto_cb *crycb_s, 249 const u32 crycb_gpa, 250 struct kvm_s390_crypto_cb *crycb_h, 251 int fmt_o, int fmt_h) 252 { 253 switch (fmt_o) { 254 case CRYCB_FORMAT2: 255 if ((crycb_gpa & PAGE_MASK) != ((crycb_gpa + 256) & PAGE_MASK)) 256 return -EACCES; 257 if (fmt_h != CRYCB_FORMAT2) 258 return -EINVAL; 259 return setup_apcb11(vcpu, (unsigned long *)&crycb_s->apcb1, 260 crycb_gpa, 261 (unsigned long *)&crycb_h->apcb1); 262 case CRYCB_FORMAT1: 263 switch (fmt_h) { 264 case CRYCB_FORMAT2: 265 return setup_apcb10(vcpu, &crycb_s->apcb1, 266 crycb_gpa, 267 &crycb_h->apcb1); 268 case CRYCB_FORMAT1: 269 return setup_apcb00(vcpu, 270 (unsigned long *) &crycb_s->apcb0, 271 crycb_gpa, 272 (unsigned long *) &crycb_h->apcb0); 273 } 274 break; 275 case CRYCB_FORMAT0: 276 if ((crycb_gpa & PAGE_MASK) != ((crycb_gpa + 32) & PAGE_MASK)) 277 return -EACCES; 278 279 switch (fmt_h) { 280 case CRYCB_FORMAT2: 281 return setup_apcb10(vcpu, &crycb_s->apcb1, 282 crycb_gpa, 283 &crycb_h->apcb1); 284 case CRYCB_FORMAT1: 285 case CRYCB_FORMAT0: 286 return setup_apcb00(vcpu, 287 (unsigned long *) &crycb_s->apcb0, 288 crycb_gpa, 289 (unsigned long *) &crycb_h->apcb0); 290 } 291 } 292 return -EINVAL; 293 } 294 295 /** 296 * shadow_crycb - Create a shadow copy of the crycb block 297 * @vcpu: a pointer to the virtual CPU 298 * @vsie_page: a pointer to internal date used for the vSIE 299 * 300 * Create a shadow copy of the crycb block and setup key wrapping, if 301 * requested for guest 3 and enabled for guest 2. 302 * 303 * We accept format-1 or format-2, but we convert format-1 into format-2 304 * in the shadow CRYCB. 305 * Using format-2 enables the firmware to choose the right format when 306 * scheduling the SIE. 307 * There is nothing to do for format-0. 308 * 309 * This function centralize the issuing of set_validity_icpt() for all 310 * the subfunctions working on the crycb. 311 * 312 * Returns: - 0 if shadowed or nothing to do 313 * - > 0 if control has to be given to guest 2 314 */ 315 static int shadow_crycb(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page) 316 { 317 struct kvm_s390_sie_block *scb_s = &vsie_page->scb_s; 318 struct kvm_s390_sie_block *scb_o = vsie_page->scb_o; 319 const uint32_t crycbd_o = READ_ONCE(scb_o->crycbd); 320 const u32 crycb_addr = crycbd_o & 0x7ffffff8U; 321 unsigned long *b1, *b2; 322 u8 ecb3_flags; 323 u32 ecd_flags; 324 int apie_h; 325 int apie_s; 326 int key_msk = test_kvm_facility(vcpu->kvm, 76); 327 int fmt_o = crycbd_o & CRYCB_FORMAT_MASK; 328 int fmt_h = vcpu->arch.sie_block->crycbd & CRYCB_FORMAT_MASK; 329 int ret = 0; 330 331 scb_s->crycbd = 0; 332 333 apie_h = vcpu->arch.sie_block->eca & ECA_APIE; 334 apie_s = apie_h & scb_o->eca; 335 if (!apie_s && (!key_msk || (fmt_o == CRYCB_FORMAT0))) 336 return 0; 337 338 if (!crycb_addr) 339 return set_validity_icpt(scb_s, 0x0039U); 340 341 if (fmt_o == CRYCB_FORMAT1) 342 if ((crycb_addr & PAGE_MASK) != 343 ((crycb_addr + 128) & PAGE_MASK)) 344 return set_validity_icpt(scb_s, 0x003CU); 345 346 if (apie_s) { 347 ret = setup_apcb(vcpu, &vsie_page->crycb, crycb_addr, 348 vcpu->kvm->arch.crypto.crycb, 349 fmt_o, fmt_h); 350 if (ret) 351 goto end; 352 scb_s->eca |= scb_o->eca & ECA_APIE; 353 } 354 355 /* we may only allow it if enabled for guest 2 */ 356 ecb3_flags = scb_o->ecb3 & vcpu->arch.sie_block->ecb3 & 357 (ECB3_AES | ECB3_DEA); 358 ecd_flags = scb_o->ecd & vcpu->arch.sie_block->ecd & 359 (ECD_ECC | ECD_HMAC); 360 if (!ecb3_flags && !ecd_flags) 361 goto end; 362 363 /* copy only the wrapping keys */ 364 if (read_guest_real(vcpu, crycb_addr + 72, 365 vsie_page->crycb.dea_wrapping_key_mask, 56)) 366 return set_validity_icpt(scb_s, 0x0035U); 367 368 scb_s->ecb3 |= ecb3_flags; 369 scb_s->ecd |= ecd_flags; 370 371 /* xor both blocks in one run */ 372 b1 = (unsigned long *) vsie_page->crycb.dea_wrapping_key_mask; 373 b2 = (unsigned long *) 374 vcpu->kvm->arch.crypto.crycb->dea_wrapping_key_mask; 375 /* as 56%8 == 0, bitmap_xor won't overwrite any data */ 376 bitmap_xor(b1, b1, b2, BITS_PER_BYTE * 56); 377 end: 378 switch (ret) { 379 case -EINVAL: 380 return set_validity_icpt(scb_s, 0x0022U); 381 case -EFAULT: 382 return set_validity_icpt(scb_s, 0x0035U); 383 case -EACCES: 384 return set_validity_icpt(scb_s, 0x003CU); 385 } 386 scb_s->crycbd = (u32)virt_to_phys(&vsie_page->crycb) | CRYCB_FORMAT2; 387 return 0; 388 } 389 390 static void shadow_esa(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page) 391 { 392 struct kvm_s390_sie_block *scb_s = &vsie_page->scb_s; 393 394 /* Ensure these bits are indeed turned off */ 395 scb_s->eca &= ~ECA_VX; 396 scb_s->ecb &= ~(ECB_GS | ECB_TE); 397 scb_s->ecb3 &= ~ECB3_RI; 398 scb_s->ecd &= ~ECD_HOSTREGMGMT; 399 } 400 401 /* shadow (round up/down) the ibc to avoid validity icpt */ 402 static void prepare_ibc(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page) 403 { 404 struct kvm_s390_sie_block *scb_s = &vsie_page->scb_s; 405 struct kvm_s390_sie_block *scb_o = vsie_page->scb_o; 406 /* READ_ONCE does not work on bitfields - use a temporary variable */ 407 const uint32_t __new_ibc = scb_o->ibc; 408 const uint32_t new_ibc = READ_ONCE(__new_ibc) & 0x0fffU; 409 __u64 min_ibc = (sclp.ibc >> 16) & 0x0fffU; 410 411 scb_s->ibc = 0; 412 /* ibc installed in g2 and requested for g3 */ 413 if (vcpu->kvm->arch.model.ibc && new_ibc) { 414 scb_s->ibc = new_ibc; 415 /* takte care of the minimum ibc level of the machine */ 416 if (scb_s->ibc < min_ibc) 417 scb_s->ibc = min_ibc; 418 /* take care of the maximum ibc level set for the guest */ 419 if (scb_s->ibc > vcpu->kvm->arch.model.ibc) 420 scb_s->ibc = vcpu->kvm->arch.model.ibc; 421 } 422 } 423 424 /* unshadow the scb, copying parameters back to the real scb */ 425 static void unshadow_scb(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page) 426 { 427 struct kvm_s390_sie_block *scb_s = &vsie_page->scb_s; 428 struct kvm_s390_sie_block *scb_o = vsie_page->scb_o; 429 430 /* interception */ 431 scb_o->icptcode = scb_s->icptcode; 432 scb_o->icptstatus = scb_s->icptstatus; 433 scb_o->ipa = scb_s->ipa; 434 scb_o->ipb = scb_s->ipb; 435 scb_o->gbea = scb_s->gbea; 436 437 /* timer */ 438 scb_o->cputm = scb_s->cputm; 439 scb_o->ckc = scb_s->ckc; 440 scb_o->todpr = scb_s->todpr; 441 442 /* guest state */ 443 scb_o->gpsw = scb_s->gpsw; 444 scb_o->gg14 = scb_s->gg14; 445 scb_o->gg15 = scb_s->gg15; 446 memcpy(scb_o->gcr, scb_s->gcr, 128); 447 scb_o->pp = scb_s->pp; 448 449 /* branch prediction */ 450 if (test_kvm_facility(vcpu->kvm, 82)) { 451 scb_o->fpf &= ~FPF_BPBC; 452 scb_o->fpf |= scb_s->fpf & FPF_BPBC; 453 } 454 455 /* interrupt intercept */ 456 switch (scb_s->icptcode) { 457 case ICPT_PROGI: 458 case ICPT_INSTPROGI: 459 case ICPT_EXTINT: 460 memcpy((void *)((u64)scb_o + 0xc0), 461 (void *)((u64)scb_s + 0xc0), 0xf0 - 0xc0); 462 break; 463 } 464 465 if (scb_s->ihcpu != 0xffffU) 466 scb_o->ihcpu = scb_s->ihcpu; 467 } 468 469 /* 470 * Setup the shadow scb by copying and checking the relevant parts of the g2 471 * provided scb. 472 * 473 * Returns: - 0 if the scb has been shadowed 474 * - > 0 if control has to be given to guest 2 475 */ 476 static int shadow_scb(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page) 477 { 478 struct kvm_s390_sie_block *scb_o = vsie_page->scb_o; 479 struct kvm_s390_sie_block *scb_s = &vsie_page->scb_s; 480 /* READ_ONCE does not work on bitfields - use a temporary variable */ 481 const uint32_t __new_prefix = scb_o->prefix; 482 uint32_t new_prefix = READ_ONCE(__new_prefix); 483 const bool wants_tx = READ_ONCE(scb_o->ecb) & ECB_TE; 484 bool had_tx = scb_s->ecb & ECB_TE; 485 unsigned long new_mso = 0; 486 int rc; 487 488 /* make sure we don't have any leftovers when reusing the scb */ 489 scb_s->icptcode = 0; 490 scb_s->eca = 0; 491 scb_s->ecb = 0; 492 scb_s->ecb2 = 0; 493 scb_s->ecb3 = 0; 494 scb_s->ecd = 0; 495 scb_s->fac = 0; 496 scb_s->fpf = 0; 497 498 rc = prepare_cpuflags(vcpu, vsie_page); 499 if (rc) 500 goto out; 501 502 /* timer */ 503 scb_s->cputm = scb_o->cputm; 504 scb_s->ckc = scb_o->ckc; 505 scb_s->todpr = scb_o->todpr; 506 scb_s->epoch = scb_o->epoch; 507 508 /* guest state */ 509 scb_s->gpsw = scb_o->gpsw; 510 scb_s->gg14 = scb_o->gg14; 511 scb_s->gg15 = scb_o->gg15; 512 memcpy(scb_s->gcr, scb_o->gcr, 128); 513 scb_s->pp = scb_o->pp; 514 515 /* interception / execution handling */ 516 scb_s->gbea = scb_o->gbea; 517 scb_s->lctl = scb_o->lctl; 518 scb_s->svcc = scb_o->svcc; 519 scb_s->ictl = scb_o->ictl; 520 /* 521 * SKEY handling functions can't deal with false setting of PTE invalid 522 * bits. Therefore we cannot provide interpretation and would later 523 * have to provide own emulation handlers. 524 */ 525 if (!(atomic_read(&scb_s->cpuflags) & CPUSTAT_KSS)) 526 scb_s->ictl |= ICTL_ISKE | ICTL_SSKE | ICTL_RRBE; 527 528 scb_s->icpua = scb_o->icpua; 529 530 if (!(atomic_read(&scb_s->cpuflags) & CPUSTAT_ZARCH)) 531 new_prefix &= GUEST_PREFIX_MASK_ESA; 532 else 533 new_prefix &= GUEST_PREFIX_MASK_ZARCH; 534 535 if (!(atomic_read(&scb_s->cpuflags) & CPUSTAT_SM)) 536 new_mso = READ_ONCE(scb_o->mso) & 0xfffffffffff00000UL; 537 /* if the hva of the prefix changes, we have to remap the prefix */ 538 if (scb_s->mso != new_mso || scb_s->prefix != new_prefix) 539 prefix_unmapped(vsie_page); 540 /* SIE will do mso/msl validity and exception checks for us */ 541 scb_s->msl = scb_o->msl & 0xfffffffffff00000UL; 542 scb_s->mso = new_mso; 543 scb_s->prefix = new_prefix; 544 545 /* We have to definitely flush the tlb if this scb never ran */ 546 if (scb_s->ihcpu != 0xffffU) 547 scb_s->ihcpu = scb_o->ihcpu; 548 549 /* MVPG and Protection Exception Interpretation are always available */ 550 scb_s->eca |= scb_o->eca & (ECA_MVPGI | ECA_PROTEXCI); 551 /* Host-protection-interruption introduced with ESOP */ 552 if (test_kvm_cpu_feat(vcpu->kvm, KVM_S390_VM_CPU_FEAT_ESOP)) 553 scb_s->ecb |= scb_o->ecb & ECB_HOSTPROTINT; 554 /* 555 * CPU Topology 556 * This facility only uses the utility field of the SCA and none of 557 * the cpu entries that are problematic with the other interpretation 558 * facilities so we can pass it through 559 */ 560 if (test_kvm_facility(vcpu->kvm, 11)) 561 scb_s->ecb |= scb_o->ecb & ECB_PTF; 562 /* transactional execution */ 563 if (test_kvm_facility(vcpu->kvm, 73) && wants_tx) { 564 /* remap the prefix is tx is toggled on */ 565 if (!had_tx) 566 prefix_unmapped(vsie_page); 567 scb_s->ecb |= ECB_TE; 568 } 569 /* specification exception interpretation */ 570 scb_s->ecb |= scb_o->ecb & ECB_SPECI; 571 /* branch prediction */ 572 if (test_kvm_facility(vcpu->kvm, 82)) 573 scb_s->fpf |= scb_o->fpf & FPF_BPBC; 574 /* SIMD */ 575 if (test_kvm_facility(vcpu->kvm, 129)) { 576 scb_s->eca |= scb_o->eca & ECA_VX; 577 scb_s->ecd |= scb_o->ecd & ECD_HOSTREGMGMT; 578 } 579 /* Run-time-Instrumentation */ 580 if (test_kvm_facility(vcpu->kvm, 64)) 581 scb_s->ecb3 |= scb_o->ecb3 & ECB3_RI; 582 /* Instruction Execution Prevention */ 583 if (test_kvm_facility(vcpu->kvm, 130)) 584 scb_s->ecb2 |= scb_o->ecb2 & ECB2_IEP; 585 /* Guarded Storage */ 586 if (test_kvm_facility(vcpu->kvm, 133)) { 587 scb_s->ecb |= scb_o->ecb & ECB_GS; 588 scb_s->ecd |= scb_o->ecd & ECD_HOSTREGMGMT; 589 } 590 if (test_kvm_cpu_feat(vcpu->kvm, KVM_S390_VM_CPU_FEAT_SIIF)) 591 scb_s->eca |= scb_o->eca & ECA_SII; 592 if (test_kvm_cpu_feat(vcpu->kvm, KVM_S390_VM_CPU_FEAT_IB)) 593 scb_s->eca |= scb_o->eca & ECA_IB; 594 if (test_kvm_cpu_feat(vcpu->kvm, KVM_S390_VM_CPU_FEAT_CEI)) 595 scb_s->eca |= scb_o->eca & ECA_CEI; 596 /* Epoch Extension */ 597 if (test_kvm_facility(vcpu->kvm, 139)) { 598 scb_s->ecd |= scb_o->ecd & ECD_MEF; 599 scb_s->epdx = scb_o->epdx; 600 } 601 602 /* etoken */ 603 if (test_kvm_facility(vcpu->kvm, 156)) 604 scb_s->ecd |= scb_o->ecd & ECD_ETOKENF; 605 606 scb_s->hpid = HPID_VSIE; 607 scb_s->cpnc = scb_o->cpnc; 608 609 if (!(atomic_read(&scb_s->cpuflags) & CPUSTAT_ZARCH)) 610 shadow_esa(vcpu, vsie_page); 611 612 prepare_ibc(vcpu, vsie_page); 613 rc = shadow_crycb(vcpu, vsie_page); 614 out: 615 if (rc) 616 unshadow_scb(vcpu, vsie_page); 617 return rc; 618 } 619 620 void kvm_s390_vsie_gmap_notifier(struct gmap *gmap, gpa_t start, gpa_t end) 621 { 622 struct vsie_page *cur, *next; 623 unsigned long prefix; 624 625 KVM_BUG_ON(!test_bit(GMAP_FLAG_SHADOW, &gmap->flags), gmap->kvm); 626 /* 627 * Only new shadow blocks are added to the list during runtime, 628 * therefore we can safely reference them all the time. 629 */ 630 list_for_each_entry_safe(cur, next, &gmap->scb_users, gmap_cache.list) { 631 prefix = cur->scb_s.prefix << GUEST_PREFIX_SHIFT; 632 /* with mso/msl, the prefix lies at an offset */ 633 prefix += cur->scb_s.mso; 634 if (prefix <= end && start <= prefix + 2 * PAGE_SIZE - 1) 635 prefix_unmapped_sync(cur); 636 } 637 } 638 639 /* 640 * Map the first prefix page and if tx is enabled also the second prefix page. 641 * 642 * The prefix will be protected, a gmap notifier will inform about unmaps. 643 * The shadow scb must not be executed until the prefix is remapped, this is 644 * guaranteed by properly handling PROG_REQUEST. 645 * 646 * Returns: - 0 on if successfully mapped or already mapped 647 * - > 0 if control has to be given to guest 2 648 * - -EAGAIN if the caller can retry immediately 649 * - -ENOMEM if out of memory 650 */ 651 static int map_prefix(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page, struct gmap *sg) 652 { 653 struct kvm_s390_sie_block *scb_s = &vsie_page->scb_s; 654 u64 prefix = scb_s->prefix << GUEST_PREFIX_SHIFT; 655 int rc; 656 657 if (prefix_is_mapped(vsie_page)) 658 return 0; 659 660 /* mark it as mapped so we can catch any concurrent unmappers */ 661 prefix_mapped(vsie_page); 662 663 /* with mso/msl, the prefix lies at offset *mso* */ 664 prefix += scb_s->mso; 665 666 rc = gaccess_shadow_fault(vcpu, sg, prefix, NULL, true); 667 if (!rc && (scb_s->ecb & ECB_TE)) 668 rc = gaccess_shadow_fault(vcpu, sg, prefix + PAGE_SIZE, NULL, true); 669 /* 670 * We don't have to mprotect, we will be called for all unshadows. 671 * SIE will detect if protection applies and trigger a validity. 672 */ 673 if (rc) 674 prefix_unmapped(vsie_page); 675 if (rc > 0 || rc == -EFAULT) 676 rc = set_validity_icpt(scb_s, 0x0037U); 677 return rc; 678 } 679 680 /* 681 * Pin the guest page given by gpa and set hpa to the pinned host address. 682 * Will always be pinned writable. 683 * 684 * Returns: - 0 on success 685 * - -EINVAL if the gpa is not valid guest storage 686 */ 687 static int pin_guest_page(struct kvm *kvm, gpa_t gpa, hpa_t *hpa) 688 { 689 struct page *page; 690 691 page = gfn_to_page(kvm, gpa_to_gfn(gpa)); 692 if (!page) 693 return -EINVAL; 694 *hpa = (hpa_t)page_to_phys(page) + (gpa & ~PAGE_MASK); 695 return 0; 696 } 697 698 /* Unpins a page previously pinned via pin_guest_page, marking it as dirty. */ 699 static void unpin_guest_page(struct kvm *kvm, gpa_t gpa, hpa_t hpa) 700 { 701 kvm_release_page_dirty(pfn_to_page(hpa >> PAGE_SHIFT)); 702 /* mark the page always as dirty for migration */ 703 mark_page_dirty(kvm, gpa_to_gfn(gpa)); 704 } 705 706 /* unpin all blocks previously pinned by pin_blocks(), marking them dirty */ 707 static void unpin_blocks(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page) 708 { 709 struct kvm_s390_sie_block *scb_s = &vsie_page->scb_s; 710 hpa_t hpa; 711 712 hpa = (u64) scb_s->scaoh << 32 | scb_s->scaol; 713 if (hpa) { 714 unpin_guest_page(vcpu->kvm, vsie_page->sca_gpa, hpa); 715 vsie_page->sca_gpa = 0; 716 scb_s->scaol = 0; 717 scb_s->scaoh = 0; 718 } 719 720 hpa = scb_s->itdba; 721 if (hpa) { 722 unpin_guest_page(vcpu->kvm, vsie_page->itdba_gpa, hpa); 723 vsie_page->itdba_gpa = 0; 724 scb_s->itdba = 0; 725 } 726 727 hpa = scb_s->gvrd; 728 if (hpa) { 729 unpin_guest_page(vcpu->kvm, vsie_page->gvrd_gpa, hpa); 730 vsie_page->gvrd_gpa = 0; 731 scb_s->gvrd = 0; 732 } 733 734 hpa = scb_s->riccbd; 735 if (hpa) { 736 unpin_guest_page(vcpu->kvm, vsie_page->riccbd_gpa, hpa); 737 vsie_page->riccbd_gpa = 0; 738 scb_s->riccbd = 0; 739 } 740 741 hpa = scb_s->sdnxo; 742 if (hpa) { 743 unpin_guest_page(vcpu->kvm, vsie_page->sdnx_gpa, hpa); 744 vsie_page->sdnx_gpa = 0; 745 scb_s->sdnxo = 0; 746 } 747 } 748 749 /* 750 * Instead of shadowing some blocks, we can simply forward them because the 751 * addresses in the scb are 64 bit long. 752 * 753 * This works as long as the data lies in one page. If blocks ever exceed one 754 * page, we have to fall back to shadowing. 755 * 756 * As we reuse the sca, the vcpu pointers contained in it are invalid. We must 757 * therefore not enable any facilities that access these pointers (e.g. SIGPIF). 758 * 759 * Returns: - 0 if all blocks were pinned. 760 * - > 0 if control has to be given to guest 2 761 * - -ENOMEM if out of memory 762 */ 763 static int pin_blocks(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page) 764 { 765 struct kvm_s390_sie_block *scb_o = vsie_page->scb_o; 766 struct kvm_s390_sie_block *scb_s = &vsie_page->scb_s; 767 hpa_t hpa; 768 gpa_t gpa; 769 int rc = 0; 770 771 gpa = READ_ONCE(scb_o->scaol) & ~0xfUL; 772 if (test_kvm_cpu_feat(vcpu->kvm, KVM_S390_VM_CPU_FEAT_64BSCAO)) 773 gpa |= (u64) READ_ONCE(scb_o->scaoh) << 32; 774 if (gpa) { 775 if (gpa < 2 * PAGE_SIZE) 776 rc = set_validity_icpt(scb_s, 0x0038U); 777 else if ((gpa & ~0x1fffUL) == kvm_s390_get_prefix(vcpu)) 778 rc = set_validity_icpt(scb_s, 0x0011U); 779 else if ((gpa & PAGE_MASK) != 780 ((gpa + offsetof(struct bsca_block, cpu[0]) - 1) & PAGE_MASK)) 781 rc = set_validity_icpt(scb_s, 0x003bU); 782 if (!rc) { 783 rc = pin_guest_page(vcpu->kvm, gpa, &hpa); 784 if (rc) 785 rc = set_validity_icpt(scb_s, 0x0034U); 786 } 787 if (rc) 788 goto unpin; 789 vsie_page->sca_gpa = gpa; 790 scb_s->scaoh = (u32)((u64)hpa >> 32); 791 scb_s->scaol = (u32)(u64)hpa; 792 } 793 794 gpa = READ_ONCE(scb_o->itdba) & ~0xffUL; 795 if (gpa && (scb_s->ecb & ECB_TE)) { 796 if (gpa < 2 * PAGE_SIZE) { 797 rc = set_validity_icpt(scb_s, 0x0080U); 798 goto unpin; 799 } 800 /* 256 bytes cannot cross page boundaries */ 801 rc = pin_guest_page(vcpu->kvm, gpa, &hpa); 802 if (rc) { 803 rc = set_validity_icpt(scb_s, 0x0080U); 804 goto unpin; 805 } 806 vsie_page->itdba_gpa = gpa; 807 scb_s->itdba = hpa; 808 } 809 810 gpa = READ_ONCE(scb_o->gvrd) & ~0x1ffUL; 811 if (gpa && (scb_s->eca & ECA_VX) && !(scb_s->ecd & ECD_HOSTREGMGMT)) { 812 if (gpa < 2 * PAGE_SIZE) { 813 rc = set_validity_icpt(scb_s, 0x1310U); 814 goto unpin; 815 } 816 /* 817 * 512 bytes vector registers cannot cross page boundaries 818 * if this block gets bigger, we have to shadow it. 819 */ 820 rc = pin_guest_page(vcpu->kvm, gpa, &hpa); 821 if (rc) { 822 rc = set_validity_icpt(scb_s, 0x1310U); 823 goto unpin; 824 } 825 vsie_page->gvrd_gpa = gpa; 826 scb_s->gvrd = hpa; 827 } 828 829 gpa = READ_ONCE(scb_o->riccbd) & ~0x3fUL; 830 if (gpa && (scb_s->ecb3 & ECB3_RI)) { 831 if (gpa < 2 * PAGE_SIZE) { 832 rc = set_validity_icpt(scb_s, 0x0043U); 833 goto unpin; 834 } 835 /* 64 bytes cannot cross page boundaries */ 836 rc = pin_guest_page(vcpu->kvm, gpa, &hpa); 837 if (rc) { 838 rc = set_validity_icpt(scb_s, 0x0043U); 839 goto unpin; 840 } 841 /* Validity 0x0044 will be checked by SIE */ 842 vsie_page->riccbd_gpa = gpa; 843 scb_s->riccbd = hpa; 844 } 845 if (((scb_s->ecb & ECB_GS) && !(scb_s->ecd & ECD_HOSTREGMGMT)) || 846 (scb_s->ecd & ECD_ETOKENF)) { 847 unsigned long sdnxc; 848 849 gpa = READ_ONCE(scb_o->sdnxo) & ~0xfUL; 850 sdnxc = READ_ONCE(scb_o->sdnxo) & 0xfUL; 851 if (!gpa || gpa < 2 * PAGE_SIZE) { 852 rc = set_validity_icpt(scb_s, 0x10b0U); 853 goto unpin; 854 } 855 if (sdnxc < 6 || sdnxc > 12) { 856 rc = set_validity_icpt(scb_s, 0x10b1U); 857 goto unpin; 858 } 859 if (gpa & ((1 << sdnxc) - 1)) { 860 rc = set_validity_icpt(scb_s, 0x10b2U); 861 goto unpin; 862 } 863 /* Due to alignment rules (checked above) this cannot 864 * cross page boundaries 865 */ 866 rc = pin_guest_page(vcpu->kvm, gpa, &hpa); 867 if (rc) { 868 rc = set_validity_icpt(scb_s, 0x10b0U); 869 goto unpin; 870 } 871 vsie_page->sdnx_gpa = gpa; 872 scb_s->sdnxo = hpa | sdnxc; 873 } 874 return 0; 875 unpin: 876 unpin_blocks(vcpu, vsie_page); 877 return rc; 878 } 879 880 /* unpin the scb provided by guest 2, marking it as dirty */ 881 static void unpin_scb(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page, 882 gpa_t gpa) 883 { 884 hpa_t hpa = virt_to_phys(vsie_page->scb_o); 885 886 if (hpa) 887 unpin_guest_page(vcpu->kvm, gpa, hpa); 888 vsie_page->scb_o = NULL; 889 } 890 891 /* 892 * Pin the scb at gpa provided by guest 2 at vsie_page->scb_o. 893 * 894 * Returns: - 0 if the scb was pinned. 895 * - > 0 if control has to be given to guest 2 896 */ 897 static int pin_scb(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page, 898 gpa_t gpa) 899 { 900 hpa_t hpa; 901 int rc; 902 903 rc = pin_guest_page(vcpu->kvm, gpa, &hpa); 904 if (rc) { 905 rc = kvm_s390_inject_program_int(vcpu, PGM_ADDRESSING); 906 WARN_ON_ONCE(rc); 907 return 1; 908 } 909 vsie_page->scb_o = phys_to_virt(hpa); 910 return 0; 911 } 912 913 /* 914 * Inject a fault into guest 2. 915 * 916 * Returns: - > 0 if control has to be given to guest 2 917 * < 0 if an error occurred during injection. 918 */ 919 static int inject_fault(struct kvm_vcpu *vcpu, __u16 code, __u64 vaddr, 920 bool write_flag) 921 { 922 struct kvm_s390_pgm_info pgm = { 923 .code = code, 924 .trans_exc_code = 925 /* 0-51: virtual address */ 926 (vaddr & 0xfffffffffffff000UL) | 927 /* 52-53: store / fetch */ 928 (((unsigned int) !write_flag) + 1) << 10, 929 /* 62-63: asce id (always primary == 0) */ 930 .exc_access_id = 0, /* always primary */ 931 .op_access_id = 0, /* not MVPG */ 932 }; 933 int rc; 934 935 if (code == PGM_PROTECTION) 936 pgm.trans_exc_code |= 0x4UL; 937 938 rc = kvm_s390_inject_prog_irq(vcpu, &pgm); 939 return rc ? rc : 1; 940 } 941 942 /* 943 * Handle a fault during vsie execution on a gmap shadow. 944 * 945 * Returns: - 0 if the fault was resolved 946 * - > 0 if control has to be given to guest 2 947 * - < 0 if an error occurred 948 */ 949 static int handle_fault(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page, struct gmap *sg) 950 { 951 bool wr = kvm_s390_cur_gmap_fault_is_write(); 952 int rc; 953 954 if ((current->thread.gmap_int_code & PGM_INT_CODE_MASK) == PGM_PROTECTION) 955 /* we can directly forward all protection exceptions */ 956 return inject_fault(vcpu, PGM_PROTECTION, 957 current->thread.gmap_teid.addr * PAGE_SIZE, 1); 958 959 rc = gaccess_shadow_fault(vcpu, sg, current->thread.gmap_teid.addr * PAGE_SIZE, NULL, wr); 960 if (rc > 0) { 961 rc = inject_fault(vcpu, rc, 962 current->thread.gmap_teid.addr * PAGE_SIZE, wr); 963 if (rc >= 0) 964 vsie_page->fault_addr = current->thread.gmap_teid.addr * PAGE_SIZE; 965 } 966 return rc; 967 } 968 969 /* 970 * Retry the previous fault that required guest 2 intervention. This avoids 971 * one superfluous SIE re-entry and direct exit. 972 * 973 * Will ignore any errors. The next SIE fault will do proper fault handling. 974 */ 975 static void handle_last_fault(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page, struct gmap *sg) 976 { 977 if (vsie_page->fault_addr) 978 gaccess_shadow_fault(vcpu, sg, vsie_page->fault_addr, NULL, true); 979 vsie_page->fault_addr = 0; 980 } 981 982 static inline void clear_vsie_icpt(struct vsie_page *vsie_page) 983 { 984 vsie_page->scb_s.icptcode = 0; 985 } 986 987 /* rewind the psw and clear the vsie icpt, so we can retry execution */ 988 static void retry_vsie_icpt(struct vsie_page *vsie_page) 989 { 990 struct kvm_s390_sie_block *scb_s = &vsie_page->scb_s; 991 int ilen = insn_length(scb_s->ipa >> 8); 992 993 /* take care of EXECUTE instructions */ 994 if (scb_s->icptstatus & 1) { 995 ilen = (scb_s->icptstatus >> 4) & 0x6; 996 if (!ilen) 997 ilen = 4; 998 } 999 scb_s->gpsw.addr = __rewind_psw(scb_s->gpsw, ilen); 1000 clear_vsie_icpt(vsie_page); 1001 } 1002 1003 /* 1004 * Try to shadow + enable the guest 2 provided facility list. 1005 * Retry instruction execution if enabled for and provided by guest 2. 1006 * 1007 * Returns: - 0 if handled (retry or guest 2 icpt) 1008 * - > 0 if control has to be given to guest 2 1009 */ 1010 static int handle_stfle(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page) 1011 { 1012 struct kvm_s390_sie_block *scb_s = &vsie_page->scb_s; 1013 __u32 fac = READ_ONCE(vsie_page->scb_o->fac); 1014 1015 /* 1016 * Alternate-STFLE-Interpretive-Execution facilities are not supported 1017 * -> format-0 flcb 1018 */ 1019 if (fac && test_kvm_facility(vcpu->kvm, 7)) { 1020 retry_vsie_icpt(vsie_page); 1021 /* 1022 * The facility list origin (FLO) is in bits 1 - 28 of the FLD 1023 * so we need to mask here before reading. 1024 */ 1025 fac = fac & 0x7ffffff8U; 1026 /* 1027 * format-0 -> size of nested guest's facility list == guest's size 1028 * guest's size == host's size, since STFLE is interpretatively executed 1029 * using a format-0 for the guest, too. 1030 */ 1031 if (read_guest_real(vcpu, fac, &vsie_page->fac, 1032 stfle_size() * sizeof(u64))) 1033 return set_validity_icpt(scb_s, 0x1090U); 1034 scb_s->fac = (u32)virt_to_phys(&vsie_page->fac); 1035 } 1036 return 0; 1037 } 1038 1039 /* 1040 * Get a register for a nested guest. 1041 * @vcpu the vcpu of the guest 1042 * @vsie_page the vsie_page for the nested guest 1043 * @reg the register number, the upper 4 bits are ignored. 1044 * returns: the value of the register. 1045 */ 1046 static u64 vsie_get_register(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page, u8 reg) 1047 { 1048 /* no need to validate the parameter and/or perform error handling */ 1049 reg &= 0xf; 1050 switch (reg) { 1051 case 15: 1052 return vsie_page->scb_s.gg15; 1053 case 14: 1054 return vsie_page->scb_s.gg14; 1055 default: 1056 return vcpu->run->s.regs.gprs[reg]; 1057 } 1058 } 1059 1060 static int vsie_handle_mvpg(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page, struct gmap *sg) 1061 { 1062 struct kvm_s390_sie_block *scb_s = &vsie_page->scb_s; 1063 unsigned long src, dest, mask, prefix; 1064 u64 *pei_block = &vsie_page->scb_o->mcic; 1065 union mvpg_pei pei_dest, pei_src; 1066 int edat, rc_dest, rc_src; 1067 union ctlreg0 cr0; 1068 1069 cr0.val = vcpu->arch.sie_block->gcr[0]; 1070 edat = cr0.edat && test_kvm_facility(vcpu->kvm, 8); 1071 mask = _kvm_s390_logical_to_effective(&scb_s->gpsw, PAGE_MASK); 1072 prefix = scb_s->prefix << GUEST_PREFIX_SHIFT; 1073 1074 dest = vsie_get_register(vcpu, vsie_page, scb_s->ipb >> 20) & mask; 1075 dest = _kvm_s390_real_to_abs(prefix, dest) + scb_s->mso; 1076 src = vsie_get_register(vcpu, vsie_page, scb_s->ipb >> 16) & mask; 1077 src = _kvm_s390_real_to_abs(prefix, src) + scb_s->mso; 1078 1079 rc_dest = gaccess_shadow_fault(vcpu, sg, dest, &pei_dest, true); 1080 rc_src = gaccess_shadow_fault(vcpu, sg, src, &pei_src, false); 1081 /* 1082 * Either everything went well, or something non-critical went wrong 1083 * e.g. because of a race. In either case, simply retry. 1084 */ 1085 if (rc_dest == -EAGAIN || rc_src == -EAGAIN || (!rc_dest && !rc_src)) { 1086 retry_vsie_icpt(vsie_page); 1087 return -EAGAIN; 1088 } 1089 /* Something more serious went wrong, propagate the error */ 1090 if (rc_dest < 0) 1091 return rc_dest; 1092 if (rc_src < 0) 1093 return rc_src; 1094 1095 /* The only possible suppressing exception: just deliver it */ 1096 if (rc_dest == PGM_TRANSLATION_SPEC || rc_src == PGM_TRANSLATION_SPEC) { 1097 clear_vsie_icpt(vsie_page); 1098 rc_dest = kvm_s390_inject_program_int(vcpu, PGM_TRANSLATION_SPEC); 1099 WARN_ON_ONCE(rc_dest); 1100 return 1; 1101 } 1102 1103 /* 1104 * Forward the PEI intercept to the guest if it was a page fault, or 1105 * also for segment and region table faults if EDAT applies. 1106 */ 1107 if (edat) { 1108 rc_dest = rc_dest == PGM_ASCE_TYPE ? rc_dest : 0; 1109 rc_src = rc_src == PGM_ASCE_TYPE ? rc_src : 0; 1110 } else { 1111 rc_dest = rc_dest != PGM_PAGE_TRANSLATION ? rc_dest : 0; 1112 rc_src = rc_src != PGM_PAGE_TRANSLATION ? rc_src : 0; 1113 } 1114 if (!rc_dest && !rc_src) { 1115 pei_block[0] = pei_dest.val; 1116 pei_block[1] = pei_src.val; 1117 return 1; 1118 } 1119 1120 retry_vsie_icpt(vsie_page); 1121 1122 /* 1123 * The host has edat, and the guest does not, or it was an ASCE type 1124 * exception. The host needs to inject the appropriate DAT interrupts 1125 * into the guest. 1126 */ 1127 if (rc_dest) 1128 return inject_fault(vcpu, rc_dest, dest, 1); 1129 return inject_fault(vcpu, rc_src, src, 0); 1130 } 1131 1132 /* 1133 * Run the vsie on a shadow scb and a shadow gmap, without any further 1134 * sanity checks, handling SIE faults. 1135 * 1136 * Returns: - 0 everything went fine 1137 * - > 0 if control has to be given to guest 2 1138 * - < 0 if an error occurred 1139 */ 1140 static int do_vsie_run(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page, struct gmap *sg) 1141 __releases(vcpu->kvm->srcu) 1142 __acquires(vcpu->kvm->srcu) 1143 { 1144 struct kvm_s390_sie_block *scb_s = &vsie_page->scb_s; 1145 struct kvm_s390_sie_block *scb_o = vsie_page->scb_o; 1146 unsigned long sie_return = SIE64_RETURN_NORMAL; 1147 int guest_bp_isolation; 1148 int rc = 0; 1149 1150 handle_last_fault(vcpu, vsie_page, sg); 1151 1152 kvm_vcpu_srcu_read_unlock(vcpu); 1153 1154 /* save current guest state of bp isolation override */ 1155 guest_bp_isolation = test_thread_flag(TIF_ISOLATE_BP_GUEST); 1156 1157 /* 1158 * The guest is running with BPBC, so we have to force it on for our 1159 * nested guest. This is done by enabling BPBC globally, so the BPBC 1160 * control in the SCB (which the nested guest can modify) is simply 1161 * ignored. 1162 */ 1163 if (test_kvm_facility(vcpu->kvm, 82) && 1164 vcpu->arch.sie_block->fpf & FPF_BPBC) 1165 set_thread_flag(TIF_ISOLATE_BP_GUEST); 1166 1167 /* 1168 * Simulate a SIE entry of the VCPU (see sie64a), so VCPU blocking 1169 * and VCPU requests also hinder the vSIE from running and lead 1170 * to an immediate exit. kvm_s390_vsie_kick() has to be used to 1171 * also kick the vSIE. 1172 */ 1173 vcpu->arch.sie_block->prog0c |= PROG_IN_SIE; 1174 current->thread.gmap_int_code = 0; 1175 barrier(); 1176 if (!kvm_s390_vcpu_sie_inhibited(vcpu)) { 1177 xfer_to_guest_mode_check: 1178 local_irq_disable(); 1179 xfer_to_guest_mode_prepare(); 1180 if (xfer_to_guest_mode_work_pending()) { 1181 local_irq_enable(); 1182 rc = kvm_xfer_to_guest_mode_handle_work(vcpu); 1183 if (rc) 1184 goto skip_sie; 1185 goto xfer_to_guest_mode_check; 1186 } 1187 guest_timing_enter_irqoff(); 1188 sie_return = kvm_s390_enter_exit_sie(scb_s, vcpu->run->s.regs.gprs, sg->asce.val); 1189 guest_timing_exit_irqoff(); 1190 local_irq_enable(); 1191 } 1192 1193 skip_sie: 1194 barrier(); 1195 vcpu->arch.sie_block->prog0c &= ~PROG_IN_SIE; 1196 1197 /* restore guest state for bp isolation override */ 1198 if (!guest_bp_isolation) 1199 clear_thread_flag(TIF_ISOLATE_BP_GUEST); 1200 1201 kvm_vcpu_srcu_read_lock(vcpu); 1202 1203 if (sie_return == SIE64_RETURN_MCCK) { 1204 kvm_s390_reinject_machine_check(vcpu, &vsie_page->mcck_info); 1205 return 0; 1206 } 1207 1208 WARN_ON_ONCE(sie_return != SIE64_RETURN_NORMAL); 1209 1210 if (rc > 0) 1211 rc = 0; /* we could still have an icpt */ 1212 else if (current->thread.gmap_int_code) 1213 return handle_fault(vcpu, vsie_page, sg); 1214 1215 switch (scb_s->icptcode) { 1216 case ICPT_INST: 1217 if (scb_s->ipa == 0xb2b0) 1218 rc = handle_stfle(vcpu, vsie_page); 1219 break; 1220 case ICPT_STOP: 1221 /* stop not requested by g2 - must have been a kick */ 1222 if (!(atomic_read(&scb_o->cpuflags) & CPUSTAT_STOP_INT)) 1223 clear_vsie_icpt(vsie_page); 1224 break; 1225 case ICPT_VALIDITY: 1226 if ((scb_s->ipa & 0xf000) != 0xf000) 1227 scb_s->ipa += 0x1000; 1228 break; 1229 case ICPT_PARTEXEC: 1230 if (scb_s->ipa == 0xb254) 1231 rc = vsie_handle_mvpg(vcpu, vsie_page, sg); 1232 break; 1233 } 1234 return rc; 1235 } 1236 1237 static void release_gmap_shadow(struct vsie_page *vsie_page) 1238 { 1239 struct gmap *gmap = vsie_page->gmap_cache.gmap; 1240 1241 lockdep_assert_held(&gmap->kvm->arch.gmap->children_lock); 1242 1243 list_del(&vsie_page->gmap_cache.list); 1244 vsie_page->gmap_cache.gmap = NULL; 1245 prefix_unmapped(vsie_page); 1246 1247 if (list_empty(&gmap->scb_users)) { 1248 gmap_remove_child(gmap); 1249 gmap_put(gmap); 1250 } 1251 } 1252 1253 static struct gmap *acquire_gmap_shadow(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page) 1254 { 1255 union ctlreg0 cr0; 1256 struct gmap *gmap; 1257 union asce asce; 1258 int edat; 1259 1260 asce.val = vcpu->arch.sie_block->gcr[1]; 1261 cr0.val = vcpu->arch.sie_block->gcr[0]; 1262 edat = cr0.edat && test_kvm_facility(vcpu->kvm, 8); 1263 edat += edat && test_kvm_facility(vcpu->kvm, 78); 1264 1265 scoped_guard(spinlock, &vcpu->kvm->arch.gmap->children_lock) { 1266 gmap = vsie_page->gmap_cache.gmap; 1267 if (gmap) { 1268 /* 1269 * ASCE or EDAT could have changed since last icpt, or the gmap 1270 * we're holding has been unshadowed. If the gmap is still valid, 1271 * we can safely reuse it. 1272 */ 1273 if (gmap_is_shadow_valid(gmap, asce, edat)) { 1274 vcpu->kvm->stat.gmap_shadow_reuse++; 1275 gmap_get(gmap); 1276 return gmap; 1277 } 1278 /* release the old shadow and mark the prefix as unmapped */ 1279 release_gmap_shadow(vsie_page); 1280 } 1281 } 1282 again: 1283 gmap = gmap_create_shadow(vcpu->arch.mc, vcpu->kvm->arch.gmap, asce, edat); 1284 if (IS_ERR(gmap)) 1285 return gmap; 1286 scoped_guard(spinlock, &vcpu->kvm->arch.gmap->children_lock) { 1287 /* unlikely race condition, remove the previous shadow */ 1288 if (vsie_page->gmap_cache.gmap) 1289 release_gmap_shadow(vsie_page); 1290 if (!gmap->parent) { 1291 gmap_put(gmap); 1292 goto again; 1293 } 1294 vcpu->kvm->stat.gmap_shadow_create++; 1295 list_add(&vsie_page->gmap_cache.list, &gmap->scb_users); 1296 vsie_page->gmap_cache.gmap = gmap; 1297 prefix_unmapped(vsie_page); 1298 } 1299 return gmap; 1300 } 1301 1302 /* 1303 * Register the shadow scb at the VCPU, e.g. for kicking out of vsie. 1304 */ 1305 static void register_shadow_scb(struct kvm_vcpu *vcpu, 1306 struct vsie_page *vsie_page) 1307 { 1308 struct kvm_s390_sie_block *scb_s = &vsie_page->scb_s; 1309 1310 WRITE_ONCE(vcpu->arch.vsie_block, &vsie_page->scb_s); 1311 /* 1312 * External calls have to lead to a kick of the vcpu and 1313 * therefore the vsie -> Simulate Wait state. 1314 */ 1315 kvm_s390_set_cpuflags(vcpu, CPUSTAT_WAIT); 1316 /* 1317 * We have to adjust the g3 epoch by the g2 epoch. The epoch will 1318 * automatically be adjusted on tod clock changes via kvm_sync_clock. 1319 */ 1320 preempt_disable(); 1321 scb_s->epoch += vcpu->kvm->arch.epoch; 1322 1323 if (scb_s->ecd & ECD_MEF) { 1324 scb_s->epdx += vcpu->kvm->arch.epdx; 1325 if (scb_s->epoch < vcpu->kvm->arch.epoch) 1326 scb_s->epdx += 1; 1327 } 1328 1329 preempt_enable(); 1330 } 1331 1332 /* 1333 * Unregister a shadow scb from a VCPU. 1334 */ 1335 static void unregister_shadow_scb(struct kvm_vcpu *vcpu) 1336 { 1337 kvm_s390_clear_cpuflags(vcpu, CPUSTAT_WAIT); 1338 WRITE_ONCE(vcpu->arch.vsie_block, NULL); 1339 } 1340 1341 /* 1342 * Run the vsie on a shadowed scb, managing the gmap shadow, handling 1343 * prefix pages and faults. 1344 * 1345 * Returns: - 0 if no errors occurred 1346 * - > 0 if control has to be given to guest 2 1347 * - -ENOMEM if out of memory 1348 */ 1349 static int vsie_run(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page) 1350 { 1351 struct kvm_s390_sie_block *scb_s = &vsie_page->scb_s; 1352 struct gmap *sg = NULL; 1353 int rc = 0; 1354 1355 while (1) { 1356 sg = acquire_gmap_shadow(vcpu, vsie_page); 1357 if (IS_ERR(sg)) { 1358 rc = PTR_ERR(sg); 1359 sg = NULL; 1360 } 1361 if (!rc) 1362 rc = map_prefix(vcpu, vsie_page, sg); 1363 if (!rc) { 1364 update_intervention_requests(vsie_page); 1365 rc = do_vsie_run(vcpu, vsie_page, sg); 1366 } 1367 atomic_andnot(PROG_BLOCK_SIE, &scb_s->prog20); 1368 1369 if (rc == -EAGAIN) 1370 rc = 0; 1371 1372 /* 1373 * Exit the loop if the guest needs to process the intercept 1374 */ 1375 if (rc || scb_s->icptcode) 1376 break; 1377 1378 /* 1379 * Exit the loop if the host needs to process an intercept, 1380 * but rewind the PSW to re-enter SIE once that's completed 1381 * instead of passing a "no action" intercept to the guest. 1382 */ 1383 if (kvm_s390_vcpu_has_irq(vcpu, 0) || 1384 kvm_s390_vcpu_sie_inhibited(vcpu)) { 1385 kvm_s390_rewind_psw(vcpu, 4); 1386 break; 1387 } 1388 if (sg) 1389 sg = gmap_put(sg); 1390 cond_resched(); 1391 } 1392 if (sg) 1393 sg = gmap_put(sg); 1394 1395 if (rc == -EFAULT) { 1396 /* 1397 * Addressing exceptions are always presentes as intercepts. 1398 * As addressing exceptions are suppressing and our guest 3 PSW 1399 * points at the responsible instruction, we have to 1400 * forward the PSW and set the ilc. If we can't read guest 3 1401 * instruction, we can use an arbitrary ilc. Let's always use 1402 * ilen = 4 for now, so we can avoid reading in guest 3 virtual 1403 * memory. (we could also fake the shadow so the hardware 1404 * handles it). 1405 */ 1406 scb_s->icptcode = ICPT_PROGI; 1407 scb_s->iprcc = PGM_ADDRESSING; 1408 scb_s->pgmilc = 4; 1409 scb_s->gpsw.addr = __rewind_psw(scb_s->gpsw, 4); 1410 rc = 1; 1411 } 1412 return rc; 1413 } 1414 1415 /* Try getting a given vsie page, returning "true" on success. */ 1416 static inline bool try_get_vsie_page(struct vsie_page *vsie_page) 1417 { 1418 if (test_bit(VSIE_PAGE_IN_USE, &vsie_page->flags)) 1419 return false; 1420 return !test_and_set_bit(VSIE_PAGE_IN_USE, &vsie_page->flags); 1421 } 1422 1423 /* Put a vsie page acquired through get_vsie_page / try_get_vsie_page. */ 1424 static void put_vsie_page(struct vsie_page *vsie_page) 1425 { 1426 clear_bit(VSIE_PAGE_IN_USE, &vsie_page->flags); 1427 } 1428 1429 /* 1430 * Get or create a vsie page for a scb address. 1431 * 1432 * Returns: - address of a vsie page (cached or new one) 1433 * - NULL if the same scb address is already used by another VCPU 1434 * - ERR_PTR(-ENOMEM) if out of memory 1435 */ 1436 static struct vsie_page *get_vsie_page(struct kvm *kvm, unsigned long addr) 1437 { 1438 struct vsie_page *vsie_page; 1439 int nr_vcpus; 1440 1441 rcu_read_lock(); 1442 vsie_page = radix_tree_lookup(&kvm->arch.vsie.addr_to_page, addr >> 9); 1443 rcu_read_unlock(); 1444 if (vsie_page) { 1445 if (try_get_vsie_page(vsie_page)) { 1446 if (vsie_page->scb_gpa == addr) 1447 return vsie_page; 1448 /* 1449 * We raced with someone reusing + putting this vsie 1450 * page before we grabbed it. 1451 */ 1452 put_vsie_page(vsie_page); 1453 } 1454 } 1455 1456 /* 1457 * We want at least #online_vcpus shadows, so every VCPU can execute 1458 * the VSIE in parallel. 1459 */ 1460 nr_vcpus = atomic_read(&kvm->online_vcpus); 1461 1462 mutex_lock(&kvm->arch.vsie.mutex); 1463 if (kvm->arch.vsie.page_count < nr_vcpus) { 1464 vsie_page = (void *)__get_free_page(GFP_KERNEL_ACCOUNT | __GFP_ZERO | GFP_DMA); 1465 if (!vsie_page) { 1466 mutex_unlock(&kvm->arch.vsie.mutex); 1467 return ERR_PTR(-ENOMEM); 1468 } 1469 __set_bit(VSIE_PAGE_IN_USE, &vsie_page->flags); 1470 kvm->arch.vsie.pages[kvm->arch.vsie.page_count] = vsie_page; 1471 kvm->arch.vsie.page_count++; 1472 } else { 1473 /* reuse an existing entry that belongs to nobody */ 1474 while (true) { 1475 vsie_page = kvm->arch.vsie.pages[kvm->arch.vsie.next]; 1476 if (try_get_vsie_page(vsie_page)) 1477 break; 1478 kvm->arch.vsie.next++; 1479 kvm->arch.vsie.next %= nr_vcpus; 1480 } 1481 if (vsie_page->scb_gpa != ULONG_MAX) 1482 radix_tree_delete(&kvm->arch.vsie.addr_to_page, 1483 vsie_page->scb_gpa >> 9); 1484 } 1485 /* Mark it as invalid until it resides in the tree. */ 1486 vsie_page->scb_gpa = ULONG_MAX; 1487 1488 /* Double use of the same address or allocation failure. */ 1489 if (radix_tree_insert(&kvm->arch.vsie.addr_to_page, addr >> 9, vsie_page)) { 1490 put_vsie_page(vsie_page); 1491 mutex_unlock(&kvm->arch.vsie.mutex); 1492 return NULL; 1493 } 1494 vsie_page->scb_gpa = addr; 1495 mutex_unlock(&kvm->arch.vsie.mutex); 1496 1497 memset(&vsie_page->scb_s, 0, sizeof(struct kvm_s390_sie_block)); 1498 if (vsie_page->gmap_cache.gmap) { 1499 scoped_guard(spinlock, &kvm->arch.gmap->children_lock) 1500 if (vsie_page->gmap_cache.gmap) 1501 release_gmap_shadow(vsie_page); 1502 } 1503 prefix_unmapped(vsie_page); 1504 vsie_page->fault_addr = 0; 1505 vsie_page->scb_s.ihcpu = 0xffffU; 1506 return vsie_page; 1507 } 1508 1509 int kvm_s390_handle_vsie(struct kvm_vcpu *vcpu) 1510 { 1511 struct vsie_page *vsie_page; 1512 unsigned long scb_addr; 1513 int rc; 1514 1515 vcpu->stat.instruction_sie++; 1516 if (!test_kvm_cpu_feat(vcpu->kvm, KVM_S390_VM_CPU_FEAT_SIEF2)) 1517 return -EOPNOTSUPP; 1518 if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE) 1519 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP); 1520 1521 BUILD_BUG_ON(sizeof(struct vsie_page) != PAGE_SIZE); 1522 scb_addr = kvm_s390_get_base_disp_s(vcpu, NULL); 1523 1524 /* 512 byte alignment */ 1525 if (unlikely(scb_addr & 0x1ffUL)) 1526 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION); 1527 1528 if (kvm_s390_vcpu_has_irq(vcpu, 0) || kvm_s390_vcpu_sie_inhibited(vcpu)) { 1529 kvm_s390_rewind_psw(vcpu, 4); 1530 return 0; 1531 } 1532 1533 vsie_page = get_vsie_page(vcpu->kvm, scb_addr); 1534 if (IS_ERR(vsie_page)) { 1535 return PTR_ERR(vsie_page); 1536 } else if (!vsie_page) { 1537 /* double use of sie control block - simply do nothing */ 1538 kvm_s390_rewind_psw(vcpu, 4); 1539 return 0; 1540 } 1541 1542 rc = pin_scb(vcpu, vsie_page, scb_addr); 1543 if (rc) 1544 goto out_put; 1545 rc = shadow_scb(vcpu, vsie_page); 1546 if (rc) 1547 goto out_unpin_scb; 1548 rc = pin_blocks(vcpu, vsie_page); 1549 if (rc) 1550 goto out_unshadow; 1551 register_shadow_scb(vcpu, vsie_page); 1552 rc = vsie_run(vcpu, vsie_page); 1553 unregister_shadow_scb(vcpu); 1554 unpin_blocks(vcpu, vsie_page); 1555 out_unshadow: 1556 unshadow_scb(vcpu, vsie_page); 1557 out_unpin_scb: 1558 unpin_scb(vcpu, vsie_page, scb_addr); 1559 out_put: 1560 put_vsie_page(vsie_page); 1561 1562 return rc < 0 ? rc : 0; 1563 } 1564 1565 /* Init the vsie data structures. To be called when a vm is initialized. */ 1566 void kvm_s390_vsie_init(struct kvm *kvm) 1567 { 1568 mutex_init(&kvm->arch.vsie.mutex); 1569 INIT_RADIX_TREE(&kvm->arch.vsie.addr_to_page, GFP_KERNEL_ACCOUNT); 1570 } 1571 1572 /* Destroy the vsie data structures. To be called when a vm is destroyed. */ 1573 void kvm_s390_vsie_destroy(struct kvm *kvm) 1574 { 1575 struct vsie_page *vsie_page; 1576 int i; 1577 1578 mutex_lock(&kvm->arch.vsie.mutex); 1579 for (i = 0; i < kvm->arch.vsie.page_count; i++) { 1580 vsie_page = kvm->arch.vsie.pages[i]; 1581 scoped_guard(spinlock, &kvm->arch.gmap->children_lock) 1582 if (vsie_page->gmap_cache.gmap) 1583 release_gmap_shadow(vsie_page); 1584 kvm->arch.vsie.pages[i] = NULL; 1585 /* free the radix tree entry */ 1586 if (vsie_page->scb_gpa != ULONG_MAX) 1587 radix_tree_delete(&kvm->arch.vsie.addr_to_page, 1588 vsie_page->scb_gpa >> 9); 1589 free_page((unsigned long)vsie_page); 1590 } 1591 kvm->arch.vsie.page_count = 0; 1592 mutex_unlock(&kvm->arch.vsie.mutex); 1593 } 1594 1595 void kvm_s390_vsie_kick(struct kvm_vcpu *vcpu) 1596 { 1597 struct kvm_s390_sie_block *scb = READ_ONCE(vcpu->arch.vsie_block); 1598 1599 /* 1600 * Even if the VCPU lets go of the shadow sie block reference, it is 1601 * still valid in the cache. So we can safely kick it. 1602 */ 1603 if (scb) { 1604 atomic_or(PROG_BLOCK_SIE, &scb->prog20); 1605 if (scb->prog0c & PROG_IN_SIE) 1606 atomic_or(CPUSTAT_STOP_INT, &scb->cpuflags); 1607 } 1608 } 1609