1 // SPDX-License-Identifier: GPL-2.0 2 /* 3 * kvm nested virtualization support for s390x 4 * 5 * Copyright IBM Corp. 2016 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 /* the pinned originial scb */ 32 struct kvm_s390_sie_block *scb_o; /* 0x0218 */ 33 /* the shadow gmap in use by the vsie_page */ 34 struct gmap *gmap; /* 0x0220 */ 35 /* address of the last reported fault to guest2 */ 36 unsigned long fault_addr; /* 0x0228 */ 37 __u8 reserved[0x0700 - 0x0230]; /* 0x0230 */ 38 struct kvm_s390_crypto_cb crycb; /* 0x0700 */ 39 __u8 fac[S390_ARCH_FAC_LIST_SIZE_BYTE]; /* 0x0800 */ 40 }; 41 42 /* trigger a validity icpt for the given scb */ 43 static int set_validity_icpt(struct kvm_s390_sie_block *scb, 44 __u16 reason_code) 45 { 46 scb->ipa = 0x1000; 47 scb->ipb = ((__u32) reason_code) << 16; 48 scb->icptcode = ICPT_VALIDITY; 49 return 1; 50 } 51 52 /* mark the prefix as unmapped, this will block the VSIE */ 53 static void prefix_unmapped(struct vsie_page *vsie_page) 54 { 55 atomic_or(PROG_REQUEST, &vsie_page->scb_s.prog20); 56 } 57 58 /* mark the prefix as unmapped and wait until the VSIE has been left */ 59 static void prefix_unmapped_sync(struct vsie_page *vsie_page) 60 { 61 prefix_unmapped(vsie_page); 62 if (vsie_page->scb_s.prog0c & PROG_IN_SIE) 63 atomic_or(CPUSTAT_STOP_INT, &vsie_page->scb_s.cpuflags); 64 while (vsie_page->scb_s.prog0c & PROG_IN_SIE) 65 cpu_relax(); 66 } 67 68 /* mark the prefix as mapped, this will allow the VSIE to run */ 69 static void prefix_mapped(struct vsie_page *vsie_page) 70 { 71 atomic_andnot(PROG_REQUEST, &vsie_page->scb_s.prog20); 72 } 73 74 /* test if the prefix is mapped into the gmap shadow */ 75 static int prefix_is_mapped(struct vsie_page *vsie_page) 76 { 77 return !(atomic_read(&vsie_page->scb_s.prog20) & PROG_REQUEST); 78 } 79 80 /* copy the updated intervention request bits into the shadow scb */ 81 static void update_intervention_requests(struct vsie_page *vsie_page) 82 { 83 const int bits = CPUSTAT_STOP_INT | CPUSTAT_IO_INT | CPUSTAT_EXT_INT; 84 int cpuflags; 85 86 cpuflags = atomic_read(&vsie_page->scb_o->cpuflags); 87 atomic_andnot(bits, &vsie_page->scb_s.cpuflags); 88 atomic_or(cpuflags & bits, &vsie_page->scb_s.cpuflags); 89 } 90 91 /* shadow (filter and validate) the cpuflags */ 92 static int prepare_cpuflags(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page) 93 { 94 struct kvm_s390_sie_block *scb_s = &vsie_page->scb_s; 95 struct kvm_s390_sie_block *scb_o = vsie_page->scb_o; 96 int newflags, cpuflags = atomic_read(&scb_o->cpuflags); 97 98 /* we don't allow ESA/390 guests */ 99 if (!(cpuflags & CPUSTAT_ZARCH)) 100 return set_validity_icpt(scb_s, 0x0001U); 101 102 if (cpuflags & (CPUSTAT_RRF | CPUSTAT_MCDS)) 103 return set_validity_icpt(scb_s, 0x0001U); 104 else if (cpuflags & (CPUSTAT_SLSV | CPUSTAT_SLSR)) 105 return set_validity_icpt(scb_s, 0x0007U); 106 107 /* intervention requests will be set later */ 108 newflags = CPUSTAT_ZARCH; 109 if (cpuflags & CPUSTAT_GED && test_kvm_facility(vcpu->kvm, 8)) 110 newflags |= CPUSTAT_GED; 111 if (cpuflags & CPUSTAT_GED2 && test_kvm_facility(vcpu->kvm, 78)) { 112 if (cpuflags & CPUSTAT_GED) 113 return set_validity_icpt(scb_s, 0x0001U); 114 newflags |= CPUSTAT_GED2; 115 } 116 if (test_kvm_cpu_feat(vcpu->kvm, KVM_S390_VM_CPU_FEAT_GPERE)) 117 newflags |= cpuflags & CPUSTAT_P; 118 if (test_kvm_cpu_feat(vcpu->kvm, KVM_S390_VM_CPU_FEAT_GSLS)) 119 newflags |= cpuflags & CPUSTAT_SM; 120 if (test_kvm_cpu_feat(vcpu->kvm, KVM_S390_VM_CPU_FEAT_IBS)) 121 newflags |= cpuflags & CPUSTAT_IBS; 122 if (test_kvm_cpu_feat(vcpu->kvm, KVM_S390_VM_CPU_FEAT_KSS)) 123 newflags |= cpuflags & CPUSTAT_KSS; 124 125 atomic_set(&scb_s->cpuflags, newflags); 126 return 0; 127 } 128 129 /* 130 * Create a shadow copy of the crycb block and setup key wrapping, if 131 * requested for guest 3 and enabled for guest 2. 132 * 133 * We only accept format-1 (no AP in g2), but convert it into format-2 134 * There is nothing to do for format-0. 135 * 136 * Returns: - 0 if shadowed or nothing to do 137 * - > 0 if control has to be given to guest 2 138 */ 139 static int shadow_crycb(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page) 140 { 141 struct kvm_s390_sie_block *scb_s = &vsie_page->scb_s; 142 struct kvm_s390_sie_block *scb_o = vsie_page->scb_o; 143 u32 crycb_addr = scb_o->crycbd & 0x7ffffff8U; 144 unsigned long *b1, *b2; 145 u8 ecb3_flags; 146 147 scb_s->crycbd = 0; 148 if (!(scb_o->crycbd & vcpu->arch.sie_block->crycbd & CRYCB_FORMAT1)) 149 return 0; 150 /* format-1 is supported with message-security-assist extension 3 */ 151 if (!test_kvm_facility(vcpu->kvm, 76)) 152 return 0; 153 /* we may only allow it if enabled for guest 2 */ 154 ecb3_flags = scb_o->ecb3 & vcpu->arch.sie_block->ecb3 & 155 (ECB3_AES | ECB3_DEA); 156 if (!ecb3_flags) 157 return 0; 158 159 if ((crycb_addr & PAGE_MASK) != ((crycb_addr + 128) & PAGE_MASK)) 160 return set_validity_icpt(scb_s, 0x003CU); 161 else if (!crycb_addr) 162 return set_validity_icpt(scb_s, 0x0039U); 163 164 /* copy only the wrapping keys */ 165 if (read_guest_real(vcpu, crycb_addr + 72, &vsie_page->crycb, 56)) 166 return set_validity_icpt(scb_s, 0x0035U); 167 168 scb_s->ecb3 |= ecb3_flags; 169 scb_s->crycbd = ((__u32)(__u64) &vsie_page->crycb) | CRYCB_FORMAT1 | 170 CRYCB_FORMAT2; 171 172 /* xor both blocks in one run */ 173 b1 = (unsigned long *) vsie_page->crycb.dea_wrapping_key_mask; 174 b2 = (unsigned long *) 175 vcpu->kvm->arch.crypto.crycb->dea_wrapping_key_mask; 176 /* as 56%8 == 0, bitmap_xor won't overwrite any data */ 177 bitmap_xor(b1, b1, b2, BITS_PER_BYTE * 56); 178 return 0; 179 } 180 181 /* shadow (round up/down) the ibc to avoid validity icpt */ 182 static void prepare_ibc(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page) 183 { 184 struct kvm_s390_sie_block *scb_s = &vsie_page->scb_s; 185 struct kvm_s390_sie_block *scb_o = vsie_page->scb_o; 186 __u64 min_ibc = (sclp.ibc >> 16) & 0x0fffU; 187 188 scb_s->ibc = 0; 189 /* ibc installed in g2 and requested for g3 */ 190 if (vcpu->kvm->arch.model.ibc && (scb_o->ibc & 0x0fffU)) { 191 scb_s->ibc = scb_o->ibc & 0x0fffU; 192 /* takte care of the minimum ibc level of the machine */ 193 if (scb_s->ibc < min_ibc) 194 scb_s->ibc = min_ibc; 195 /* take care of the maximum ibc level set for the guest */ 196 if (scb_s->ibc > vcpu->kvm->arch.model.ibc) 197 scb_s->ibc = vcpu->kvm->arch.model.ibc; 198 } 199 } 200 201 /* unshadow the scb, copying parameters back to the real scb */ 202 static void unshadow_scb(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page) 203 { 204 struct kvm_s390_sie_block *scb_s = &vsie_page->scb_s; 205 struct kvm_s390_sie_block *scb_o = vsie_page->scb_o; 206 207 /* interception */ 208 scb_o->icptcode = scb_s->icptcode; 209 scb_o->icptstatus = scb_s->icptstatus; 210 scb_o->ipa = scb_s->ipa; 211 scb_o->ipb = scb_s->ipb; 212 scb_o->gbea = scb_s->gbea; 213 214 /* timer */ 215 scb_o->cputm = scb_s->cputm; 216 scb_o->ckc = scb_s->ckc; 217 scb_o->todpr = scb_s->todpr; 218 219 /* guest state */ 220 scb_o->gpsw = scb_s->gpsw; 221 scb_o->gg14 = scb_s->gg14; 222 scb_o->gg15 = scb_s->gg15; 223 memcpy(scb_o->gcr, scb_s->gcr, 128); 224 scb_o->pp = scb_s->pp; 225 226 /* branch prediction */ 227 if (test_kvm_facility(vcpu->kvm, 82)) { 228 scb_o->fpf &= ~FPF_BPBC; 229 scb_o->fpf |= scb_s->fpf & FPF_BPBC; 230 } 231 232 /* interrupt intercept */ 233 switch (scb_s->icptcode) { 234 case ICPT_PROGI: 235 case ICPT_INSTPROGI: 236 case ICPT_EXTINT: 237 memcpy((void *)((u64)scb_o + 0xc0), 238 (void *)((u64)scb_s + 0xc0), 0xf0 - 0xc0); 239 break; 240 case ICPT_PARTEXEC: 241 /* MVPG only */ 242 memcpy((void *)((u64)scb_o + 0xc0), 243 (void *)((u64)scb_s + 0xc0), 0xd0 - 0xc0); 244 break; 245 } 246 247 if (scb_s->ihcpu != 0xffffU) 248 scb_o->ihcpu = scb_s->ihcpu; 249 } 250 251 /* 252 * Setup the shadow scb by copying and checking the relevant parts of the g2 253 * provided scb. 254 * 255 * Returns: - 0 if the scb has been shadowed 256 * - > 0 if control has to be given to guest 2 257 */ 258 static int shadow_scb(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page) 259 { 260 struct kvm_s390_sie_block *scb_o = vsie_page->scb_o; 261 struct kvm_s390_sie_block *scb_s = &vsie_page->scb_s; 262 bool had_tx = scb_s->ecb & ECB_TE; 263 unsigned long new_mso = 0; 264 int rc; 265 266 /* make sure we don't have any leftovers when reusing the scb */ 267 scb_s->icptcode = 0; 268 scb_s->eca = 0; 269 scb_s->ecb = 0; 270 scb_s->ecb2 = 0; 271 scb_s->ecb3 = 0; 272 scb_s->ecd = 0; 273 scb_s->fac = 0; 274 scb_s->fpf = 0; 275 276 rc = prepare_cpuflags(vcpu, vsie_page); 277 if (rc) 278 goto out; 279 280 /* timer */ 281 scb_s->cputm = scb_o->cputm; 282 scb_s->ckc = scb_o->ckc; 283 scb_s->todpr = scb_o->todpr; 284 scb_s->epoch = scb_o->epoch; 285 286 /* guest state */ 287 scb_s->gpsw = scb_o->gpsw; 288 scb_s->gg14 = scb_o->gg14; 289 scb_s->gg15 = scb_o->gg15; 290 memcpy(scb_s->gcr, scb_o->gcr, 128); 291 scb_s->pp = scb_o->pp; 292 293 /* interception / execution handling */ 294 scb_s->gbea = scb_o->gbea; 295 scb_s->lctl = scb_o->lctl; 296 scb_s->svcc = scb_o->svcc; 297 scb_s->ictl = scb_o->ictl; 298 /* 299 * SKEY handling functions can't deal with false setting of PTE invalid 300 * bits. Therefore we cannot provide interpretation and would later 301 * have to provide own emulation handlers. 302 */ 303 if (!(atomic_read(&scb_s->cpuflags) & CPUSTAT_KSS)) 304 scb_s->ictl |= ICTL_ISKE | ICTL_SSKE | ICTL_RRBE; 305 306 scb_s->icpua = scb_o->icpua; 307 308 if (!(atomic_read(&scb_s->cpuflags) & CPUSTAT_SM)) 309 new_mso = scb_o->mso & 0xfffffffffff00000UL; 310 /* if the hva of the prefix changes, we have to remap the prefix */ 311 if (scb_s->mso != new_mso || scb_s->prefix != scb_o->prefix) 312 prefix_unmapped(vsie_page); 313 /* SIE will do mso/msl validity and exception checks for us */ 314 scb_s->msl = scb_o->msl & 0xfffffffffff00000UL; 315 scb_s->mso = new_mso; 316 scb_s->prefix = scb_o->prefix; 317 318 /* We have to definetly flush the tlb if this scb never ran */ 319 if (scb_s->ihcpu != 0xffffU) 320 scb_s->ihcpu = scb_o->ihcpu; 321 322 /* MVPG and Protection Exception Interpretation are always available */ 323 scb_s->eca |= scb_o->eca & (ECA_MVPGI | ECA_PROTEXCI); 324 /* Host-protection-interruption introduced with ESOP */ 325 if (test_kvm_cpu_feat(vcpu->kvm, KVM_S390_VM_CPU_FEAT_ESOP)) 326 scb_s->ecb |= scb_o->ecb & ECB_HOSTPROTINT; 327 /* transactional execution */ 328 if (test_kvm_facility(vcpu->kvm, 73)) { 329 /* remap the prefix is tx is toggled on */ 330 if ((scb_o->ecb & ECB_TE) && !had_tx) 331 prefix_unmapped(vsie_page); 332 scb_s->ecb |= scb_o->ecb & ECB_TE; 333 } 334 /* branch prediction */ 335 if (test_kvm_facility(vcpu->kvm, 82)) 336 scb_s->fpf |= scb_o->fpf & FPF_BPBC; 337 /* SIMD */ 338 if (test_kvm_facility(vcpu->kvm, 129)) { 339 scb_s->eca |= scb_o->eca & ECA_VX; 340 scb_s->ecd |= scb_o->ecd & ECD_HOSTREGMGMT; 341 } 342 /* Run-time-Instrumentation */ 343 if (test_kvm_facility(vcpu->kvm, 64)) 344 scb_s->ecb3 |= scb_o->ecb3 & ECB3_RI; 345 /* Instruction Execution Prevention */ 346 if (test_kvm_facility(vcpu->kvm, 130)) 347 scb_s->ecb2 |= scb_o->ecb2 & ECB2_IEP; 348 /* Guarded Storage */ 349 if (test_kvm_facility(vcpu->kvm, 133)) { 350 scb_s->ecb |= scb_o->ecb & ECB_GS; 351 scb_s->ecd |= scb_o->ecd & ECD_HOSTREGMGMT; 352 } 353 if (test_kvm_cpu_feat(vcpu->kvm, KVM_S390_VM_CPU_FEAT_SIIF)) 354 scb_s->eca |= scb_o->eca & ECA_SII; 355 if (test_kvm_cpu_feat(vcpu->kvm, KVM_S390_VM_CPU_FEAT_IB)) 356 scb_s->eca |= scb_o->eca & ECA_IB; 357 if (test_kvm_cpu_feat(vcpu->kvm, KVM_S390_VM_CPU_FEAT_CEI)) 358 scb_s->eca |= scb_o->eca & ECA_CEI; 359 /* Epoch Extension */ 360 if (test_kvm_facility(vcpu->kvm, 139)) 361 scb_s->ecd |= scb_o->ecd & ECD_MEF; 362 363 prepare_ibc(vcpu, vsie_page); 364 rc = shadow_crycb(vcpu, vsie_page); 365 out: 366 if (rc) 367 unshadow_scb(vcpu, vsie_page); 368 return rc; 369 } 370 371 void kvm_s390_vsie_gmap_notifier(struct gmap *gmap, unsigned long start, 372 unsigned long end) 373 { 374 struct kvm *kvm = gmap->private; 375 struct vsie_page *cur; 376 unsigned long prefix; 377 struct page *page; 378 int i; 379 380 if (!gmap_is_shadow(gmap)) 381 return; 382 if (start >= 1UL << 31) 383 /* We are only interested in prefix pages */ 384 return; 385 386 /* 387 * Only new shadow blocks are added to the list during runtime, 388 * therefore we can safely reference them all the time. 389 */ 390 for (i = 0; i < kvm->arch.vsie.page_count; i++) { 391 page = READ_ONCE(kvm->arch.vsie.pages[i]); 392 if (!page) 393 continue; 394 cur = page_to_virt(page); 395 if (READ_ONCE(cur->gmap) != gmap) 396 continue; 397 prefix = cur->scb_s.prefix << GUEST_PREFIX_SHIFT; 398 /* with mso/msl, the prefix lies at an offset */ 399 prefix += cur->scb_s.mso; 400 if (prefix <= end && start <= prefix + 2 * PAGE_SIZE - 1) 401 prefix_unmapped_sync(cur); 402 } 403 } 404 405 /* 406 * Map the first prefix page and if tx is enabled also the second prefix page. 407 * 408 * The prefix will be protected, a gmap notifier will inform about unmaps. 409 * The shadow scb must not be executed until the prefix is remapped, this is 410 * guaranteed by properly handling PROG_REQUEST. 411 * 412 * Returns: - 0 on if successfully mapped or already mapped 413 * - > 0 if control has to be given to guest 2 414 * - -EAGAIN if the caller can retry immediately 415 * - -ENOMEM if out of memory 416 */ 417 static int map_prefix(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page) 418 { 419 struct kvm_s390_sie_block *scb_s = &vsie_page->scb_s; 420 u64 prefix = scb_s->prefix << GUEST_PREFIX_SHIFT; 421 int rc; 422 423 if (prefix_is_mapped(vsie_page)) 424 return 0; 425 426 /* mark it as mapped so we can catch any concurrent unmappers */ 427 prefix_mapped(vsie_page); 428 429 /* with mso/msl, the prefix lies at offset *mso* */ 430 prefix += scb_s->mso; 431 432 rc = kvm_s390_shadow_fault(vcpu, vsie_page->gmap, prefix); 433 if (!rc && (scb_s->ecb & ECB_TE)) 434 rc = kvm_s390_shadow_fault(vcpu, vsie_page->gmap, 435 prefix + PAGE_SIZE); 436 /* 437 * We don't have to mprotect, we will be called for all unshadows. 438 * SIE will detect if protection applies and trigger a validity. 439 */ 440 if (rc) 441 prefix_unmapped(vsie_page); 442 if (rc > 0 || rc == -EFAULT) 443 rc = set_validity_icpt(scb_s, 0x0037U); 444 return rc; 445 } 446 447 /* 448 * Pin the guest page given by gpa and set hpa to the pinned host address. 449 * Will always be pinned writable. 450 * 451 * Returns: - 0 on success 452 * - -EINVAL if the gpa is not valid guest storage 453 */ 454 static int pin_guest_page(struct kvm *kvm, gpa_t gpa, hpa_t *hpa) 455 { 456 struct page *page; 457 458 page = gfn_to_page(kvm, gpa_to_gfn(gpa)); 459 if (is_error_page(page)) 460 return -EINVAL; 461 *hpa = (hpa_t) page_to_virt(page) + (gpa & ~PAGE_MASK); 462 return 0; 463 } 464 465 /* Unpins a page previously pinned via pin_guest_page, marking it as dirty. */ 466 static void unpin_guest_page(struct kvm *kvm, gpa_t gpa, hpa_t hpa) 467 { 468 kvm_release_pfn_dirty(hpa >> PAGE_SHIFT); 469 /* mark the page always as dirty for migration */ 470 mark_page_dirty(kvm, gpa_to_gfn(gpa)); 471 } 472 473 /* unpin all blocks previously pinned by pin_blocks(), marking them dirty */ 474 static void unpin_blocks(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page) 475 { 476 struct kvm_s390_sie_block *scb_o = vsie_page->scb_o; 477 struct kvm_s390_sie_block *scb_s = &vsie_page->scb_s; 478 hpa_t hpa; 479 gpa_t gpa; 480 481 hpa = (u64) scb_s->scaoh << 32 | scb_s->scaol; 482 if (hpa) { 483 gpa = scb_o->scaol & ~0xfUL; 484 if (test_kvm_cpu_feat(vcpu->kvm, KVM_S390_VM_CPU_FEAT_64BSCAO)) 485 gpa |= (u64) scb_o->scaoh << 32; 486 unpin_guest_page(vcpu->kvm, gpa, hpa); 487 scb_s->scaol = 0; 488 scb_s->scaoh = 0; 489 } 490 491 hpa = scb_s->itdba; 492 if (hpa) { 493 gpa = scb_o->itdba & ~0xffUL; 494 unpin_guest_page(vcpu->kvm, gpa, hpa); 495 scb_s->itdba = 0; 496 } 497 498 hpa = scb_s->gvrd; 499 if (hpa) { 500 gpa = scb_o->gvrd & ~0x1ffUL; 501 unpin_guest_page(vcpu->kvm, gpa, hpa); 502 scb_s->gvrd = 0; 503 } 504 505 hpa = scb_s->riccbd; 506 if (hpa) { 507 gpa = scb_o->riccbd & ~0x3fUL; 508 unpin_guest_page(vcpu->kvm, gpa, hpa); 509 scb_s->riccbd = 0; 510 } 511 512 hpa = scb_s->sdnxo; 513 if (hpa) { 514 gpa = scb_o->sdnxo; 515 unpin_guest_page(vcpu->kvm, gpa, hpa); 516 scb_s->sdnxo = 0; 517 } 518 } 519 520 /* 521 * Instead of shadowing some blocks, we can simply forward them because the 522 * addresses in the scb are 64 bit long. 523 * 524 * This works as long as the data lies in one page. If blocks ever exceed one 525 * page, we have to fall back to shadowing. 526 * 527 * As we reuse the sca, the vcpu pointers contained in it are invalid. We must 528 * therefore not enable any facilities that access these pointers (e.g. SIGPIF). 529 * 530 * Returns: - 0 if all blocks were pinned. 531 * - > 0 if control has to be given to guest 2 532 * - -ENOMEM if out of memory 533 */ 534 static int pin_blocks(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page) 535 { 536 struct kvm_s390_sie_block *scb_o = vsie_page->scb_o; 537 struct kvm_s390_sie_block *scb_s = &vsie_page->scb_s; 538 hpa_t hpa; 539 gpa_t gpa; 540 int rc = 0; 541 542 gpa = scb_o->scaol & ~0xfUL; 543 if (test_kvm_cpu_feat(vcpu->kvm, KVM_S390_VM_CPU_FEAT_64BSCAO)) 544 gpa |= (u64) scb_o->scaoh << 32; 545 if (gpa) { 546 if (!(gpa & ~0x1fffUL)) 547 rc = set_validity_icpt(scb_s, 0x0038U); 548 else if ((gpa & ~0x1fffUL) == kvm_s390_get_prefix(vcpu)) 549 rc = set_validity_icpt(scb_s, 0x0011U); 550 else if ((gpa & PAGE_MASK) != 551 ((gpa + sizeof(struct bsca_block) - 1) & PAGE_MASK)) 552 rc = set_validity_icpt(scb_s, 0x003bU); 553 if (!rc) { 554 rc = pin_guest_page(vcpu->kvm, gpa, &hpa); 555 if (rc) 556 rc = set_validity_icpt(scb_s, 0x0034U); 557 } 558 if (rc) 559 goto unpin; 560 scb_s->scaoh = (u32)((u64)hpa >> 32); 561 scb_s->scaol = (u32)(u64)hpa; 562 } 563 564 gpa = scb_o->itdba & ~0xffUL; 565 if (gpa && (scb_s->ecb & ECB_TE)) { 566 if (!(gpa & ~0x1fffU)) { 567 rc = set_validity_icpt(scb_s, 0x0080U); 568 goto unpin; 569 } 570 /* 256 bytes cannot cross page boundaries */ 571 rc = pin_guest_page(vcpu->kvm, gpa, &hpa); 572 if (rc) { 573 rc = set_validity_icpt(scb_s, 0x0080U); 574 goto unpin; 575 } 576 scb_s->itdba = hpa; 577 } 578 579 gpa = scb_o->gvrd & ~0x1ffUL; 580 if (gpa && (scb_s->eca & ECA_VX) && !(scb_s->ecd & ECD_HOSTREGMGMT)) { 581 if (!(gpa & ~0x1fffUL)) { 582 rc = set_validity_icpt(scb_s, 0x1310U); 583 goto unpin; 584 } 585 /* 586 * 512 bytes vector registers cannot cross page boundaries 587 * if this block gets bigger, we have to shadow it. 588 */ 589 rc = pin_guest_page(vcpu->kvm, gpa, &hpa); 590 if (rc) { 591 rc = set_validity_icpt(scb_s, 0x1310U); 592 goto unpin; 593 } 594 scb_s->gvrd = hpa; 595 } 596 597 gpa = scb_o->riccbd & ~0x3fUL; 598 if (gpa && (scb_s->ecb3 & ECB3_RI)) { 599 if (!(gpa & ~0x1fffUL)) { 600 rc = set_validity_icpt(scb_s, 0x0043U); 601 goto unpin; 602 } 603 /* 64 bytes cannot cross page boundaries */ 604 rc = pin_guest_page(vcpu->kvm, gpa, &hpa); 605 if (rc) { 606 rc = set_validity_icpt(scb_s, 0x0043U); 607 goto unpin; 608 } 609 /* Validity 0x0044 will be checked by SIE */ 610 scb_s->riccbd = hpa; 611 } 612 if ((scb_s->ecb & ECB_GS) && !(scb_s->ecd & ECD_HOSTREGMGMT)) { 613 unsigned long sdnxc; 614 615 gpa = scb_o->sdnxo & ~0xfUL; 616 sdnxc = scb_o->sdnxo & 0xfUL; 617 if (!gpa || !(gpa & ~0x1fffUL)) { 618 rc = set_validity_icpt(scb_s, 0x10b0U); 619 goto unpin; 620 } 621 if (sdnxc < 6 || sdnxc > 12) { 622 rc = set_validity_icpt(scb_s, 0x10b1U); 623 goto unpin; 624 } 625 if (gpa & ((1 << sdnxc) - 1)) { 626 rc = set_validity_icpt(scb_s, 0x10b2U); 627 goto unpin; 628 } 629 /* Due to alignment rules (checked above) this cannot 630 * cross page boundaries 631 */ 632 rc = pin_guest_page(vcpu->kvm, gpa, &hpa); 633 if (rc) { 634 rc = set_validity_icpt(scb_s, 0x10b0U); 635 goto unpin; 636 } 637 scb_s->sdnxo = hpa | sdnxc; 638 } 639 return 0; 640 unpin: 641 unpin_blocks(vcpu, vsie_page); 642 return rc; 643 } 644 645 /* unpin the scb provided by guest 2, marking it as dirty */ 646 static void unpin_scb(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page, 647 gpa_t gpa) 648 { 649 hpa_t hpa = (hpa_t) vsie_page->scb_o; 650 651 if (hpa) 652 unpin_guest_page(vcpu->kvm, gpa, hpa); 653 vsie_page->scb_o = NULL; 654 } 655 656 /* 657 * Pin the scb at gpa provided by guest 2 at vsie_page->scb_o. 658 * 659 * Returns: - 0 if the scb was pinned. 660 * - > 0 if control has to be given to guest 2 661 */ 662 static int pin_scb(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page, 663 gpa_t gpa) 664 { 665 hpa_t hpa; 666 int rc; 667 668 rc = pin_guest_page(vcpu->kvm, gpa, &hpa); 669 if (rc) { 670 rc = kvm_s390_inject_program_int(vcpu, PGM_ADDRESSING); 671 WARN_ON_ONCE(rc); 672 return 1; 673 } 674 vsie_page->scb_o = (struct kvm_s390_sie_block *) hpa; 675 return 0; 676 } 677 678 /* 679 * Inject a fault into guest 2. 680 * 681 * Returns: - > 0 if control has to be given to guest 2 682 * < 0 if an error occurred during injection. 683 */ 684 static int inject_fault(struct kvm_vcpu *vcpu, __u16 code, __u64 vaddr, 685 bool write_flag) 686 { 687 struct kvm_s390_pgm_info pgm = { 688 .code = code, 689 .trans_exc_code = 690 /* 0-51: virtual address */ 691 (vaddr & 0xfffffffffffff000UL) | 692 /* 52-53: store / fetch */ 693 (((unsigned int) !write_flag) + 1) << 10, 694 /* 62-63: asce id (alway primary == 0) */ 695 .exc_access_id = 0, /* always primary */ 696 .op_access_id = 0, /* not MVPG */ 697 }; 698 int rc; 699 700 if (code == PGM_PROTECTION) 701 pgm.trans_exc_code |= 0x4UL; 702 703 rc = kvm_s390_inject_prog_irq(vcpu, &pgm); 704 return rc ? rc : 1; 705 } 706 707 /* 708 * Handle a fault during vsie execution on a gmap shadow. 709 * 710 * Returns: - 0 if the fault was resolved 711 * - > 0 if control has to be given to guest 2 712 * - < 0 if an error occurred 713 */ 714 static int handle_fault(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page) 715 { 716 int rc; 717 718 if (current->thread.gmap_int_code == PGM_PROTECTION) 719 /* we can directly forward all protection exceptions */ 720 return inject_fault(vcpu, PGM_PROTECTION, 721 current->thread.gmap_addr, 1); 722 723 rc = kvm_s390_shadow_fault(vcpu, vsie_page->gmap, 724 current->thread.gmap_addr); 725 if (rc > 0) { 726 rc = inject_fault(vcpu, rc, 727 current->thread.gmap_addr, 728 current->thread.gmap_write_flag); 729 if (rc >= 0) 730 vsie_page->fault_addr = current->thread.gmap_addr; 731 } 732 return rc; 733 } 734 735 /* 736 * Retry the previous fault that required guest 2 intervention. This avoids 737 * one superfluous SIE re-entry and direct exit. 738 * 739 * Will ignore any errors. The next SIE fault will do proper fault handling. 740 */ 741 static void handle_last_fault(struct kvm_vcpu *vcpu, 742 struct vsie_page *vsie_page) 743 { 744 if (vsie_page->fault_addr) 745 kvm_s390_shadow_fault(vcpu, vsie_page->gmap, 746 vsie_page->fault_addr); 747 vsie_page->fault_addr = 0; 748 } 749 750 static inline void clear_vsie_icpt(struct vsie_page *vsie_page) 751 { 752 vsie_page->scb_s.icptcode = 0; 753 } 754 755 /* rewind the psw and clear the vsie icpt, so we can retry execution */ 756 static void retry_vsie_icpt(struct vsie_page *vsie_page) 757 { 758 struct kvm_s390_sie_block *scb_s = &vsie_page->scb_s; 759 int ilen = insn_length(scb_s->ipa >> 8); 760 761 /* take care of EXECUTE instructions */ 762 if (scb_s->icptstatus & 1) { 763 ilen = (scb_s->icptstatus >> 4) & 0x6; 764 if (!ilen) 765 ilen = 4; 766 } 767 scb_s->gpsw.addr = __rewind_psw(scb_s->gpsw, ilen); 768 clear_vsie_icpt(vsie_page); 769 } 770 771 /* 772 * Try to shadow + enable the guest 2 provided facility list. 773 * Retry instruction execution if enabled for and provided by guest 2. 774 * 775 * Returns: - 0 if handled (retry or guest 2 icpt) 776 * - > 0 if control has to be given to guest 2 777 */ 778 static int handle_stfle(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page) 779 { 780 struct kvm_s390_sie_block *scb_s = &vsie_page->scb_s; 781 __u32 fac = vsie_page->scb_o->fac & 0x7ffffff8U; 782 783 if (fac && test_kvm_facility(vcpu->kvm, 7)) { 784 retry_vsie_icpt(vsie_page); 785 if (read_guest_real(vcpu, fac, &vsie_page->fac, 786 sizeof(vsie_page->fac))) 787 return set_validity_icpt(scb_s, 0x1090U); 788 scb_s->fac = (__u32)(__u64) &vsie_page->fac; 789 } 790 return 0; 791 } 792 793 /* 794 * Run the vsie on a shadow scb and a shadow gmap, without any further 795 * sanity checks, handling SIE faults. 796 * 797 * Returns: - 0 everything went fine 798 * - > 0 if control has to be given to guest 2 799 * - < 0 if an error occurred 800 */ 801 static int do_vsie_run(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page) 802 { 803 struct kvm_s390_sie_block *scb_s = &vsie_page->scb_s; 804 struct kvm_s390_sie_block *scb_o = vsie_page->scb_o; 805 int rc; 806 807 handle_last_fault(vcpu, vsie_page); 808 809 if (need_resched()) 810 schedule(); 811 if (test_cpu_flag(CIF_MCCK_PENDING)) 812 s390_handle_mcck(); 813 814 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx); 815 local_irq_disable(); 816 guest_enter_irqoff(); 817 local_irq_enable(); 818 819 rc = sie64a(scb_s, vcpu->run->s.regs.gprs); 820 821 local_irq_disable(); 822 guest_exit_irqoff(); 823 local_irq_enable(); 824 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu); 825 826 if (rc == -EINTR) { 827 VCPU_EVENT(vcpu, 3, "%s", "machine check"); 828 kvm_s390_reinject_machine_check(vcpu, &vsie_page->mcck_info); 829 return 0; 830 } 831 832 if (rc > 0) 833 rc = 0; /* we could still have an icpt */ 834 else if (rc == -EFAULT) 835 return handle_fault(vcpu, vsie_page); 836 837 switch (scb_s->icptcode) { 838 case ICPT_INST: 839 if (scb_s->ipa == 0xb2b0) 840 rc = handle_stfle(vcpu, vsie_page); 841 break; 842 case ICPT_STOP: 843 /* stop not requested by g2 - must have been a kick */ 844 if (!(atomic_read(&scb_o->cpuflags) & CPUSTAT_STOP_INT)) 845 clear_vsie_icpt(vsie_page); 846 break; 847 case ICPT_VALIDITY: 848 if ((scb_s->ipa & 0xf000) != 0xf000) 849 scb_s->ipa += 0x1000; 850 break; 851 } 852 return rc; 853 } 854 855 static void release_gmap_shadow(struct vsie_page *vsie_page) 856 { 857 if (vsie_page->gmap) 858 gmap_put(vsie_page->gmap); 859 WRITE_ONCE(vsie_page->gmap, NULL); 860 prefix_unmapped(vsie_page); 861 } 862 863 static int acquire_gmap_shadow(struct kvm_vcpu *vcpu, 864 struct vsie_page *vsie_page) 865 { 866 unsigned long asce; 867 union ctlreg0 cr0; 868 struct gmap *gmap; 869 int edat; 870 871 asce = vcpu->arch.sie_block->gcr[1]; 872 cr0.val = vcpu->arch.sie_block->gcr[0]; 873 edat = cr0.edat && test_kvm_facility(vcpu->kvm, 8); 874 edat += edat && test_kvm_facility(vcpu->kvm, 78); 875 876 /* 877 * ASCE or EDAT could have changed since last icpt, or the gmap 878 * we're holding has been unshadowed. If the gmap is still valid, 879 * we can safely reuse it. 880 */ 881 if (vsie_page->gmap && gmap_shadow_valid(vsie_page->gmap, asce, edat)) 882 return 0; 883 884 /* release the old shadow - if any, and mark the prefix as unmapped */ 885 release_gmap_shadow(vsie_page); 886 gmap = gmap_shadow(vcpu->arch.gmap, asce, edat); 887 if (IS_ERR(gmap)) 888 return PTR_ERR(gmap); 889 gmap->private = vcpu->kvm; 890 WRITE_ONCE(vsie_page->gmap, gmap); 891 return 0; 892 } 893 894 /* 895 * Register the shadow scb at the VCPU, e.g. for kicking out of vsie. 896 */ 897 static void register_shadow_scb(struct kvm_vcpu *vcpu, 898 struct vsie_page *vsie_page) 899 { 900 struct kvm_s390_sie_block *scb_s = &vsie_page->scb_s; 901 902 WRITE_ONCE(vcpu->arch.vsie_block, &vsie_page->scb_s); 903 /* 904 * External calls have to lead to a kick of the vcpu and 905 * therefore the vsie -> Simulate Wait state. 906 */ 907 atomic_or(CPUSTAT_WAIT, &vcpu->arch.sie_block->cpuflags); 908 /* 909 * We have to adjust the g3 epoch by the g2 epoch. The epoch will 910 * automatically be adjusted on tod clock changes via kvm_sync_clock. 911 */ 912 preempt_disable(); 913 scb_s->epoch += vcpu->kvm->arch.epoch; 914 915 if (scb_s->ecd & ECD_MEF) { 916 scb_s->epdx += vcpu->kvm->arch.epdx; 917 if (scb_s->epoch < vcpu->kvm->arch.epoch) 918 scb_s->epdx += 1; 919 } 920 921 preempt_enable(); 922 } 923 924 /* 925 * Unregister a shadow scb from a VCPU. 926 */ 927 static void unregister_shadow_scb(struct kvm_vcpu *vcpu) 928 { 929 atomic_andnot(CPUSTAT_WAIT, &vcpu->arch.sie_block->cpuflags); 930 WRITE_ONCE(vcpu->arch.vsie_block, NULL); 931 } 932 933 /* 934 * Run the vsie on a shadowed scb, managing the gmap shadow, handling 935 * prefix pages and faults. 936 * 937 * Returns: - 0 if no errors occurred 938 * - > 0 if control has to be given to guest 2 939 * - -ENOMEM if out of memory 940 */ 941 static int vsie_run(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page) 942 { 943 struct kvm_s390_sie_block *scb_s = &vsie_page->scb_s; 944 int rc = 0; 945 946 while (1) { 947 rc = acquire_gmap_shadow(vcpu, vsie_page); 948 if (!rc) 949 rc = map_prefix(vcpu, vsie_page); 950 if (!rc) { 951 gmap_enable(vsie_page->gmap); 952 update_intervention_requests(vsie_page); 953 rc = do_vsie_run(vcpu, vsie_page); 954 gmap_enable(vcpu->arch.gmap); 955 } 956 atomic_andnot(PROG_BLOCK_SIE, &scb_s->prog20); 957 958 if (rc == -EAGAIN) 959 rc = 0; 960 if (rc || scb_s->icptcode || signal_pending(current) || 961 kvm_s390_vcpu_has_irq(vcpu, 0)) 962 break; 963 } 964 965 if (rc == -EFAULT) { 966 /* 967 * Addressing exceptions are always presentes as intercepts. 968 * As addressing exceptions are suppressing and our guest 3 PSW 969 * points at the responsible instruction, we have to 970 * forward the PSW and set the ilc. If we can't read guest 3 971 * instruction, we can use an arbitrary ilc. Let's always use 972 * ilen = 4 for now, so we can avoid reading in guest 3 virtual 973 * memory. (we could also fake the shadow so the hardware 974 * handles it). 975 */ 976 scb_s->icptcode = ICPT_PROGI; 977 scb_s->iprcc = PGM_ADDRESSING; 978 scb_s->pgmilc = 4; 979 scb_s->gpsw.addr = __rewind_psw(scb_s->gpsw, 4); 980 } 981 return rc; 982 } 983 984 /* 985 * Get or create a vsie page for a scb address. 986 * 987 * Returns: - address of a vsie page (cached or new one) 988 * - NULL if the same scb address is already used by another VCPU 989 * - ERR_PTR(-ENOMEM) if out of memory 990 */ 991 static struct vsie_page *get_vsie_page(struct kvm *kvm, unsigned long addr) 992 { 993 struct vsie_page *vsie_page; 994 struct page *page; 995 int nr_vcpus; 996 997 rcu_read_lock(); 998 page = radix_tree_lookup(&kvm->arch.vsie.addr_to_page, addr >> 9); 999 rcu_read_unlock(); 1000 if (page) { 1001 if (page_ref_inc_return(page) == 2) 1002 return page_to_virt(page); 1003 page_ref_dec(page); 1004 } 1005 1006 /* 1007 * We want at least #online_vcpus shadows, so every VCPU can execute 1008 * the VSIE in parallel. 1009 */ 1010 nr_vcpus = atomic_read(&kvm->online_vcpus); 1011 1012 mutex_lock(&kvm->arch.vsie.mutex); 1013 if (kvm->arch.vsie.page_count < nr_vcpus) { 1014 page = alloc_page(GFP_KERNEL | __GFP_ZERO | GFP_DMA); 1015 if (!page) { 1016 mutex_unlock(&kvm->arch.vsie.mutex); 1017 return ERR_PTR(-ENOMEM); 1018 } 1019 page_ref_inc(page); 1020 kvm->arch.vsie.pages[kvm->arch.vsie.page_count] = page; 1021 kvm->arch.vsie.page_count++; 1022 } else { 1023 /* reuse an existing entry that belongs to nobody */ 1024 while (true) { 1025 page = kvm->arch.vsie.pages[kvm->arch.vsie.next]; 1026 if (page_ref_inc_return(page) == 2) 1027 break; 1028 page_ref_dec(page); 1029 kvm->arch.vsie.next++; 1030 kvm->arch.vsie.next %= nr_vcpus; 1031 } 1032 radix_tree_delete(&kvm->arch.vsie.addr_to_page, page->index >> 9); 1033 } 1034 page->index = addr; 1035 /* double use of the same address */ 1036 if (radix_tree_insert(&kvm->arch.vsie.addr_to_page, addr >> 9, page)) { 1037 page_ref_dec(page); 1038 mutex_unlock(&kvm->arch.vsie.mutex); 1039 return NULL; 1040 } 1041 mutex_unlock(&kvm->arch.vsie.mutex); 1042 1043 vsie_page = page_to_virt(page); 1044 memset(&vsie_page->scb_s, 0, sizeof(struct kvm_s390_sie_block)); 1045 release_gmap_shadow(vsie_page); 1046 vsie_page->fault_addr = 0; 1047 vsie_page->scb_s.ihcpu = 0xffffU; 1048 return vsie_page; 1049 } 1050 1051 /* put a vsie page acquired via get_vsie_page */ 1052 static void put_vsie_page(struct kvm *kvm, struct vsie_page *vsie_page) 1053 { 1054 struct page *page = pfn_to_page(__pa(vsie_page) >> PAGE_SHIFT); 1055 1056 page_ref_dec(page); 1057 } 1058 1059 int kvm_s390_handle_vsie(struct kvm_vcpu *vcpu) 1060 { 1061 struct vsie_page *vsie_page; 1062 unsigned long scb_addr; 1063 int rc; 1064 1065 vcpu->stat.instruction_sie++; 1066 if (!test_kvm_cpu_feat(vcpu->kvm, KVM_S390_VM_CPU_FEAT_SIEF2)) 1067 return -EOPNOTSUPP; 1068 if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE) 1069 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP); 1070 1071 BUILD_BUG_ON(sizeof(struct vsie_page) != PAGE_SIZE); 1072 scb_addr = kvm_s390_get_base_disp_s(vcpu, NULL); 1073 1074 /* 512 byte alignment */ 1075 if (unlikely(scb_addr & 0x1ffUL)) 1076 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION); 1077 1078 if (signal_pending(current) || kvm_s390_vcpu_has_irq(vcpu, 0)) 1079 return 0; 1080 1081 vsie_page = get_vsie_page(vcpu->kvm, scb_addr); 1082 if (IS_ERR(vsie_page)) 1083 return PTR_ERR(vsie_page); 1084 else if (!vsie_page) 1085 /* double use of sie control block - simply do nothing */ 1086 return 0; 1087 1088 rc = pin_scb(vcpu, vsie_page, scb_addr); 1089 if (rc) 1090 goto out_put; 1091 rc = shadow_scb(vcpu, vsie_page); 1092 if (rc) 1093 goto out_unpin_scb; 1094 rc = pin_blocks(vcpu, vsie_page); 1095 if (rc) 1096 goto out_unshadow; 1097 register_shadow_scb(vcpu, vsie_page); 1098 rc = vsie_run(vcpu, vsie_page); 1099 unregister_shadow_scb(vcpu); 1100 unpin_blocks(vcpu, vsie_page); 1101 out_unshadow: 1102 unshadow_scb(vcpu, vsie_page); 1103 out_unpin_scb: 1104 unpin_scb(vcpu, vsie_page, scb_addr); 1105 out_put: 1106 put_vsie_page(vcpu->kvm, vsie_page); 1107 1108 return rc < 0 ? rc : 0; 1109 } 1110 1111 /* Init the vsie data structures. To be called when a vm is initialized. */ 1112 void kvm_s390_vsie_init(struct kvm *kvm) 1113 { 1114 mutex_init(&kvm->arch.vsie.mutex); 1115 INIT_RADIX_TREE(&kvm->arch.vsie.addr_to_page, GFP_KERNEL); 1116 } 1117 1118 /* Destroy the vsie data structures. To be called when a vm is destroyed. */ 1119 void kvm_s390_vsie_destroy(struct kvm *kvm) 1120 { 1121 struct vsie_page *vsie_page; 1122 struct page *page; 1123 int i; 1124 1125 mutex_lock(&kvm->arch.vsie.mutex); 1126 for (i = 0; i < kvm->arch.vsie.page_count; i++) { 1127 page = kvm->arch.vsie.pages[i]; 1128 kvm->arch.vsie.pages[i] = NULL; 1129 vsie_page = page_to_virt(page); 1130 release_gmap_shadow(vsie_page); 1131 /* free the radix tree entry */ 1132 radix_tree_delete(&kvm->arch.vsie.addr_to_page, page->index >> 9); 1133 __free_page(page); 1134 } 1135 kvm->arch.vsie.page_count = 0; 1136 mutex_unlock(&kvm->arch.vsie.mutex); 1137 } 1138 1139 void kvm_s390_vsie_kick(struct kvm_vcpu *vcpu) 1140 { 1141 struct kvm_s390_sie_block *scb = READ_ONCE(vcpu->arch.vsie_block); 1142 1143 /* 1144 * Even if the VCPU lets go of the shadow sie block reference, it is 1145 * still valid in the cache. So we can safely kick it. 1146 */ 1147 if (scb) { 1148 atomic_or(PROG_BLOCK_SIE, &scb->prog20); 1149 if (scb->prog0c & PROG_IN_SIE) 1150 atomic_or(CPUSTAT_STOP_INT, &scb->cpuflags); 1151 } 1152 } 1153