1 // SPDX-License-Identifier: GPL-2.0 2 /* 3 * kvm nested virtualization support for s390x 4 * 5 * Copyright IBM Corp. 2016, 2018 6 * 7 * Author(s): David Hildenbrand <dahi@linux.vnet.ibm.com> 8 */ 9 #include <linux/vmalloc.h> 10 #include <linux/kvm_host.h> 11 #include <linux/bug.h> 12 #include <linux/list.h> 13 #include <linux/bitmap.h> 14 #include <linux/sched/signal.h> 15 16 #include <asm/gmap.h> 17 #include <asm/mmu_context.h> 18 #include <asm/sclp.h> 19 #include <asm/nmi.h> 20 #include <asm/dis.h> 21 #include <asm/facility.h> 22 #include "kvm-s390.h" 23 #include "gaccess.h" 24 25 struct vsie_page { 26 struct kvm_s390_sie_block scb_s; /* 0x0000 */ 27 /* 28 * the backup info for machine check. ensure it's at 29 * the same offset as that in struct sie_page! 30 */ 31 struct mcck_volatile_info mcck_info; /* 0x0200 */ 32 /* 33 * The pinned original scb. Be aware that other VCPUs can modify 34 * it while we read from it. Values that are used for conditions or 35 * are reused conditionally, should be accessed via READ_ONCE. 36 */ 37 struct kvm_s390_sie_block *scb_o; /* 0x0218 */ 38 /* the shadow gmap in use by the vsie_page */ 39 struct gmap *gmap; /* 0x0220 */ 40 /* address of the last reported fault to guest2 */ 41 unsigned long fault_addr; /* 0x0228 */ 42 /* calculated guest addresses of satellite control blocks */ 43 gpa_t sca_gpa; /* 0x0230 */ 44 gpa_t itdba_gpa; /* 0x0238 */ 45 gpa_t gvrd_gpa; /* 0x0240 */ 46 gpa_t riccbd_gpa; /* 0x0248 */ 47 gpa_t sdnx_gpa; /* 0x0250 */ 48 __u8 reserved[0x0700 - 0x0258]; /* 0x0258 */ 49 struct kvm_s390_crypto_cb crycb; /* 0x0700 */ 50 __u8 fac[S390_ARCH_FAC_LIST_SIZE_BYTE]; /* 0x0800 */ 51 }; 52 53 /* trigger a validity icpt for the given scb */ 54 static int set_validity_icpt(struct kvm_s390_sie_block *scb, 55 __u16 reason_code) 56 { 57 scb->ipa = 0x1000; 58 scb->ipb = ((__u32) reason_code) << 16; 59 scb->icptcode = ICPT_VALIDITY; 60 return 1; 61 } 62 63 /* mark the prefix as unmapped, this will block the VSIE */ 64 static void prefix_unmapped(struct vsie_page *vsie_page) 65 { 66 atomic_or(PROG_REQUEST, &vsie_page->scb_s.prog20); 67 } 68 69 /* mark the prefix as unmapped and wait until the VSIE has been left */ 70 static void prefix_unmapped_sync(struct vsie_page *vsie_page) 71 { 72 prefix_unmapped(vsie_page); 73 if (vsie_page->scb_s.prog0c & PROG_IN_SIE) 74 atomic_or(CPUSTAT_STOP_INT, &vsie_page->scb_s.cpuflags); 75 while (vsie_page->scb_s.prog0c & PROG_IN_SIE) 76 cpu_relax(); 77 } 78 79 /* mark the prefix as mapped, this will allow the VSIE to run */ 80 static void prefix_mapped(struct vsie_page *vsie_page) 81 { 82 atomic_andnot(PROG_REQUEST, &vsie_page->scb_s.prog20); 83 } 84 85 /* test if the prefix is mapped into the gmap shadow */ 86 static int prefix_is_mapped(struct vsie_page *vsie_page) 87 { 88 return !(atomic_read(&vsie_page->scb_s.prog20) & PROG_REQUEST); 89 } 90 91 /* copy the updated intervention request bits into the shadow scb */ 92 static void update_intervention_requests(struct vsie_page *vsie_page) 93 { 94 const int bits = CPUSTAT_STOP_INT | CPUSTAT_IO_INT | CPUSTAT_EXT_INT; 95 int cpuflags; 96 97 cpuflags = atomic_read(&vsie_page->scb_o->cpuflags); 98 atomic_andnot(bits, &vsie_page->scb_s.cpuflags); 99 atomic_or(cpuflags & bits, &vsie_page->scb_s.cpuflags); 100 } 101 102 /* shadow (filter and validate) the cpuflags */ 103 static int prepare_cpuflags(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page) 104 { 105 struct kvm_s390_sie_block *scb_s = &vsie_page->scb_s; 106 struct kvm_s390_sie_block *scb_o = vsie_page->scb_o; 107 int newflags, cpuflags = atomic_read(&scb_o->cpuflags); 108 109 /* we don't allow ESA/390 guests */ 110 if (!(cpuflags & CPUSTAT_ZARCH)) 111 return set_validity_icpt(scb_s, 0x0001U); 112 113 if (cpuflags & (CPUSTAT_RRF | CPUSTAT_MCDS)) 114 return set_validity_icpt(scb_s, 0x0001U); 115 else if (cpuflags & (CPUSTAT_SLSV | CPUSTAT_SLSR)) 116 return set_validity_icpt(scb_s, 0x0007U); 117 118 /* intervention requests will be set later */ 119 newflags = CPUSTAT_ZARCH; 120 if (cpuflags & CPUSTAT_GED && test_kvm_facility(vcpu->kvm, 8)) 121 newflags |= CPUSTAT_GED; 122 if (cpuflags & CPUSTAT_GED2 && test_kvm_facility(vcpu->kvm, 78)) { 123 if (cpuflags & CPUSTAT_GED) 124 return set_validity_icpt(scb_s, 0x0001U); 125 newflags |= CPUSTAT_GED2; 126 } 127 if (test_kvm_cpu_feat(vcpu->kvm, KVM_S390_VM_CPU_FEAT_GPERE)) 128 newflags |= cpuflags & CPUSTAT_P; 129 if (test_kvm_cpu_feat(vcpu->kvm, KVM_S390_VM_CPU_FEAT_GSLS)) 130 newflags |= cpuflags & CPUSTAT_SM; 131 if (test_kvm_cpu_feat(vcpu->kvm, KVM_S390_VM_CPU_FEAT_IBS)) 132 newflags |= cpuflags & CPUSTAT_IBS; 133 if (test_kvm_cpu_feat(vcpu->kvm, KVM_S390_VM_CPU_FEAT_KSS)) 134 newflags |= cpuflags & CPUSTAT_KSS; 135 136 atomic_set(&scb_s->cpuflags, newflags); 137 return 0; 138 } 139 /* Copy to APCB FORMAT1 from APCB FORMAT0 */ 140 static int setup_apcb10(struct kvm_vcpu *vcpu, struct kvm_s390_apcb1 *apcb_s, 141 unsigned long crycb_gpa, struct kvm_s390_apcb1 *apcb_h) 142 { 143 struct kvm_s390_apcb0 tmp; 144 unsigned long apcb_gpa; 145 146 apcb_gpa = crycb_gpa + offsetof(struct kvm_s390_crypto_cb, apcb0); 147 148 if (read_guest_real(vcpu, apcb_gpa, &tmp, 149 sizeof(struct kvm_s390_apcb0))) 150 return -EFAULT; 151 152 apcb_s->apm[0] = apcb_h->apm[0] & tmp.apm[0]; 153 apcb_s->aqm[0] = apcb_h->aqm[0] & tmp.aqm[0] & 0xffff000000000000UL; 154 apcb_s->adm[0] = apcb_h->adm[0] & tmp.adm[0] & 0xffff000000000000UL; 155 156 return 0; 157 158 } 159 160 /** 161 * setup_apcb00 - Copy to APCB FORMAT0 from APCB FORMAT0 162 * @vcpu: pointer to the virtual CPU 163 * @apcb_s: pointer to start of apcb in the shadow crycb 164 * @crycb_gpa: guest physical address to start of original guest crycb 165 * @apcb_h: pointer to start of apcb in the guest1 166 * 167 * Returns 0 and -EFAULT on error reading guest apcb 168 */ 169 static int setup_apcb00(struct kvm_vcpu *vcpu, unsigned long *apcb_s, 170 unsigned long crycb_gpa, unsigned long *apcb_h) 171 { 172 unsigned long apcb_gpa; 173 174 apcb_gpa = crycb_gpa + offsetof(struct kvm_s390_crypto_cb, apcb0); 175 176 if (read_guest_real(vcpu, apcb_gpa, apcb_s, 177 sizeof(struct kvm_s390_apcb0))) 178 return -EFAULT; 179 180 bitmap_and(apcb_s, apcb_s, apcb_h, 181 BITS_PER_BYTE * sizeof(struct kvm_s390_apcb0)); 182 183 return 0; 184 } 185 186 /** 187 * setup_apcb11 - Copy the FORMAT1 APCB from the guest to the shadow CRYCB 188 * @vcpu: pointer to the virtual CPU 189 * @apcb_s: pointer to start of apcb in the shadow crycb 190 * @crycb_gpa: guest physical address to start of original guest crycb 191 * @apcb_h: pointer to start of apcb in the host 192 * 193 * Returns 0 and -EFAULT on error reading guest apcb 194 */ 195 static int setup_apcb11(struct kvm_vcpu *vcpu, unsigned long *apcb_s, 196 unsigned long crycb_gpa, 197 unsigned long *apcb_h) 198 { 199 unsigned long apcb_gpa; 200 201 apcb_gpa = crycb_gpa + offsetof(struct kvm_s390_crypto_cb, apcb1); 202 203 if (read_guest_real(vcpu, apcb_gpa, apcb_s, 204 sizeof(struct kvm_s390_apcb1))) 205 return -EFAULT; 206 207 bitmap_and(apcb_s, apcb_s, apcb_h, 208 BITS_PER_BYTE * sizeof(struct kvm_s390_apcb1)); 209 210 return 0; 211 } 212 213 /** 214 * setup_apcb - Create a shadow copy of the apcb. 215 * @vcpu: pointer to the virtual CPU 216 * @crycb_s: pointer to shadow crycb 217 * @crycb_gpa: guest physical address of original guest crycb 218 * @crycb_h: pointer to the host crycb 219 * @fmt_o: format of the original guest crycb. 220 * @fmt_h: format of the host crycb. 221 * 222 * Checks the compatibility between the guest and host crycb and calls the 223 * appropriate copy function. 224 * 225 * Return 0 or an error number if the guest and host crycb are incompatible. 226 */ 227 static int setup_apcb(struct kvm_vcpu *vcpu, struct kvm_s390_crypto_cb *crycb_s, 228 const u32 crycb_gpa, 229 struct kvm_s390_crypto_cb *crycb_h, 230 int fmt_o, int fmt_h) 231 { 232 switch (fmt_o) { 233 case CRYCB_FORMAT2: 234 if ((crycb_gpa & PAGE_MASK) != ((crycb_gpa + 256) & PAGE_MASK)) 235 return -EACCES; 236 if (fmt_h != CRYCB_FORMAT2) 237 return -EINVAL; 238 return setup_apcb11(vcpu, (unsigned long *)&crycb_s->apcb1, 239 crycb_gpa, 240 (unsigned long *)&crycb_h->apcb1); 241 case CRYCB_FORMAT1: 242 switch (fmt_h) { 243 case CRYCB_FORMAT2: 244 return setup_apcb10(vcpu, &crycb_s->apcb1, 245 crycb_gpa, 246 &crycb_h->apcb1); 247 case CRYCB_FORMAT1: 248 return setup_apcb00(vcpu, 249 (unsigned long *) &crycb_s->apcb0, 250 crycb_gpa, 251 (unsigned long *) &crycb_h->apcb0); 252 } 253 break; 254 case CRYCB_FORMAT0: 255 if ((crycb_gpa & PAGE_MASK) != ((crycb_gpa + 32) & PAGE_MASK)) 256 return -EACCES; 257 258 switch (fmt_h) { 259 case CRYCB_FORMAT2: 260 return setup_apcb10(vcpu, &crycb_s->apcb1, 261 crycb_gpa, 262 &crycb_h->apcb1); 263 case CRYCB_FORMAT1: 264 case CRYCB_FORMAT0: 265 return setup_apcb00(vcpu, 266 (unsigned long *) &crycb_s->apcb0, 267 crycb_gpa, 268 (unsigned long *) &crycb_h->apcb0); 269 } 270 } 271 return -EINVAL; 272 } 273 274 /** 275 * shadow_crycb - Create a shadow copy of the crycb block 276 * @vcpu: a pointer to the virtual CPU 277 * @vsie_page: a pointer to internal date used for the vSIE 278 * 279 * Create a shadow copy of the crycb block and setup key wrapping, if 280 * requested for guest 3 and enabled for guest 2. 281 * 282 * We accept format-1 or format-2, but we convert format-1 into format-2 283 * in the shadow CRYCB. 284 * Using format-2 enables the firmware to choose the right format when 285 * scheduling the SIE. 286 * There is nothing to do for format-0. 287 * 288 * This function centralize the issuing of set_validity_icpt() for all 289 * the subfunctions working on the crycb. 290 * 291 * Returns: - 0 if shadowed or nothing to do 292 * - > 0 if control has to be given to guest 2 293 */ 294 static int shadow_crycb(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page) 295 { 296 struct kvm_s390_sie_block *scb_s = &vsie_page->scb_s; 297 struct kvm_s390_sie_block *scb_o = vsie_page->scb_o; 298 const uint32_t crycbd_o = READ_ONCE(scb_o->crycbd); 299 const u32 crycb_addr = crycbd_o & 0x7ffffff8U; 300 unsigned long *b1, *b2; 301 u8 ecb3_flags; 302 u32 ecd_flags; 303 int apie_h; 304 int apie_s; 305 int key_msk = test_kvm_facility(vcpu->kvm, 76); 306 int fmt_o = crycbd_o & CRYCB_FORMAT_MASK; 307 int fmt_h = vcpu->arch.sie_block->crycbd & CRYCB_FORMAT_MASK; 308 int ret = 0; 309 310 scb_s->crycbd = 0; 311 312 apie_h = vcpu->arch.sie_block->eca & ECA_APIE; 313 apie_s = apie_h & scb_o->eca; 314 if (!apie_s && (!key_msk || (fmt_o == CRYCB_FORMAT0))) 315 return 0; 316 317 if (!crycb_addr) 318 return set_validity_icpt(scb_s, 0x0039U); 319 320 if (fmt_o == CRYCB_FORMAT1) 321 if ((crycb_addr & PAGE_MASK) != 322 ((crycb_addr + 128) & PAGE_MASK)) 323 return set_validity_icpt(scb_s, 0x003CU); 324 325 if (apie_s) { 326 ret = setup_apcb(vcpu, &vsie_page->crycb, crycb_addr, 327 vcpu->kvm->arch.crypto.crycb, 328 fmt_o, fmt_h); 329 if (ret) 330 goto end; 331 scb_s->eca |= scb_o->eca & ECA_APIE; 332 } 333 334 /* we may only allow it if enabled for guest 2 */ 335 ecb3_flags = scb_o->ecb3 & vcpu->arch.sie_block->ecb3 & 336 (ECB3_AES | ECB3_DEA); 337 ecd_flags = scb_o->ecd & vcpu->arch.sie_block->ecd & ECD_ECC; 338 if (!ecb3_flags && !ecd_flags) 339 goto end; 340 341 /* copy only the wrapping keys */ 342 if (read_guest_real(vcpu, crycb_addr + 72, 343 vsie_page->crycb.dea_wrapping_key_mask, 56)) 344 return set_validity_icpt(scb_s, 0x0035U); 345 346 scb_s->ecb3 |= ecb3_flags; 347 scb_s->ecd |= ecd_flags; 348 349 /* xor both blocks in one run */ 350 b1 = (unsigned long *) vsie_page->crycb.dea_wrapping_key_mask; 351 b2 = (unsigned long *) 352 vcpu->kvm->arch.crypto.crycb->dea_wrapping_key_mask; 353 /* as 56%8 == 0, bitmap_xor won't overwrite any data */ 354 bitmap_xor(b1, b1, b2, BITS_PER_BYTE * 56); 355 end: 356 switch (ret) { 357 case -EINVAL: 358 return set_validity_icpt(scb_s, 0x0022U); 359 case -EFAULT: 360 return set_validity_icpt(scb_s, 0x0035U); 361 case -EACCES: 362 return set_validity_icpt(scb_s, 0x003CU); 363 } 364 scb_s->crycbd = ((__u32)(__u64) &vsie_page->crycb) | CRYCB_FORMAT2; 365 return 0; 366 } 367 368 /* shadow (round up/down) the ibc to avoid validity icpt */ 369 static void prepare_ibc(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page) 370 { 371 struct kvm_s390_sie_block *scb_s = &vsie_page->scb_s; 372 struct kvm_s390_sie_block *scb_o = vsie_page->scb_o; 373 /* READ_ONCE does not work on bitfields - use a temporary variable */ 374 const uint32_t __new_ibc = scb_o->ibc; 375 const uint32_t new_ibc = READ_ONCE(__new_ibc) & 0x0fffU; 376 __u64 min_ibc = (sclp.ibc >> 16) & 0x0fffU; 377 378 scb_s->ibc = 0; 379 /* ibc installed in g2 and requested for g3 */ 380 if (vcpu->kvm->arch.model.ibc && new_ibc) { 381 scb_s->ibc = new_ibc; 382 /* takte care of the minimum ibc level of the machine */ 383 if (scb_s->ibc < min_ibc) 384 scb_s->ibc = min_ibc; 385 /* take care of the maximum ibc level set for the guest */ 386 if (scb_s->ibc > vcpu->kvm->arch.model.ibc) 387 scb_s->ibc = vcpu->kvm->arch.model.ibc; 388 } 389 } 390 391 /* unshadow the scb, copying parameters back to the real scb */ 392 static void unshadow_scb(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page) 393 { 394 struct kvm_s390_sie_block *scb_s = &vsie_page->scb_s; 395 struct kvm_s390_sie_block *scb_o = vsie_page->scb_o; 396 397 /* interception */ 398 scb_o->icptcode = scb_s->icptcode; 399 scb_o->icptstatus = scb_s->icptstatus; 400 scb_o->ipa = scb_s->ipa; 401 scb_o->ipb = scb_s->ipb; 402 scb_o->gbea = scb_s->gbea; 403 404 /* timer */ 405 scb_o->cputm = scb_s->cputm; 406 scb_o->ckc = scb_s->ckc; 407 scb_o->todpr = scb_s->todpr; 408 409 /* guest state */ 410 scb_o->gpsw = scb_s->gpsw; 411 scb_o->gg14 = scb_s->gg14; 412 scb_o->gg15 = scb_s->gg15; 413 memcpy(scb_o->gcr, scb_s->gcr, 128); 414 scb_o->pp = scb_s->pp; 415 416 /* branch prediction */ 417 if (test_kvm_facility(vcpu->kvm, 82)) { 418 scb_o->fpf &= ~FPF_BPBC; 419 scb_o->fpf |= scb_s->fpf & FPF_BPBC; 420 } 421 422 /* interrupt intercept */ 423 switch (scb_s->icptcode) { 424 case ICPT_PROGI: 425 case ICPT_INSTPROGI: 426 case ICPT_EXTINT: 427 memcpy((void *)((u64)scb_o + 0xc0), 428 (void *)((u64)scb_s + 0xc0), 0xf0 - 0xc0); 429 break; 430 } 431 432 if (scb_s->ihcpu != 0xffffU) 433 scb_o->ihcpu = scb_s->ihcpu; 434 } 435 436 /* 437 * Setup the shadow scb by copying and checking the relevant parts of the g2 438 * provided scb. 439 * 440 * Returns: - 0 if the scb has been shadowed 441 * - > 0 if control has to be given to guest 2 442 */ 443 static int shadow_scb(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page) 444 { 445 struct kvm_s390_sie_block *scb_o = vsie_page->scb_o; 446 struct kvm_s390_sie_block *scb_s = &vsie_page->scb_s; 447 /* READ_ONCE does not work on bitfields - use a temporary variable */ 448 const uint32_t __new_prefix = scb_o->prefix; 449 const uint32_t new_prefix = READ_ONCE(__new_prefix); 450 const bool wants_tx = READ_ONCE(scb_o->ecb) & ECB_TE; 451 bool had_tx = scb_s->ecb & ECB_TE; 452 unsigned long new_mso = 0; 453 int rc; 454 455 /* make sure we don't have any leftovers when reusing the scb */ 456 scb_s->icptcode = 0; 457 scb_s->eca = 0; 458 scb_s->ecb = 0; 459 scb_s->ecb2 = 0; 460 scb_s->ecb3 = 0; 461 scb_s->ecd = 0; 462 scb_s->fac = 0; 463 scb_s->fpf = 0; 464 465 rc = prepare_cpuflags(vcpu, vsie_page); 466 if (rc) 467 goto out; 468 469 /* timer */ 470 scb_s->cputm = scb_o->cputm; 471 scb_s->ckc = scb_o->ckc; 472 scb_s->todpr = scb_o->todpr; 473 scb_s->epoch = scb_o->epoch; 474 475 /* guest state */ 476 scb_s->gpsw = scb_o->gpsw; 477 scb_s->gg14 = scb_o->gg14; 478 scb_s->gg15 = scb_o->gg15; 479 memcpy(scb_s->gcr, scb_o->gcr, 128); 480 scb_s->pp = scb_o->pp; 481 482 /* interception / execution handling */ 483 scb_s->gbea = scb_o->gbea; 484 scb_s->lctl = scb_o->lctl; 485 scb_s->svcc = scb_o->svcc; 486 scb_s->ictl = scb_o->ictl; 487 /* 488 * SKEY handling functions can't deal with false setting of PTE invalid 489 * bits. Therefore we cannot provide interpretation and would later 490 * have to provide own emulation handlers. 491 */ 492 if (!(atomic_read(&scb_s->cpuflags) & CPUSTAT_KSS)) 493 scb_s->ictl |= ICTL_ISKE | ICTL_SSKE | ICTL_RRBE; 494 495 scb_s->icpua = scb_o->icpua; 496 497 if (!(atomic_read(&scb_s->cpuflags) & CPUSTAT_SM)) 498 new_mso = READ_ONCE(scb_o->mso) & 0xfffffffffff00000UL; 499 /* if the hva of the prefix changes, we have to remap the prefix */ 500 if (scb_s->mso != new_mso || scb_s->prefix != new_prefix) 501 prefix_unmapped(vsie_page); 502 /* SIE will do mso/msl validity and exception checks for us */ 503 scb_s->msl = scb_o->msl & 0xfffffffffff00000UL; 504 scb_s->mso = new_mso; 505 scb_s->prefix = new_prefix; 506 507 /* We have to definitely flush the tlb if this scb never ran */ 508 if (scb_s->ihcpu != 0xffffU) 509 scb_s->ihcpu = scb_o->ihcpu; 510 511 /* MVPG and Protection Exception Interpretation are always available */ 512 scb_s->eca |= scb_o->eca & (ECA_MVPGI | ECA_PROTEXCI); 513 /* Host-protection-interruption introduced with ESOP */ 514 if (test_kvm_cpu_feat(vcpu->kvm, KVM_S390_VM_CPU_FEAT_ESOP)) 515 scb_s->ecb |= scb_o->ecb & ECB_HOSTPROTINT; 516 /* 517 * CPU Topology 518 * This facility only uses the utility field of the SCA and none of 519 * the cpu entries that are problematic with the other interpretation 520 * facilities so we can pass it through 521 */ 522 if (test_kvm_facility(vcpu->kvm, 11)) 523 scb_s->ecb |= scb_o->ecb & ECB_PTF; 524 /* transactional execution */ 525 if (test_kvm_facility(vcpu->kvm, 73) && wants_tx) { 526 /* remap the prefix is tx is toggled on */ 527 if (!had_tx) 528 prefix_unmapped(vsie_page); 529 scb_s->ecb |= ECB_TE; 530 } 531 /* specification exception interpretation */ 532 scb_s->ecb |= scb_o->ecb & ECB_SPECI; 533 /* branch prediction */ 534 if (test_kvm_facility(vcpu->kvm, 82)) 535 scb_s->fpf |= scb_o->fpf & FPF_BPBC; 536 /* SIMD */ 537 if (test_kvm_facility(vcpu->kvm, 129)) { 538 scb_s->eca |= scb_o->eca & ECA_VX; 539 scb_s->ecd |= scb_o->ecd & ECD_HOSTREGMGMT; 540 } 541 /* Run-time-Instrumentation */ 542 if (test_kvm_facility(vcpu->kvm, 64)) 543 scb_s->ecb3 |= scb_o->ecb3 & ECB3_RI; 544 /* Instruction Execution Prevention */ 545 if (test_kvm_facility(vcpu->kvm, 130)) 546 scb_s->ecb2 |= scb_o->ecb2 & ECB2_IEP; 547 /* Guarded Storage */ 548 if (test_kvm_facility(vcpu->kvm, 133)) { 549 scb_s->ecb |= scb_o->ecb & ECB_GS; 550 scb_s->ecd |= scb_o->ecd & ECD_HOSTREGMGMT; 551 } 552 if (test_kvm_cpu_feat(vcpu->kvm, KVM_S390_VM_CPU_FEAT_SIIF)) 553 scb_s->eca |= scb_o->eca & ECA_SII; 554 if (test_kvm_cpu_feat(vcpu->kvm, KVM_S390_VM_CPU_FEAT_IB)) 555 scb_s->eca |= scb_o->eca & ECA_IB; 556 if (test_kvm_cpu_feat(vcpu->kvm, KVM_S390_VM_CPU_FEAT_CEI)) 557 scb_s->eca |= scb_o->eca & ECA_CEI; 558 /* Epoch Extension */ 559 if (test_kvm_facility(vcpu->kvm, 139)) { 560 scb_s->ecd |= scb_o->ecd & ECD_MEF; 561 scb_s->epdx = scb_o->epdx; 562 } 563 564 /* etoken */ 565 if (test_kvm_facility(vcpu->kvm, 156)) 566 scb_s->ecd |= scb_o->ecd & ECD_ETOKENF; 567 568 scb_s->hpid = HPID_VSIE; 569 scb_s->cpnc = scb_o->cpnc; 570 571 prepare_ibc(vcpu, vsie_page); 572 rc = shadow_crycb(vcpu, vsie_page); 573 out: 574 if (rc) 575 unshadow_scb(vcpu, vsie_page); 576 return rc; 577 } 578 579 void kvm_s390_vsie_gmap_notifier(struct gmap *gmap, unsigned long start, 580 unsigned long end) 581 { 582 struct kvm *kvm = gmap->private; 583 struct vsie_page *cur; 584 unsigned long prefix; 585 struct page *page; 586 int i; 587 588 if (!gmap_is_shadow(gmap)) 589 return; 590 /* 591 * Only new shadow blocks are added to the list during runtime, 592 * therefore we can safely reference them all the time. 593 */ 594 for (i = 0; i < kvm->arch.vsie.page_count; i++) { 595 page = READ_ONCE(kvm->arch.vsie.pages[i]); 596 if (!page) 597 continue; 598 cur = page_to_virt(page); 599 if (READ_ONCE(cur->gmap) != gmap) 600 continue; 601 prefix = cur->scb_s.prefix << GUEST_PREFIX_SHIFT; 602 /* with mso/msl, the prefix lies at an offset */ 603 prefix += cur->scb_s.mso; 604 if (prefix <= end && start <= prefix + 2 * PAGE_SIZE - 1) 605 prefix_unmapped_sync(cur); 606 } 607 } 608 609 /* 610 * Map the first prefix page and if tx is enabled also the second prefix page. 611 * 612 * The prefix will be protected, a gmap notifier will inform about unmaps. 613 * The shadow scb must not be executed until the prefix is remapped, this is 614 * guaranteed by properly handling PROG_REQUEST. 615 * 616 * Returns: - 0 on if successfully mapped or already mapped 617 * - > 0 if control has to be given to guest 2 618 * - -EAGAIN if the caller can retry immediately 619 * - -ENOMEM if out of memory 620 */ 621 static int map_prefix(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page) 622 { 623 struct kvm_s390_sie_block *scb_s = &vsie_page->scb_s; 624 u64 prefix = scb_s->prefix << GUEST_PREFIX_SHIFT; 625 int rc; 626 627 if (prefix_is_mapped(vsie_page)) 628 return 0; 629 630 /* mark it as mapped so we can catch any concurrent unmappers */ 631 prefix_mapped(vsie_page); 632 633 /* with mso/msl, the prefix lies at offset *mso* */ 634 prefix += scb_s->mso; 635 636 rc = kvm_s390_shadow_fault(vcpu, vsie_page->gmap, prefix, NULL); 637 if (!rc && (scb_s->ecb & ECB_TE)) 638 rc = kvm_s390_shadow_fault(vcpu, vsie_page->gmap, 639 prefix + PAGE_SIZE, NULL); 640 /* 641 * We don't have to mprotect, we will be called for all unshadows. 642 * SIE will detect if protection applies and trigger a validity. 643 */ 644 if (rc) 645 prefix_unmapped(vsie_page); 646 if (rc > 0 || rc == -EFAULT) 647 rc = set_validity_icpt(scb_s, 0x0037U); 648 return rc; 649 } 650 651 /* 652 * Pin the guest page given by gpa and set hpa to the pinned host address. 653 * Will always be pinned writable. 654 * 655 * Returns: - 0 on success 656 * - -EINVAL if the gpa is not valid guest storage 657 */ 658 static int pin_guest_page(struct kvm *kvm, gpa_t gpa, hpa_t *hpa) 659 { 660 struct page *page; 661 662 page = gfn_to_page(kvm, gpa_to_gfn(gpa)); 663 if (is_error_page(page)) 664 return -EINVAL; 665 *hpa = (hpa_t)page_to_phys(page) + (gpa & ~PAGE_MASK); 666 return 0; 667 } 668 669 /* Unpins a page previously pinned via pin_guest_page, marking it as dirty. */ 670 static void unpin_guest_page(struct kvm *kvm, gpa_t gpa, hpa_t hpa) 671 { 672 kvm_release_pfn_dirty(hpa >> PAGE_SHIFT); 673 /* mark the page always as dirty for migration */ 674 mark_page_dirty(kvm, gpa_to_gfn(gpa)); 675 } 676 677 /* unpin all blocks previously pinned by pin_blocks(), marking them dirty */ 678 static void unpin_blocks(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page) 679 { 680 struct kvm_s390_sie_block *scb_s = &vsie_page->scb_s; 681 hpa_t hpa; 682 683 hpa = (u64) scb_s->scaoh << 32 | scb_s->scaol; 684 if (hpa) { 685 unpin_guest_page(vcpu->kvm, vsie_page->sca_gpa, hpa); 686 vsie_page->sca_gpa = 0; 687 scb_s->scaol = 0; 688 scb_s->scaoh = 0; 689 } 690 691 hpa = scb_s->itdba; 692 if (hpa) { 693 unpin_guest_page(vcpu->kvm, vsie_page->itdba_gpa, hpa); 694 vsie_page->itdba_gpa = 0; 695 scb_s->itdba = 0; 696 } 697 698 hpa = scb_s->gvrd; 699 if (hpa) { 700 unpin_guest_page(vcpu->kvm, vsie_page->gvrd_gpa, hpa); 701 vsie_page->gvrd_gpa = 0; 702 scb_s->gvrd = 0; 703 } 704 705 hpa = scb_s->riccbd; 706 if (hpa) { 707 unpin_guest_page(vcpu->kvm, vsie_page->riccbd_gpa, hpa); 708 vsie_page->riccbd_gpa = 0; 709 scb_s->riccbd = 0; 710 } 711 712 hpa = scb_s->sdnxo; 713 if (hpa) { 714 unpin_guest_page(vcpu->kvm, vsie_page->sdnx_gpa, hpa); 715 vsie_page->sdnx_gpa = 0; 716 scb_s->sdnxo = 0; 717 } 718 } 719 720 /* 721 * Instead of shadowing some blocks, we can simply forward them because the 722 * addresses in the scb are 64 bit long. 723 * 724 * This works as long as the data lies in one page. If blocks ever exceed one 725 * page, we have to fall back to shadowing. 726 * 727 * As we reuse the sca, the vcpu pointers contained in it are invalid. We must 728 * therefore not enable any facilities that access these pointers (e.g. SIGPIF). 729 * 730 * Returns: - 0 if all blocks were pinned. 731 * - > 0 if control has to be given to guest 2 732 * - -ENOMEM if out of memory 733 */ 734 static int pin_blocks(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page) 735 { 736 struct kvm_s390_sie_block *scb_o = vsie_page->scb_o; 737 struct kvm_s390_sie_block *scb_s = &vsie_page->scb_s; 738 hpa_t hpa; 739 gpa_t gpa; 740 int rc = 0; 741 742 gpa = READ_ONCE(scb_o->scaol) & ~0xfUL; 743 if (test_kvm_cpu_feat(vcpu->kvm, KVM_S390_VM_CPU_FEAT_64BSCAO)) 744 gpa |= (u64) READ_ONCE(scb_o->scaoh) << 32; 745 if (gpa) { 746 if (gpa < 2 * PAGE_SIZE) 747 rc = set_validity_icpt(scb_s, 0x0038U); 748 else if ((gpa & ~0x1fffUL) == kvm_s390_get_prefix(vcpu)) 749 rc = set_validity_icpt(scb_s, 0x0011U); 750 else if ((gpa & PAGE_MASK) != 751 ((gpa + sizeof(struct bsca_block) - 1) & PAGE_MASK)) 752 rc = set_validity_icpt(scb_s, 0x003bU); 753 if (!rc) { 754 rc = pin_guest_page(vcpu->kvm, gpa, &hpa); 755 if (rc) 756 rc = set_validity_icpt(scb_s, 0x0034U); 757 } 758 if (rc) 759 goto unpin; 760 vsie_page->sca_gpa = gpa; 761 scb_s->scaoh = (u32)((u64)hpa >> 32); 762 scb_s->scaol = (u32)(u64)hpa; 763 } 764 765 gpa = READ_ONCE(scb_o->itdba) & ~0xffUL; 766 if (gpa && (scb_s->ecb & ECB_TE)) { 767 if (gpa < 2 * PAGE_SIZE) { 768 rc = set_validity_icpt(scb_s, 0x0080U); 769 goto unpin; 770 } 771 /* 256 bytes cannot cross page boundaries */ 772 rc = pin_guest_page(vcpu->kvm, gpa, &hpa); 773 if (rc) { 774 rc = set_validity_icpt(scb_s, 0x0080U); 775 goto unpin; 776 } 777 vsie_page->itdba_gpa = gpa; 778 scb_s->itdba = hpa; 779 } 780 781 gpa = READ_ONCE(scb_o->gvrd) & ~0x1ffUL; 782 if (gpa && (scb_s->eca & ECA_VX) && !(scb_s->ecd & ECD_HOSTREGMGMT)) { 783 if (gpa < 2 * PAGE_SIZE) { 784 rc = set_validity_icpt(scb_s, 0x1310U); 785 goto unpin; 786 } 787 /* 788 * 512 bytes vector registers cannot cross page boundaries 789 * if this block gets bigger, we have to shadow it. 790 */ 791 rc = pin_guest_page(vcpu->kvm, gpa, &hpa); 792 if (rc) { 793 rc = set_validity_icpt(scb_s, 0x1310U); 794 goto unpin; 795 } 796 vsie_page->gvrd_gpa = gpa; 797 scb_s->gvrd = hpa; 798 } 799 800 gpa = READ_ONCE(scb_o->riccbd) & ~0x3fUL; 801 if (gpa && (scb_s->ecb3 & ECB3_RI)) { 802 if (gpa < 2 * PAGE_SIZE) { 803 rc = set_validity_icpt(scb_s, 0x0043U); 804 goto unpin; 805 } 806 /* 64 bytes cannot cross page boundaries */ 807 rc = pin_guest_page(vcpu->kvm, gpa, &hpa); 808 if (rc) { 809 rc = set_validity_icpt(scb_s, 0x0043U); 810 goto unpin; 811 } 812 /* Validity 0x0044 will be checked by SIE */ 813 vsie_page->riccbd_gpa = gpa; 814 scb_s->riccbd = hpa; 815 } 816 if (((scb_s->ecb & ECB_GS) && !(scb_s->ecd & ECD_HOSTREGMGMT)) || 817 (scb_s->ecd & ECD_ETOKENF)) { 818 unsigned long sdnxc; 819 820 gpa = READ_ONCE(scb_o->sdnxo) & ~0xfUL; 821 sdnxc = READ_ONCE(scb_o->sdnxo) & 0xfUL; 822 if (!gpa || gpa < 2 * PAGE_SIZE) { 823 rc = set_validity_icpt(scb_s, 0x10b0U); 824 goto unpin; 825 } 826 if (sdnxc < 6 || sdnxc > 12) { 827 rc = set_validity_icpt(scb_s, 0x10b1U); 828 goto unpin; 829 } 830 if (gpa & ((1 << sdnxc) - 1)) { 831 rc = set_validity_icpt(scb_s, 0x10b2U); 832 goto unpin; 833 } 834 /* Due to alignment rules (checked above) this cannot 835 * cross page boundaries 836 */ 837 rc = pin_guest_page(vcpu->kvm, gpa, &hpa); 838 if (rc) { 839 rc = set_validity_icpt(scb_s, 0x10b0U); 840 goto unpin; 841 } 842 vsie_page->sdnx_gpa = gpa; 843 scb_s->sdnxo = hpa | sdnxc; 844 } 845 return 0; 846 unpin: 847 unpin_blocks(vcpu, vsie_page); 848 return rc; 849 } 850 851 /* unpin the scb provided by guest 2, marking it as dirty */ 852 static void unpin_scb(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page, 853 gpa_t gpa) 854 { 855 hpa_t hpa = (hpa_t) vsie_page->scb_o; 856 857 if (hpa) 858 unpin_guest_page(vcpu->kvm, gpa, hpa); 859 vsie_page->scb_o = NULL; 860 } 861 862 /* 863 * Pin the scb at gpa provided by guest 2 at vsie_page->scb_o. 864 * 865 * Returns: - 0 if the scb was pinned. 866 * - > 0 if control has to be given to guest 2 867 */ 868 static int pin_scb(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page, 869 gpa_t gpa) 870 { 871 hpa_t hpa; 872 int rc; 873 874 rc = pin_guest_page(vcpu->kvm, gpa, &hpa); 875 if (rc) { 876 rc = kvm_s390_inject_program_int(vcpu, PGM_ADDRESSING); 877 WARN_ON_ONCE(rc); 878 return 1; 879 } 880 vsie_page->scb_o = phys_to_virt(hpa); 881 return 0; 882 } 883 884 /* 885 * Inject a fault into guest 2. 886 * 887 * Returns: - > 0 if control has to be given to guest 2 888 * < 0 if an error occurred during injection. 889 */ 890 static int inject_fault(struct kvm_vcpu *vcpu, __u16 code, __u64 vaddr, 891 bool write_flag) 892 { 893 struct kvm_s390_pgm_info pgm = { 894 .code = code, 895 .trans_exc_code = 896 /* 0-51: virtual address */ 897 (vaddr & 0xfffffffffffff000UL) | 898 /* 52-53: store / fetch */ 899 (((unsigned int) !write_flag) + 1) << 10, 900 /* 62-63: asce id (always primary == 0) */ 901 .exc_access_id = 0, /* always primary */ 902 .op_access_id = 0, /* not MVPG */ 903 }; 904 int rc; 905 906 if (code == PGM_PROTECTION) 907 pgm.trans_exc_code |= 0x4UL; 908 909 rc = kvm_s390_inject_prog_irq(vcpu, &pgm); 910 return rc ? rc : 1; 911 } 912 913 /* 914 * Handle a fault during vsie execution on a gmap shadow. 915 * 916 * Returns: - 0 if the fault was resolved 917 * - > 0 if control has to be given to guest 2 918 * - < 0 if an error occurred 919 */ 920 static int handle_fault(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page) 921 { 922 int rc; 923 924 if (current->thread.gmap_int_code == PGM_PROTECTION) 925 /* we can directly forward all protection exceptions */ 926 return inject_fault(vcpu, PGM_PROTECTION, 927 current->thread.gmap_addr, 1); 928 929 rc = kvm_s390_shadow_fault(vcpu, vsie_page->gmap, 930 current->thread.gmap_addr, NULL); 931 if (rc > 0) { 932 rc = inject_fault(vcpu, rc, 933 current->thread.gmap_addr, 934 current->thread.gmap_write_flag); 935 if (rc >= 0) 936 vsie_page->fault_addr = current->thread.gmap_addr; 937 } 938 return rc; 939 } 940 941 /* 942 * Retry the previous fault that required guest 2 intervention. This avoids 943 * one superfluous SIE re-entry and direct exit. 944 * 945 * Will ignore any errors. The next SIE fault will do proper fault handling. 946 */ 947 static void handle_last_fault(struct kvm_vcpu *vcpu, 948 struct vsie_page *vsie_page) 949 { 950 if (vsie_page->fault_addr) 951 kvm_s390_shadow_fault(vcpu, vsie_page->gmap, 952 vsie_page->fault_addr, NULL); 953 vsie_page->fault_addr = 0; 954 } 955 956 static inline void clear_vsie_icpt(struct vsie_page *vsie_page) 957 { 958 vsie_page->scb_s.icptcode = 0; 959 } 960 961 /* rewind the psw and clear the vsie icpt, so we can retry execution */ 962 static void retry_vsie_icpt(struct vsie_page *vsie_page) 963 { 964 struct kvm_s390_sie_block *scb_s = &vsie_page->scb_s; 965 int ilen = insn_length(scb_s->ipa >> 8); 966 967 /* take care of EXECUTE instructions */ 968 if (scb_s->icptstatus & 1) { 969 ilen = (scb_s->icptstatus >> 4) & 0x6; 970 if (!ilen) 971 ilen = 4; 972 } 973 scb_s->gpsw.addr = __rewind_psw(scb_s->gpsw, ilen); 974 clear_vsie_icpt(vsie_page); 975 } 976 977 /* 978 * Try to shadow + enable the guest 2 provided facility list. 979 * Retry instruction execution if enabled for and provided by guest 2. 980 * 981 * Returns: - 0 if handled (retry or guest 2 icpt) 982 * - > 0 if control has to be given to guest 2 983 */ 984 static int handle_stfle(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page) 985 { 986 struct kvm_s390_sie_block *scb_s = &vsie_page->scb_s; 987 __u32 fac = READ_ONCE(vsie_page->scb_o->fac); 988 989 /* 990 * Alternate-STFLE-Interpretive-Execution facilities are not supported 991 * -> format-0 flcb 992 */ 993 if (fac && test_kvm_facility(vcpu->kvm, 7)) { 994 retry_vsie_icpt(vsie_page); 995 /* 996 * The facility list origin (FLO) is in bits 1 - 28 of the FLD 997 * so we need to mask here before reading. 998 */ 999 fac = fac & 0x7ffffff8U; 1000 /* 1001 * format-0 -> size of nested guest's facility list == guest's size 1002 * guest's size == host's size, since STFLE is interpretatively executed 1003 * using a format-0 for the guest, too. 1004 */ 1005 if (read_guest_real(vcpu, fac, &vsie_page->fac, 1006 stfle_size() * sizeof(u64))) 1007 return set_validity_icpt(scb_s, 0x1090U); 1008 scb_s->fac = (__u32)(__u64) &vsie_page->fac; 1009 } 1010 return 0; 1011 } 1012 1013 /* 1014 * Get a register for a nested guest. 1015 * @vcpu the vcpu of the guest 1016 * @vsie_page the vsie_page for the nested guest 1017 * @reg the register number, the upper 4 bits are ignored. 1018 * returns: the value of the register. 1019 */ 1020 static u64 vsie_get_register(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page, u8 reg) 1021 { 1022 /* no need to validate the parameter and/or perform error handling */ 1023 reg &= 0xf; 1024 switch (reg) { 1025 case 15: 1026 return vsie_page->scb_s.gg15; 1027 case 14: 1028 return vsie_page->scb_s.gg14; 1029 default: 1030 return vcpu->run->s.regs.gprs[reg]; 1031 } 1032 } 1033 1034 static int vsie_handle_mvpg(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page) 1035 { 1036 struct kvm_s390_sie_block *scb_s = &vsie_page->scb_s; 1037 unsigned long pei_dest, pei_src, src, dest, mask, prefix; 1038 u64 *pei_block = &vsie_page->scb_o->mcic; 1039 int edat, rc_dest, rc_src; 1040 union ctlreg0 cr0; 1041 1042 cr0.val = vcpu->arch.sie_block->gcr[0]; 1043 edat = cr0.edat && test_kvm_facility(vcpu->kvm, 8); 1044 mask = _kvm_s390_logical_to_effective(&scb_s->gpsw, PAGE_MASK); 1045 prefix = scb_s->prefix << GUEST_PREFIX_SHIFT; 1046 1047 dest = vsie_get_register(vcpu, vsie_page, scb_s->ipb >> 20) & mask; 1048 dest = _kvm_s390_real_to_abs(prefix, dest) + scb_s->mso; 1049 src = vsie_get_register(vcpu, vsie_page, scb_s->ipb >> 16) & mask; 1050 src = _kvm_s390_real_to_abs(prefix, src) + scb_s->mso; 1051 1052 rc_dest = kvm_s390_shadow_fault(vcpu, vsie_page->gmap, dest, &pei_dest); 1053 rc_src = kvm_s390_shadow_fault(vcpu, vsie_page->gmap, src, &pei_src); 1054 /* 1055 * Either everything went well, or something non-critical went wrong 1056 * e.g. because of a race. In either case, simply retry. 1057 */ 1058 if (rc_dest == -EAGAIN || rc_src == -EAGAIN || (!rc_dest && !rc_src)) { 1059 retry_vsie_icpt(vsie_page); 1060 return -EAGAIN; 1061 } 1062 /* Something more serious went wrong, propagate the error */ 1063 if (rc_dest < 0) 1064 return rc_dest; 1065 if (rc_src < 0) 1066 return rc_src; 1067 1068 /* The only possible suppressing exception: just deliver it */ 1069 if (rc_dest == PGM_TRANSLATION_SPEC || rc_src == PGM_TRANSLATION_SPEC) { 1070 clear_vsie_icpt(vsie_page); 1071 rc_dest = kvm_s390_inject_program_int(vcpu, PGM_TRANSLATION_SPEC); 1072 WARN_ON_ONCE(rc_dest); 1073 return 1; 1074 } 1075 1076 /* 1077 * Forward the PEI intercept to the guest if it was a page fault, or 1078 * also for segment and region table faults if EDAT applies. 1079 */ 1080 if (edat) { 1081 rc_dest = rc_dest == PGM_ASCE_TYPE ? rc_dest : 0; 1082 rc_src = rc_src == PGM_ASCE_TYPE ? rc_src : 0; 1083 } else { 1084 rc_dest = rc_dest != PGM_PAGE_TRANSLATION ? rc_dest : 0; 1085 rc_src = rc_src != PGM_PAGE_TRANSLATION ? rc_src : 0; 1086 } 1087 if (!rc_dest && !rc_src) { 1088 pei_block[0] = pei_dest; 1089 pei_block[1] = pei_src; 1090 return 1; 1091 } 1092 1093 retry_vsie_icpt(vsie_page); 1094 1095 /* 1096 * The host has edat, and the guest does not, or it was an ASCE type 1097 * exception. The host needs to inject the appropriate DAT interrupts 1098 * into the guest. 1099 */ 1100 if (rc_dest) 1101 return inject_fault(vcpu, rc_dest, dest, 1); 1102 return inject_fault(vcpu, rc_src, src, 0); 1103 } 1104 1105 /* 1106 * Run the vsie on a shadow scb and a shadow gmap, without any further 1107 * sanity checks, handling SIE faults. 1108 * 1109 * Returns: - 0 everything went fine 1110 * - > 0 if control has to be given to guest 2 1111 * - < 0 if an error occurred 1112 */ 1113 static int do_vsie_run(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page) 1114 __releases(vcpu->kvm->srcu) 1115 __acquires(vcpu->kvm->srcu) 1116 { 1117 struct kvm_s390_sie_block *scb_s = &vsie_page->scb_s; 1118 struct kvm_s390_sie_block *scb_o = vsie_page->scb_o; 1119 int guest_bp_isolation; 1120 int rc = 0; 1121 1122 handle_last_fault(vcpu, vsie_page); 1123 1124 kvm_vcpu_srcu_read_unlock(vcpu); 1125 1126 /* save current guest state of bp isolation override */ 1127 guest_bp_isolation = test_thread_flag(TIF_ISOLATE_BP_GUEST); 1128 1129 /* 1130 * The guest is running with BPBC, so we have to force it on for our 1131 * nested guest. This is done by enabling BPBC globally, so the BPBC 1132 * control in the SCB (which the nested guest can modify) is simply 1133 * ignored. 1134 */ 1135 if (test_kvm_facility(vcpu->kvm, 82) && 1136 vcpu->arch.sie_block->fpf & FPF_BPBC) 1137 set_thread_flag(TIF_ISOLATE_BP_GUEST); 1138 1139 local_irq_disable(); 1140 guest_enter_irqoff(); 1141 local_irq_enable(); 1142 1143 /* 1144 * Simulate a SIE entry of the VCPU (see sie64a), so VCPU blocking 1145 * and VCPU requests also hinder the vSIE from running and lead 1146 * to an immediate exit. kvm_s390_vsie_kick() has to be used to 1147 * also kick the vSIE. 1148 */ 1149 vcpu->arch.sie_block->prog0c |= PROG_IN_SIE; 1150 barrier(); 1151 if (!kvm_s390_vcpu_sie_inhibited(vcpu)) 1152 rc = sie64a(scb_s, vcpu->run->s.regs.gprs); 1153 barrier(); 1154 vcpu->arch.sie_block->prog0c &= ~PROG_IN_SIE; 1155 1156 local_irq_disable(); 1157 guest_exit_irqoff(); 1158 local_irq_enable(); 1159 1160 /* restore guest state for bp isolation override */ 1161 if (!guest_bp_isolation) 1162 clear_thread_flag(TIF_ISOLATE_BP_GUEST); 1163 1164 kvm_vcpu_srcu_read_lock(vcpu); 1165 1166 if (rc == -EINTR) { 1167 VCPU_EVENT(vcpu, 3, "%s", "machine check"); 1168 kvm_s390_reinject_machine_check(vcpu, &vsie_page->mcck_info); 1169 return 0; 1170 } 1171 1172 if (rc > 0) 1173 rc = 0; /* we could still have an icpt */ 1174 else if (rc == -EFAULT) 1175 return handle_fault(vcpu, vsie_page); 1176 1177 switch (scb_s->icptcode) { 1178 case ICPT_INST: 1179 if (scb_s->ipa == 0xb2b0) 1180 rc = handle_stfle(vcpu, vsie_page); 1181 break; 1182 case ICPT_STOP: 1183 /* stop not requested by g2 - must have been a kick */ 1184 if (!(atomic_read(&scb_o->cpuflags) & CPUSTAT_STOP_INT)) 1185 clear_vsie_icpt(vsie_page); 1186 break; 1187 case ICPT_VALIDITY: 1188 if ((scb_s->ipa & 0xf000) != 0xf000) 1189 scb_s->ipa += 0x1000; 1190 break; 1191 case ICPT_PARTEXEC: 1192 if (scb_s->ipa == 0xb254) 1193 rc = vsie_handle_mvpg(vcpu, vsie_page); 1194 break; 1195 } 1196 return rc; 1197 } 1198 1199 static void release_gmap_shadow(struct vsie_page *vsie_page) 1200 { 1201 if (vsie_page->gmap) 1202 gmap_put(vsie_page->gmap); 1203 WRITE_ONCE(vsie_page->gmap, NULL); 1204 prefix_unmapped(vsie_page); 1205 } 1206 1207 static int acquire_gmap_shadow(struct kvm_vcpu *vcpu, 1208 struct vsie_page *vsie_page) 1209 { 1210 unsigned long asce; 1211 union ctlreg0 cr0; 1212 struct gmap *gmap; 1213 int edat; 1214 1215 asce = vcpu->arch.sie_block->gcr[1]; 1216 cr0.val = vcpu->arch.sie_block->gcr[0]; 1217 edat = cr0.edat && test_kvm_facility(vcpu->kvm, 8); 1218 edat += edat && test_kvm_facility(vcpu->kvm, 78); 1219 1220 /* 1221 * ASCE or EDAT could have changed since last icpt, or the gmap 1222 * we're holding has been unshadowed. If the gmap is still valid, 1223 * we can safely reuse it. 1224 */ 1225 if (vsie_page->gmap && gmap_shadow_valid(vsie_page->gmap, asce, edat)) { 1226 vcpu->kvm->stat.gmap_shadow_reuse++; 1227 return 0; 1228 } 1229 1230 /* release the old shadow - if any, and mark the prefix as unmapped */ 1231 release_gmap_shadow(vsie_page); 1232 gmap = gmap_shadow(vcpu->arch.gmap, asce, edat); 1233 if (IS_ERR(gmap)) 1234 return PTR_ERR(gmap); 1235 vcpu->kvm->stat.gmap_shadow_create++; 1236 WRITE_ONCE(vsie_page->gmap, gmap); 1237 return 0; 1238 } 1239 1240 /* 1241 * Register the shadow scb at the VCPU, e.g. for kicking out of vsie. 1242 */ 1243 static void register_shadow_scb(struct kvm_vcpu *vcpu, 1244 struct vsie_page *vsie_page) 1245 { 1246 struct kvm_s390_sie_block *scb_s = &vsie_page->scb_s; 1247 1248 WRITE_ONCE(vcpu->arch.vsie_block, &vsie_page->scb_s); 1249 /* 1250 * External calls have to lead to a kick of the vcpu and 1251 * therefore the vsie -> Simulate Wait state. 1252 */ 1253 kvm_s390_set_cpuflags(vcpu, CPUSTAT_WAIT); 1254 /* 1255 * We have to adjust the g3 epoch by the g2 epoch. The epoch will 1256 * automatically be adjusted on tod clock changes via kvm_sync_clock. 1257 */ 1258 preempt_disable(); 1259 scb_s->epoch += vcpu->kvm->arch.epoch; 1260 1261 if (scb_s->ecd & ECD_MEF) { 1262 scb_s->epdx += vcpu->kvm->arch.epdx; 1263 if (scb_s->epoch < vcpu->kvm->arch.epoch) 1264 scb_s->epdx += 1; 1265 } 1266 1267 preempt_enable(); 1268 } 1269 1270 /* 1271 * Unregister a shadow scb from a VCPU. 1272 */ 1273 static void unregister_shadow_scb(struct kvm_vcpu *vcpu) 1274 { 1275 kvm_s390_clear_cpuflags(vcpu, CPUSTAT_WAIT); 1276 WRITE_ONCE(vcpu->arch.vsie_block, NULL); 1277 } 1278 1279 /* 1280 * Run the vsie on a shadowed scb, managing the gmap shadow, handling 1281 * prefix pages and faults. 1282 * 1283 * Returns: - 0 if no errors occurred 1284 * - > 0 if control has to be given to guest 2 1285 * - -ENOMEM if out of memory 1286 */ 1287 static int vsie_run(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page) 1288 { 1289 struct kvm_s390_sie_block *scb_s = &vsie_page->scb_s; 1290 int rc = 0; 1291 1292 while (1) { 1293 rc = acquire_gmap_shadow(vcpu, vsie_page); 1294 if (!rc) 1295 rc = map_prefix(vcpu, vsie_page); 1296 if (!rc) { 1297 gmap_enable(vsie_page->gmap); 1298 update_intervention_requests(vsie_page); 1299 rc = do_vsie_run(vcpu, vsie_page); 1300 gmap_enable(vcpu->arch.gmap); 1301 } 1302 atomic_andnot(PROG_BLOCK_SIE, &scb_s->prog20); 1303 1304 if (rc == -EAGAIN) 1305 rc = 0; 1306 if (rc || scb_s->icptcode || signal_pending(current) || 1307 kvm_s390_vcpu_has_irq(vcpu, 0) || 1308 kvm_s390_vcpu_sie_inhibited(vcpu)) 1309 break; 1310 cond_resched(); 1311 } 1312 1313 if (rc == -EFAULT) { 1314 /* 1315 * Addressing exceptions are always presentes as intercepts. 1316 * As addressing exceptions are suppressing and our guest 3 PSW 1317 * points at the responsible instruction, we have to 1318 * forward the PSW and set the ilc. If we can't read guest 3 1319 * instruction, we can use an arbitrary ilc. Let's always use 1320 * ilen = 4 for now, so we can avoid reading in guest 3 virtual 1321 * memory. (we could also fake the shadow so the hardware 1322 * handles it). 1323 */ 1324 scb_s->icptcode = ICPT_PROGI; 1325 scb_s->iprcc = PGM_ADDRESSING; 1326 scb_s->pgmilc = 4; 1327 scb_s->gpsw.addr = __rewind_psw(scb_s->gpsw, 4); 1328 rc = 1; 1329 } 1330 return rc; 1331 } 1332 1333 /* 1334 * Get or create a vsie page for a scb address. 1335 * 1336 * Returns: - address of a vsie page (cached or new one) 1337 * - NULL if the same scb address is already used by another VCPU 1338 * - ERR_PTR(-ENOMEM) if out of memory 1339 */ 1340 static struct vsie_page *get_vsie_page(struct kvm *kvm, unsigned long addr) 1341 { 1342 struct vsie_page *vsie_page; 1343 struct page *page; 1344 int nr_vcpus; 1345 1346 rcu_read_lock(); 1347 page = radix_tree_lookup(&kvm->arch.vsie.addr_to_page, addr >> 9); 1348 rcu_read_unlock(); 1349 if (page) { 1350 if (page_ref_inc_return(page) == 2) 1351 return page_to_virt(page); 1352 page_ref_dec(page); 1353 } 1354 1355 /* 1356 * We want at least #online_vcpus shadows, so every VCPU can execute 1357 * the VSIE in parallel. 1358 */ 1359 nr_vcpus = atomic_read(&kvm->online_vcpus); 1360 1361 mutex_lock(&kvm->arch.vsie.mutex); 1362 if (kvm->arch.vsie.page_count < nr_vcpus) { 1363 page = alloc_page(GFP_KERNEL_ACCOUNT | __GFP_ZERO | GFP_DMA); 1364 if (!page) { 1365 mutex_unlock(&kvm->arch.vsie.mutex); 1366 return ERR_PTR(-ENOMEM); 1367 } 1368 page_ref_inc(page); 1369 kvm->arch.vsie.pages[kvm->arch.vsie.page_count] = page; 1370 kvm->arch.vsie.page_count++; 1371 } else { 1372 /* reuse an existing entry that belongs to nobody */ 1373 while (true) { 1374 page = kvm->arch.vsie.pages[kvm->arch.vsie.next]; 1375 if (page_ref_inc_return(page) == 2) 1376 break; 1377 page_ref_dec(page); 1378 kvm->arch.vsie.next++; 1379 kvm->arch.vsie.next %= nr_vcpus; 1380 } 1381 radix_tree_delete(&kvm->arch.vsie.addr_to_page, page->index >> 9); 1382 } 1383 page->index = addr; 1384 /* double use of the same address */ 1385 if (radix_tree_insert(&kvm->arch.vsie.addr_to_page, addr >> 9, page)) { 1386 page_ref_dec(page); 1387 mutex_unlock(&kvm->arch.vsie.mutex); 1388 return NULL; 1389 } 1390 mutex_unlock(&kvm->arch.vsie.mutex); 1391 1392 vsie_page = page_to_virt(page); 1393 memset(&vsie_page->scb_s, 0, sizeof(struct kvm_s390_sie_block)); 1394 release_gmap_shadow(vsie_page); 1395 vsie_page->fault_addr = 0; 1396 vsie_page->scb_s.ihcpu = 0xffffU; 1397 return vsie_page; 1398 } 1399 1400 /* put a vsie page acquired via get_vsie_page */ 1401 static void put_vsie_page(struct kvm *kvm, struct vsie_page *vsie_page) 1402 { 1403 struct page *page = pfn_to_page(__pa(vsie_page) >> PAGE_SHIFT); 1404 1405 page_ref_dec(page); 1406 } 1407 1408 int kvm_s390_handle_vsie(struct kvm_vcpu *vcpu) 1409 { 1410 struct vsie_page *vsie_page; 1411 unsigned long scb_addr; 1412 int rc; 1413 1414 vcpu->stat.instruction_sie++; 1415 if (!test_kvm_cpu_feat(vcpu->kvm, KVM_S390_VM_CPU_FEAT_SIEF2)) 1416 return -EOPNOTSUPP; 1417 if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE) 1418 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP); 1419 1420 BUILD_BUG_ON(sizeof(struct vsie_page) != PAGE_SIZE); 1421 scb_addr = kvm_s390_get_base_disp_s(vcpu, NULL); 1422 1423 /* 512 byte alignment */ 1424 if (unlikely(scb_addr & 0x1ffUL)) 1425 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION); 1426 1427 if (signal_pending(current) || kvm_s390_vcpu_has_irq(vcpu, 0) || 1428 kvm_s390_vcpu_sie_inhibited(vcpu)) 1429 return 0; 1430 1431 vsie_page = get_vsie_page(vcpu->kvm, scb_addr); 1432 if (IS_ERR(vsie_page)) 1433 return PTR_ERR(vsie_page); 1434 else if (!vsie_page) 1435 /* double use of sie control block - simply do nothing */ 1436 return 0; 1437 1438 rc = pin_scb(vcpu, vsie_page, scb_addr); 1439 if (rc) 1440 goto out_put; 1441 rc = shadow_scb(vcpu, vsie_page); 1442 if (rc) 1443 goto out_unpin_scb; 1444 rc = pin_blocks(vcpu, vsie_page); 1445 if (rc) 1446 goto out_unshadow; 1447 register_shadow_scb(vcpu, vsie_page); 1448 rc = vsie_run(vcpu, vsie_page); 1449 unregister_shadow_scb(vcpu); 1450 unpin_blocks(vcpu, vsie_page); 1451 out_unshadow: 1452 unshadow_scb(vcpu, vsie_page); 1453 out_unpin_scb: 1454 unpin_scb(vcpu, vsie_page, scb_addr); 1455 out_put: 1456 put_vsie_page(vcpu->kvm, vsie_page); 1457 1458 return rc < 0 ? rc : 0; 1459 } 1460 1461 /* Init the vsie data structures. To be called when a vm is initialized. */ 1462 void kvm_s390_vsie_init(struct kvm *kvm) 1463 { 1464 mutex_init(&kvm->arch.vsie.mutex); 1465 INIT_RADIX_TREE(&kvm->arch.vsie.addr_to_page, GFP_KERNEL_ACCOUNT); 1466 } 1467 1468 /* Destroy the vsie data structures. To be called when a vm is destroyed. */ 1469 void kvm_s390_vsie_destroy(struct kvm *kvm) 1470 { 1471 struct vsie_page *vsie_page; 1472 struct page *page; 1473 int i; 1474 1475 mutex_lock(&kvm->arch.vsie.mutex); 1476 for (i = 0; i < kvm->arch.vsie.page_count; i++) { 1477 page = kvm->arch.vsie.pages[i]; 1478 kvm->arch.vsie.pages[i] = NULL; 1479 vsie_page = page_to_virt(page); 1480 release_gmap_shadow(vsie_page); 1481 /* free the radix tree entry */ 1482 radix_tree_delete(&kvm->arch.vsie.addr_to_page, page->index >> 9); 1483 __free_page(page); 1484 } 1485 kvm->arch.vsie.page_count = 0; 1486 mutex_unlock(&kvm->arch.vsie.mutex); 1487 } 1488 1489 void kvm_s390_vsie_kick(struct kvm_vcpu *vcpu) 1490 { 1491 struct kvm_s390_sie_block *scb = READ_ONCE(vcpu->arch.vsie_block); 1492 1493 /* 1494 * Even if the VCPU lets go of the shadow sie block reference, it is 1495 * still valid in the cache. So we can safely kick it. 1496 */ 1497 if (scb) { 1498 atomic_or(PROG_BLOCK_SIE, &scb->prog20); 1499 if (scb->prog0c & PROG_IN_SIE) 1500 atomic_or(CPUSTAT_STOP_INT, &scb->cpuflags); 1501 } 1502 } 1503