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