vsie.c

   1// SPDX-License-Identifier: GPL-2.0
   2/*
   3 * kvm nested virtualization support for s390x
   4 *
   5 * Copyright IBM Corp. 2016, 2018
   6 *
   7 *    Author(s): David Hildenbrand <dahi@linux.vnet.ibm.com>
   8 */
   9#include <linux/vmalloc.h>
  10#include <linux/kvm_host.h>
  11#include <linux/bug.h>
  12#include <linux/list.h>
  13#include <linux/bitmap.h>
  14#include <linux/sched/signal.h>
  15#include <linux/io.h>
  16#include <linux/mman.h>
  17
  18#include <asm/gmap.h>
  19#include <asm/mmu_context.h>
  20#include <asm/sclp.h>
  21#include <asm/nmi.h>
  22#include <asm/dis.h>
  23#include <asm/facility.h>
  24#include "kvm-s390.h"
  25#include "gaccess.h"
  26
  27enum vsie_page_flags {
  28	VSIE_PAGE_IN_USE = 0,
  29};
  30
  31struct vsie_page {
  32	struct kvm_s390_sie_block scb_s;	/* 0x0000 */
  33	/*
  34	 * the backup info for machine check. ensure it's at
  35	 * the same offset as that in struct sie_page!
  36	 */
  37	struct mcck_volatile_info mcck_info;    /* 0x0200 */
  38	/*
  39	 * The pinned original scb. Be aware that other VCPUs can modify
  40	 * it while we read from it. Values that are used for conditions or
  41	 * are reused conditionally, should be accessed via READ_ONCE.
  42	 */
  43	struct kvm_s390_sie_block *scb_o;	/* 0x0218 */
  44	/* the shadow gmap in use by the vsie_page */
  45	struct gmap *gmap;			/* 0x0220 */
  46	/* address of the last reported fault to guest2 */
  47	unsigned long fault_addr;		/* 0x0228 */
  48	/* calculated guest addresses of satellite control blocks */
  49	gpa_t sca_gpa;				/* 0x0230 */
  50	gpa_t itdba_gpa;			/* 0x0238 */
  51	gpa_t gvrd_gpa;				/* 0x0240 */
  52	gpa_t riccbd_gpa;			/* 0x0248 */
  53	gpa_t sdnx_gpa;				/* 0x0250 */
  54	/*
  55	 * guest address of the original SCB. Remains set for free vsie
  56	 * pages, so we can properly look them up in our addr_to_page
  57	 * radix tree.
  58	 */
  59	gpa_t scb_gpa;				/* 0x0258 */
  60	/*
  61	 * Flags: must be set/cleared atomically after the vsie page can be
  62	 * looked up by other CPUs.
  63	 */
  64	unsigned long flags;			/* 0x0260 */
  65	__u8 reserved[0x0700 - 0x0268];		/* 0x0268 */
  66	struct kvm_s390_crypto_cb crycb;	/* 0x0700 */
  67	__u8 fac[S390_ARCH_FAC_LIST_SIZE_BYTE];	/* 0x0800 */
  68};
  69
  70/**
  71 * gmap_shadow_valid() - check if a shadow guest address space matches the
  72 *                       given properties and is still valid
  73 * @sg: pointer to the shadow guest address space structure
  74 * @asce: ASCE for which the shadow table is requested
  75 * @edat_level: edat level to be used for the shadow translation
  76 *
  77 * Returns 1 if the gmap shadow is still valid and matches the given
  78 * properties, the caller can continue using it. Returns 0 otherwise; the
  79 * caller has to request a new shadow gmap in this case.
  80 */
  81int gmap_shadow_valid(struct gmap *sg, unsigned long asce, int edat_level)
  82{
  83	if (sg->removed)
  84		return 0;
  85	return sg->orig_asce == asce && sg->edat_level == edat_level;
  86}
  87
  88/* trigger a validity icpt for the given scb */
  89static int set_validity_icpt(struct kvm_s390_sie_block *scb,
  90			     __u16 reason_code)
  91{
  92	scb->ipa = 0x1000;
  93	scb->ipb = ((__u32) reason_code) << 16;
  94	scb->icptcode = ICPT_VALIDITY;
  95	return 1;
  96}
  97
  98/* mark the prefix as unmapped, this will block the VSIE */
  99static void prefix_unmapped(struct vsie_page *vsie_page)
 100{
 101	atomic_or(PROG_REQUEST, &vsie_page->scb_s.prog20);
 102}
 103
 104/* mark the prefix as unmapped and wait until the VSIE has been left */
 105static void prefix_unmapped_sync(struct vsie_page *vsie_page)
 106{
 107	prefix_unmapped(vsie_page);
 108	if (vsie_page->scb_s.prog0c & PROG_IN_SIE)
 109		atomic_or(CPUSTAT_STOP_INT, &vsie_page->scb_s.cpuflags);
 110	while (vsie_page->scb_s.prog0c & PROG_IN_SIE)
 111		cpu_relax();
 112}
 113
 114/* mark the prefix as mapped, this will allow the VSIE to run */
 115static void prefix_mapped(struct vsie_page *vsie_page)
 116{
 117	atomic_andnot(PROG_REQUEST, &vsie_page->scb_s.prog20);
 118}
 119
 120/* test if the prefix is mapped into the gmap shadow */
 121static int prefix_is_mapped(struct vsie_page *vsie_page)
 122{
 123	return !(atomic_read(&vsie_page->scb_s.prog20) & PROG_REQUEST);
 124}
 125
 126/* copy the updated intervention request bits into the shadow scb */
 127static void update_intervention_requests(struct vsie_page *vsie_page)
 128{
 129	const int bits = CPUSTAT_STOP_INT | CPUSTAT_IO_INT | CPUSTAT_EXT_INT;
 130	int cpuflags;
 131
 132	cpuflags = atomic_read(&vsie_page->scb_o->cpuflags);
 133	atomic_andnot(bits, &vsie_page->scb_s.cpuflags);
 134	atomic_or(cpuflags & bits, &vsie_page->scb_s.cpuflags);
 135}
 136
 137/* shadow (filter and validate) the cpuflags  */
 138static int prepare_cpuflags(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page)
 139{
 140	struct kvm_s390_sie_block *scb_s = &vsie_page->scb_s;
 141	struct kvm_s390_sie_block *scb_o = vsie_page->scb_o;
 142	int newflags, cpuflags = atomic_read(&scb_o->cpuflags);
 143
 144	/* we don't allow ESA/390 guests */
 145	if (!(cpuflags & CPUSTAT_ZARCH))
 146		return set_validity_icpt(scb_s, 0x0001U);
 147
 148	if (cpuflags & (CPUSTAT_RRF | CPUSTAT_MCDS))
 149		return set_validity_icpt(scb_s, 0x0001U);
 150	else if (cpuflags & (CPUSTAT_SLSV | CPUSTAT_SLSR))
 151		return set_validity_icpt(scb_s, 0x0007U);
 152
 153	/* intervention requests will be set later */
 154	newflags = CPUSTAT_ZARCH;
 155	if (cpuflags & CPUSTAT_GED && test_kvm_facility(vcpu->kvm, 8))
 156		newflags |= CPUSTAT_GED;
 157	if (cpuflags & CPUSTAT_GED2 && test_kvm_facility(vcpu->kvm, 78)) {
 158		if (cpuflags & CPUSTAT_GED)
 159			return set_validity_icpt(scb_s, 0x0001U);
 160		newflags |= CPUSTAT_GED2;
 161	}
 162	if (test_kvm_cpu_feat(vcpu->kvm, KVM_S390_VM_CPU_FEAT_GPERE))
 163		newflags |= cpuflags & CPUSTAT_P;
 164	if (test_kvm_cpu_feat(vcpu->kvm, KVM_S390_VM_CPU_FEAT_GSLS))
 165		newflags |= cpuflags & CPUSTAT_SM;
 166	if (test_kvm_cpu_feat(vcpu->kvm, KVM_S390_VM_CPU_FEAT_IBS))
 167		newflags |= cpuflags & CPUSTAT_IBS;
 168	if (test_kvm_cpu_feat(vcpu->kvm, KVM_S390_VM_CPU_FEAT_KSS))
 169		newflags |= cpuflags & CPUSTAT_KSS;
 170
 171	atomic_set(&scb_s->cpuflags, newflags);
 172	return 0;
 173}
 174/* Copy to APCB FORMAT1 from APCB FORMAT0 */
 175static int setup_apcb10(struct kvm_vcpu *vcpu, struct kvm_s390_apcb1 *apcb_s,
 176			unsigned long crycb_gpa, struct kvm_s390_apcb1 *apcb_h)
 177{
 178	struct kvm_s390_apcb0 tmp;
 179	unsigned long apcb_gpa;
 180
 181	apcb_gpa = crycb_gpa + offsetof(struct kvm_s390_crypto_cb, apcb0);
 182
 183	if (read_guest_real(vcpu, apcb_gpa, &tmp,
 184			    sizeof(struct kvm_s390_apcb0)))
 185		return -EFAULT;
 186
 187	apcb_s->apm[0] = apcb_h->apm[0] & tmp.apm[0];
 188	apcb_s->aqm[0] = apcb_h->aqm[0] & tmp.aqm[0] & 0xffff000000000000UL;
 189	apcb_s->adm[0] = apcb_h->adm[0] & tmp.adm[0] & 0xffff000000000000UL;
 190
 191	return 0;
 192
 193}
 194
 195/**
 196 * setup_apcb00 - Copy to APCB FORMAT0 from APCB FORMAT0
 197 * @vcpu: pointer to the virtual CPU
 198 * @apcb_s: pointer to start of apcb in the shadow crycb
 199 * @crycb_gpa: guest physical address to start of original guest crycb
 200 * @apcb_h: pointer to start of apcb in the guest1
 201 *
 202 * Returns 0 and -EFAULT on error reading guest apcb
 203 */
 204static int setup_apcb00(struct kvm_vcpu *vcpu, unsigned long *apcb_s,
 205			unsigned long crycb_gpa, unsigned long *apcb_h)
 206{
 207	unsigned long apcb_gpa;
 208
 209	apcb_gpa = crycb_gpa + offsetof(struct kvm_s390_crypto_cb, apcb0);
 210
 211	if (read_guest_real(vcpu, apcb_gpa, apcb_s,
 212			    sizeof(struct kvm_s390_apcb0)))
 213		return -EFAULT;
 214
 215	bitmap_and(apcb_s, apcb_s, apcb_h,
 216		   BITS_PER_BYTE * sizeof(struct kvm_s390_apcb0));
 217
 218	return 0;
 219}
 220
 221/**
 222 * setup_apcb11 - Copy the FORMAT1 APCB from the guest to the shadow CRYCB
 223 * @vcpu: pointer to the virtual CPU
 224 * @apcb_s: pointer to start of apcb in the shadow crycb
 225 * @crycb_gpa: guest physical address to start of original guest crycb
 226 * @apcb_h: pointer to start of apcb in the host
 227 *
 228 * Returns 0 and -EFAULT on error reading guest apcb
 229 */
 230static int setup_apcb11(struct kvm_vcpu *vcpu, unsigned long *apcb_s,
 231			unsigned long crycb_gpa,
 232			unsigned long *apcb_h)
 233{
 234	unsigned long apcb_gpa;
 235
 236	apcb_gpa = crycb_gpa + offsetof(struct kvm_s390_crypto_cb, apcb1);
 237
 238	if (read_guest_real(vcpu, apcb_gpa, apcb_s,
 239			    sizeof(struct kvm_s390_apcb1)))
 240		return -EFAULT;
 241
 242	bitmap_and(apcb_s, apcb_s, apcb_h,
 243		   BITS_PER_BYTE * sizeof(struct kvm_s390_apcb1));
 244
 245	return 0;
 246}
 247
 248/**
 249 * setup_apcb - Create a shadow copy of the apcb.
 250 * @vcpu: pointer to the virtual CPU
 251 * @crycb_s: pointer to shadow crycb
 252 * @crycb_gpa: guest physical address of original guest crycb
 253 * @crycb_h: pointer to the host crycb
 254 * @fmt_o: format of the original guest crycb.
 255 * @fmt_h: format of the host crycb.
 256 *
 257 * Checks the compatibility between the guest and host crycb and calls the
 258 * appropriate copy function.
 259 *
 260 * Return 0 or an error number if the guest and host crycb are incompatible.
 261 */
 262static int setup_apcb(struct kvm_vcpu *vcpu, struct kvm_s390_crypto_cb *crycb_s,
 263	       const u32 crycb_gpa,
 264	       struct kvm_s390_crypto_cb *crycb_h,
 265	       int fmt_o, int fmt_h)
 266{
 267	switch (fmt_o) {
 268	case CRYCB_FORMAT2:
 269		if ((crycb_gpa & PAGE_MASK) != ((crycb_gpa + 256) & PAGE_MASK))
 270			return -EACCES;
 271		if (fmt_h != CRYCB_FORMAT2)
 272			return -EINVAL;
 273		return setup_apcb11(vcpu, (unsigned long *)&crycb_s->apcb1,
 274				    crycb_gpa,
 275				    (unsigned long *)&crycb_h->apcb1);
 276	case CRYCB_FORMAT1:
 277		switch (fmt_h) {
 278		case CRYCB_FORMAT2:
 279			return setup_apcb10(vcpu, &crycb_s->apcb1,
 280					    crycb_gpa,
 281					    &crycb_h->apcb1);
 282		case CRYCB_FORMAT1:
 283			return setup_apcb00(vcpu,
 284					    (unsigned long *) &crycb_s->apcb0,
 285					    crycb_gpa,
 286					    (unsigned long *) &crycb_h->apcb0);
 287		}
 288		break;
 289	case CRYCB_FORMAT0:
 290		if ((crycb_gpa & PAGE_MASK) != ((crycb_gpa + 32) & PAGE_MASK))
 291			return -EACCES;
 292
 293		switch (fmt_h) {
 294		case CRYCB_FORMAT2:
 295			return setup_apcb10(vcpu, &crycb_s->apcb1,
 296					    crycb_gpa,
 297					    &crycb_h->apcb1);
 298		case CRYCB_FORMAT1:
 299		case CRYCB_FORMAT0:
 300			return setup_apcb00(vcpu,
 301					    (unsigned long *) &crycb_s->apcb0,
 302					    crycb_gpa,
 303					    (unsigned long *) &crycb_h->apcb0);
 304		}
 305	}
 306	return -EINVAL;
 307}
 308
 309/**
 310 * shadow_crycb - Create a shadow copy of the crycb block
 311 * @vcpu: a pointer to the virtual CPU
 312 * @vsie_page: a pointer to internal date used for the vSIE
 313 *
 314 * Create a shadow copy of the crycb block and setup key wrapping, if
 315 * requested for guest 3 and enabled for guest 2.
 316 *
 317 * We accept format-1 or format-2, but we convert format-1 into format-2
 318 * in the shadow CRYCB.
 319 * Using format-2 enables the firmware to choose the right format when
 320 * scheduling the SIE.
 321 * There is nothing to do for format-0.
 322 *
 323 * This function centralize the issuing of set_validity_icpt() for all
 324 * the subfunctions working on the crycb.
 325 *
 326 * Returns: - 0 if shadowed or nothing to do
 327 *          - > 0 if control has to be given to guest 2
 328 */
 329static int shadow_crycb(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page)
 330{
 331	struct kvm_s390_sie_block *scb_s = &vsie_page->scb_s;
 332	struct kvm_s390_sie_block *scb_o = vsie_page->scb_o;
 333	const uint32_t crycbd_o = READ_ONCE(scb_o->crycbd);
 334	const u32 crycb_addr = crycbd_o & 0x7ffffff8U;
 335	unsigned long *b1, *b2;
 336	u8 ecb3_flags;
 337	u32 ecd_flags;
 338	int apie_h;
 339	int apie_s;
 340	int key_msk = test_kvm_facility(vcpu->kvm, 76);
 341	int fmt_o = crycbd_o & CRYCB_FORMAT_MASK;
 342	int fmt_h = vcpu->arch.sie_block->crycbd & CRYCB_FORMAT_MASK;
 343	int ret = 0;
 344
 345	scb_s->crycbd = 0;
 346
 347	apie_h = vcpu->arch.sie_block->eca & ECA_APIE;
 348	apie_s = apie_h & scb_o->eca;
 349	if (!apie_s && (!key_msk || (fmt_o == CRYCB_FORMAT0)))
 350		return 0;
 351
 352	if (!crycb_addr)
 353		return set_validity_icpt(scb_s, 0x0039U);
 354
 355	if (fmt_o == CRYCB_FORMAT1)
 356		if ((crycb_addr & PAGE_MASK) !=
 357		    ((crycb_addr + 128) & PAGE_MASK))
 358			return set_validity_icpt(scb_s, 0x003CU);
 359
 360	if (apie_s) {
 361		ret = setup_apcb(vcpu, &vsie_page->crycb, crycb_addr,
 362				 vcpu->kvm->arch.crypto.crycb,
 363				 fmt_o, fmt_h);
 364		if (ret)
 365			goto end;
 366		scb_s->eca |= scb_o->eca & ECA_APIE;
 367	}
 368
 369	/* we may only allow it if enabled for guest 2 */
 370	ecb3_flags = scb_o->ecb3 & vcpu->arch.sie_block->ecb3 &
 371		     (ECB3_AES | ECB3_DEA);
 372	ecd_flags = scb_o->ecd & vcpu->arch.sie_block->ecd &
 373		     (ECD_ECC | ECD_HMAC);
 374	if (!ecb3_flags && !ecd_flags)
 375		goto end;
 376
 377	/* copy only the wrapping keys */
 378	if (read_guest_real(vcpu, crycb_addr + 72,
 379			    vsie_page->crycb.dea_wrapping_key_mask, 56))
 380		return set_validity_icpt(scb_s, 0x0035U);
 381
 382	scb_s->ecb3 |= ecb3_flags;
 383	scb_s->ecd |= ecd_flags;
 384
 385	/* xor both blocks in one run */
 386	b1 = (unsigned long *) vsie_page->crycb.dea_wrapping_key_mask;
 387	b2 = (unsigned long *)
 388			    vcpu->kvm->arch.crypto.crycb->dea_wrapping_key_mask;
 389	/* as 56%8 == 0, bitmap_xor won't overwrite any data */
 390	bitmap_xor(b1, b1, b2, BITS_PER_BYTE * 56);
 391end:
 392	switch (ret) {
 393	case -EINVAL:
 394		return set_validity_icpt(scb_s, 0x0022U);
 395	case -EFAULT:
 396		return set_validity_icpt(scb_s, 0x0035U);
 397	case -EACCES:
 398		return set_validity_icpt(scb_s, 0x003CU);
 399	}
 400	scb_s->crycbd = (u32)virt_to_phys(&vsie_page->crycb) | CRYCB_FORMAT2;
 401	return 0;
 402}
 403
 404/* shadow (round up/down) the ibc to avoid validity icpt */
 405static void prepare_ibc(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page)
 406{
 407	struct kvm_s390_sie_block *scb_s = &vsie_page->scb_s;
 408	struct kvm_s390_sie_block *scb_o = vsie_page->scb_o;
 409	/* READ_ONCE does not work on bitfields - use a temporary variable */
 410	const uint32_t __new_ibc = scb_o->ibc;
 411	const uint32_t new_ibc = READ_ONCE(__new_ibc) & 0x0fffU;
 412	__u64 min_ibc = (sclp.ibc >> 16) & 0x0fffU;
 413
 414	scb_s->ibc = 0;
 415	/* ibc installed in g2 and requested for g3 */
 416	if (vcpu->kvm->arch.model.ibc && new_ibc) {
 417		scb_s->ibc = new_ibc;
 418		/* takte care of the minimum ibc level of the machine */
 419		if (scb_s->ibc < min_ibc)
 420			scb_s->ibc = min_ibc;
 421		/* take care of the maximum ibc level set for the guest */
 422		if (scb_s->ibc > vcpu->kvm->arch.model.ibc)
 423			scb_s->ibc = vcpu->kvm->arch.model.ibc;
 424	}
 425}
 426
 427/* unshadow the scb, copying parameters back to the real scb */
 428static void unshadow_scb(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page)
 429{
 430	struct kvm_s390_sie_block *scb_s = &vsie_page->scb_s;
 431	struct kvm_s390_sie_block *scb_o = vsie_page->scb_o;
 432
 433	/* interception */
 434	scb_o->icptcode = scb_s->icptcode;
 435	scb_o->icptstatus = scb_s->icptstatus;
 436	scb_o->ipa = scb_s->ipa;
 437	scb_o->ipb = scb_s->ipb;
 438	scb_o->gbea = scb_s->gbea;
 439
 440	/* timer */
 441	scb_o->cputm = scb_s->cputm;
 442	scb_o->ckc = scb_s->ckc;
 443	scb_o->todpr = scb_s->todpr;
 444
 445	/* guest state */
 446	scb_o->gpsw = scb_s->gpsw;
 447	scb_o->gg14 = scb_s->gg14;
 448	scb_o->gg15 = scb_s->gg15;
 449	memcpy(scb_o->gcr, scb_s->gcr, 128);
 450	scb_o->pp = scb_s->pp;
 451
 452	/* branch prediction */
 453	if (test_kvm_facility(vcpu->kvm, 82)) {
 454		scb_o->fpf &= ~FPF_BPBC;
 455		scb_o->fpf |= scb_s->fpf & FPF_BPBC;
 456	}
 457
 458	/* interrupt intercept */
 459	switch (scb_s->icptcode) {
 460	case ICPT_PROGI:
 461	case ICPT_INSTPROGI:
 462	case ICPT_EXTINT:
 463		memcpy((void *)((u64)scb_o + 0xc0),
 464		       (void *)((u64)scb_s + 0xc0), 0xf0 - 0xc0);
 465		break;
 466	}
 467
 468	if (scb_s->ihcpu != 0xffffU)
 469		scb_o->ihcpu = scb_s->ihcpu;
 470}
 471
 472/*
 473 * Setup the shadow scb by copying and checking the relevant parts of the g2
 474 * provided scb.
 475 *
 476 * Returns: - 0 if the scb has been shadowed
 477 *          - > 0 if control has to be given to guest 2
 478 */
 479static int shadow_scb(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page)
 480{
 481	struct kvm_s390_sie_block *scb_o = vsie_page->scb_o;
 482	struct kvm_s390_sie_block *scb_s = &vsie_page->scb_s;
 483	/* READ_ONCE does not work on bitfields - use a temporary variable */
 484	const uint32_t __new_prefix = scb_o->prefix;
 485	const uint32_t new_prefix = READ_ONCE(__new_prefix);
 486	const bool wants_tx = READ_ONCE(scb_o->ecb) & ECB_TE;
 487	bool had_tx = scb_s->ecb & ECB_TE;
 488	unsigned long new_mso = 0;
 489	int rc;
 490
 491	/* make sure we don't have any leftovers when reusing the scb */
 492	scb_s->icptcode = 0;
 493	scb_s->eca = 0;
 494	scb_s->ecb = 0;
 495	scb_s->ecb2 = 0;
 496	scb_s->ecb3 = 0;
 497	scb_s->ecd = 0;
 498	scb_s->fac = 0;
 499	scb_s->fpf = 0;
 500
 501	rc = prepare_cpuflags(vcpu, vsie_page);
 502	if (rc)
 503		goto out;
 504
 505	/* timer */
 506	scb_s->cputm = scb_o->cputm;
 507	scb_s->ckc = scb_o->ckc;
 508	scb_s->todpr = scb_o->todpr;
 509	scb_s->epoch = scb_o->epoch;
 510
 511	/* guest state */
 512	scb_s->gpsw = scb_o->gpsw;
 513	scb_s->gg14 = scb_o->gg14;
 514	scb_s->gg15 = scb_o->gg15;
 515	memcpy(scb_s->gcr, scb_o->gcr, 128);
 516	scb_s->pp = scb_o->pp;
 517
 518	/* interception / execution handling */
 519	scb_s->gbea = scb_o->gbea;
 520	scb_s->lctl = scb_o->lctl;
 521	scb_s->svcc = scb_o->svcc;
 522	scb_s->ictl = scb_o->ictl;
 523	/*
 524	 * SKEY handling functions can't deal with false setting of PTE invalid
 525	 * bits. Therefore we cannot provide interpretation and would later
 526	 * have to provide own emulation handlers.
 527	 */
 528	if (!(atomic_read(&scb_s->cpuflags) & CPUSTAT_KSS))
 529		scb_s->ictl |= ICTL_ISKE | ICTL_SSKE | ICTL_RRBE;
 530
 531	scb_s->icpua = scb_o->icpua;
 532
 533	if (!(atomic_read(&scb_s->cpuflags) & CPUSTAT_SM))
 534		new_mso = READ_ONCE(scb_o->mso) & 0xfffffffffff00000UL;
 535	/* if the hva of the prefix changes, we have to remap the prefix */
 536	if (scb_s->mso != new_mso || scb_s->prefix != new_prefix)
 537		prefix_unmapped(vsie_page);
 538	 /* SIE will do mso/msl validity and exception checks for us */
 539	scb_s->msl = scb_o->msl & 0xfffffffffff00000UL;
 540	scb_s->mso = new_mso;
 541	scb_s->prefix = new_prefix;
 542
 543	/* We have to definitely flush the tlb if this scb never ran */
 544	if (scb_s->ihcpu != 0xffffU)
 545		scb_s->ihcpu = scb_o->ihcpu;
 546
 547	/* MVPG and Protection Exception Interpretation are always available */
 548	scb_s->eca |= scb_o->eca & (ECA_MVPGI | ECA_PROTEXCI);
 549	/* Host-protection-interruption introduced with ESOP */
 550	if (test_kvm_cpu_feat(vcpu->kvm, KVM_S390_VM_CPU_FEAT_ESOP))
 551		scb_s->ecb |= scb_o->ecb & ECB_HOSTPROTINT;
 552	/*
 553	 * CPU Topology
 554	 * This facility only uses the utility field of the SCA and none of
 555	 * the cpu entries that are problematic with the other interpretation
 556	 * facilities so we can pass it through
 557	 */
 558	if (test_kvm_facility(vcpu->kvm, 11))
 559		scb_s->ecb |= scb_o->ecb & ECB_PTF;
 560	/* transactional execution */
 561	if (test_kvm_facility(vcpu->kvm, 73) && wants_tx) {
 562		/* remap the prefix is tx is toggled on */
 563		if (!had_tx)
 564			prefix_unmapped(vsie_page);
 565		scb_s->ecb |= ECB_TE;
 566	}
 567	/* specification exception interpretation */
 568	scb_s->ecb |= scb_o->ecb & ECB_SPECI;
 569	/* branch prediction */
 570	if (test_kvm_facility(vcpu->kvm, 82))
 571		scb_s->fpf |= scb_o->fpf & FPF_BPBC;
 572	/* SIMD */
 573	if (test_kvm_facility(vcpu->kvm, 129)) {
 574		scb_s->eca |= scb_o->eca & ECA_VX;
 575		scb_s->ecd |= scb_o->ecd & ECD_HOSTREGMGMT;
 576	}
 577	/* Run-time-Instrumentation */
 578	if (test_kvm_facility(vcpu->kvm, 64))
 579		scb_s->ecb3 |= scb_o->ecb3 & ECB3_RI;
 580	/* Instruction Execution Prevention */
 581	if (test_kvm_facility(vcpu->kvm, 130))
 582		scb_s->ecb2 |= scb_o->ecb2 & ECB2_IEP;
 583	/* Guarded Storage */
 584	if (test_kvm_facility(vcpu->kvm, 133)) {
 585		scb_s->ecb |= scb_o->ecb & ECB_GS;
 586		scb_s->ecd |= scb_o->ecd & ECD_HOSTREGMGMT;
 587	}
 588	if (test_kvm_cpu_feat(vcpu->kvm, KVM_S390_VM_CPU_FEAT_SIIF))
 589		scb_s->eca |= scb_o->eca & ECA_SII;
 590	if (test_kvm_cpu_feat(vcpu->kvm, KVM_S390_VM_CPU_FEAT_IB))
 591		scb_s->eca |= scb_o->eca & ECA_IB;
 592	if (test_kvm_cpu_feat(vcpu->kvm, KVM_S390_VM_CPU_FEAT_CEI))
 593		scb_s->eca |= scb_o->eca & ECA_CEI;
 594	/* Epoch Extension */
 595	if (test_kvm_facility(vcpu->kvm, 139)) {
 596		scb_s->ecd |= scb_o->ecd & ECD_MEF;
 597		scb_s->epdx = scb_o->epdx;
 598	}
 599
 600	/* etoken */
 601	if (test_kvm_facility(vcpu->kvm, 156))
 602		scb_s->ecd |= scb_o->ecd & ECD_ETOKENF;
 603
 604	scb_s->hpid = HPID_VSIE;
 605	scb_s->cpnc = scb_o->cpnc;
 606
 607	prepare_ibc(vcpu, vsie_page);
 608	rc = shadow_crycb(vcpu, vsie_page);
 609out:
 610	if (rc)
 611		unshadow_scb(vcpu, vsie_page);
 612	return rc;
 613}
 614
 615void kvm_s390_vsie_gmap_notifier(struct gmap *gmap, unsigned long start,
 616				 unsigned long end)
 617{
 618	struct kvm *kvm = gmap->private;
 619	struct vsie_page *cur;
 620	unsigned long prefix;
 621	int i;
 622
 623	if (!gmap_is_shadow(gmap))
 624		return;
 625	/*
 626	 * Only new shadow blocks are added to the list during runtime,
 627	 * therefore we can safely reference them all the time.
 628	 */
 629	for (i = 0; i < kvm->arch.vsie.page_count; i++) {
 630		cur = READ_ONCE(kvm->arch.vsie.pages[i]);
 631		if (!cur)
 632			continue;
 633		if (READ_ONCE(cur->gmap) != gmap)
 634			continue;
 635		prefix = cur->scb_s.prefix << GUEST_PREFIX_SHIFT;
 636		/* with mso/msl, the prefix lies at an offset */
 637		prefix += cur->scb_s.mso;
 638		if (prefix <= end && start <= prefix + 2 * PAGE_SIZE - 1)
 639			prefix_unmapped_sync(cur);
 640	}
 641}
 642
 643/*
 644 * Map the first prefix page and if tx is enabled also the second prefix page.
 645 *
 646 * The prefix will be protected, a gmap notifier will inform about unmaps.
 647 * The shadow scb must not be executed until the prefix is remapped, this is
 648 * guaranteed by properly handling PROG_REQUEST.
 649 *
 650 * Returns: - 0 on if successfully mapped or already mapped
 651 *          - > 0 if control has to be given to guest 2
 652 *          - -EAGAIN if the caller can retry immediately
 653 *          - -ENOMEM if out of memory
 654 */
 655static int map_prefix(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page)
 656{
 657	struct kvm_s390_sie_block *scb_s = &vsie_page->scb_s;
 658	u64 prefix = scb_s->prefix << GUEST_PREFIX_SHIFT;
 659	int rc;
 660
 661	if (prefix_is_mapped(vsie_page))
 662		return 0;
 663
 664	/* mark it as mapped so we can catch any concurrent unmappers */
 665	prefix_mapped(vsie_page);
 666
 667	/* with mso/msl, the prefix lies at offset *mso* */
 668	prefix += scb_s->mso;
 669
 670	rc = kvm_s390_shadow_fault(vcpu, vsie_page->gmap, prefix, NULL);
 671	if (!rc && (scb_s->ecb & ECB_TE))
 672		rc = kvm_s390_shadow_fault(vcpu, vsie_page->gmap,
 673					   prefix + PAGE_SIZE, NULL);
 674	/*
 675	 * We don't have to mprotect, we will be called for all unshadows.
 676	 * SIE will detect if protection applies and trigger a validity.
 677	 */
 678	if (rc)
 679		prefix_unmapped(vsie_page);
 680	if (rc > 0 || rc == -EFAULT)
 681		rc = set_validity_icpt(scb_s, 0x0037U);
 682	return rc;
 683}
 684
 685/*
 686 * Pin the guest page given by gpa and set hpa to the pinned host address.
 687 * Will always be pinned writable.
 688 *
 689 * Returns: - 0 on success
 690 *          - -EINVAL if the gpa is not valid guest storage
 691 */
 692static int pin_guest_page(struct kvm *kvm, gpa_t gpa, hpa_t *hpa)
 693{
 694	struct page *page;
 695
 696	page = gfn_to_page(kvm, gpa_to_gfn(gpa));
 697	if (!page)
 698		return -EINVAL;
 699	*hpa = (hpa_t)page_to_phys(page) + (gpa & ~PAGE_MASK);
 700	return 0;
 701}
 702
 703/* Unpins a page previously pinned via pin_guest_page, marking it as dirty. */
 704static void unpin_guest_page(struct kvm *kvm, gpa_t gpa, hpa_t hpa)
 705{
 706	kvm_release_page_dirty(pfn_to_page(hpa >> PAGE_SHIFT));
 707	/* mark the page always as dirty for migration */
 708	mark_page_dirty(kvm, gpa_to_gfn(gpa));
 709}
 710
 711/* unpin all blocks previously pinned by pin_blocks(), marking them dirty */
 712static void unpin_blocks(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page)
 713{
 714	struct kvm_s390_sie_block *scb_s = &vsie_page->scb_s;
 715	hpa_t hpa;
 716
 717	hpa = (u64) scb_s->scaoh << 32 | scb_s->scaol;
 718	if (hpa) {
 719		unpin_guest_page(vcpu->kvm, vsie_page->sca_gpa, hpa);
 720		vsie_page->sca_gpa = 0;
 721		scb_s->scaol = 0;
 722		scb_s->scaoh = 0;
 723	}
 724
 725	hpa = scb_s->itdba;
 726	if (hpa) {
 727		unpin_guest_page(vcpu->kvm, vsie_page->itdba_gpa, hpa);
 728		vsie_page->itdba_gpa = 0;
 729		scb_s->itdba = 0;
 730	}
 731
 732	hpa = scb_s->gvrd;
 733	if (hpa) {
 734		unpin_guest_page(vcpu->kvm, vsie_page->gvrd_gpa, hpa);
 735		vsie_page->gvrd_gpa = 0;
 736		scb_s->gvrd = 0;
 737	}
 738
 739	hpa = scb_s->riccbd;
 740	if (hpa) {
 741		unpin_guest_page(vcpu->kvm, vsie_page->riccbd_gpa, hpa);
 742		vsie_page->riccbd_gpa = 0;
 743		scb_s->riccbd = 0;
 744	}
 745
 746	hpa = scb_s->sdnxo;
 747	if (hpa) {
 748		unpin_guest_page(vcpu->kvm, vsie_page->sdnx_gpa, hpa);
 749		vsie_page->sdnx_gpa = 0;
 750		scb_s->sdnxo = 0;
 751	}
 752}
 753
 754/*
 755 * Instead of shadowing some blocks, we can simply forward them because the
 756 * addresses in the scb are 64 bit long.
 757 *
 758 * This works as long as the data lies in one page. If blocks ever exceed one
 759 * page, we have to fall back to shadowing.
 760 *
 761 * As we reuse the sca, the vcpu pointers contained in it are invalid. We must
 762 * therefore not enable any facilities that access these pointers (e.g. SIGPIF).
 763 *
 764 * Returns: - 0 if all blocks were pinned.
 765 *          - > 0 if control has to be given to guest 2
 766 *          - -ENOMEM if out of memory
 767 */
 768static int pin_blocks(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page)
 769{
 770	struct kvm_s390_sie_block *scb_o = vsie_page->scb_o;
 771	struct kvm_s390_sie_block *scb_s = &vsie_page->scb_s;
 772	hpa_t hpa;
 773	gpa_t gpa;
 774	int rc = 0;
 775
 776	gpa = READ_ONCE(scb_o->scaol) & ~0xfUL;
 777	if (test_kvm_cpu_feat(vcpu->kvm, KVM_S390_VM_CPU_FEAT_64BSCAO))
 778		gpa |= (u64) READ_ONCE(scb_o->scaoh) << 32;
 779	if (gpa) {
 780		if (gpa < 2 * PAGE_SIZE)
 781			rc = set_validity_icpt(scb_s, 0x0038U);
 782		else if ((gpa & ~0x1fffUL) == kvm_s390_get_prefix(vcpu))
 783			rc = set_validity_icpt(scb_s, 0x0011U);
 784		else if ((gpa & PAGE_MASK) !=
 785			 ((gpa + offsetof(struct bsca_block, cpu[0]) - 1) & PAGE_MASK))
 786			rc = set_validity_icpt(scb_s, 0x003bU);
 787		if (!rc) {
 788			rc = pin_guest_page(vcpu->kvm, gpa, &hpa);
 789			if (rc)
 790				rc = set_validity_icpt(scb_s, 0x0034U);
 791		}
 792		if (rc)
 793			goto unpin;
 794		vsie_page->sca_gpa = gpa;
 795		scb_s->scaoh = (u32)((u64)hpa >> 32);
 796		scb_s->scaol = (u32)(u64)hpa;
 797	}
 798
 799	gpa = READ_ONCE(scb_o->itdba) & ~0xffUL;
 800	if (gpa && (scb_s->ecb & ECB_TE)) {
 801		if (gpa < 2 * PAGE_SIZE) {
 802			rc = set_validity_icpt(scb_s, 0x0080U);
 803			goto unpin;
 804		}
 805		/* 256 bytes cannot cross page boundaries */
 806		rc = pin_guest_page(vcpu->kvm, gpa, &hpa);
 807		if (rc) {
 808			rc = set_validity_icpt(scb_s, 0x0080U);
 809			goto unpin;
 810		}
 811		vsie_page->itdba_gpa = gpa;
 812		scb_s->itdba = hpa;
 813	}
 814
 815	gpa = READ_ONCE(scb_o->gvrd) & ~0x1ffUL;
 816	if (gpa && (scb_s->eca & ECA_VX) && !(scb_s->ecd & ECD_HOSTREGMGMT)) {
 817		if (gpa < 2 * PAGE_SIZE) {
 818			rc = set_validity_icpt(scb_s, 0x1310U);
 819			goto unpin;
 820		}
 821		/*
 822		 * 512 bytes vector registers cannot cross page boundaries
 823		 * if this block gets bigger, we have to shadow it.
 824		 */
 825		rc = pin_guest_page(vcpu->kvm, gpa, &hpa);
 826		if (rc) {
 827			rc = set_validity_icpt(scb_s, 0x1310U);
 828			goto unpin;
 829		}
 830		vsie_page->gvrd_gpa = gpa;
 831		scb_s->gvrd = hpa;
 832	}
 833
 834	gpa = READ_ONCE(scb_o->riccbd) & ~0x3fUL;
 835	if (gpa && (scb_s->ecb3 & ECB3_RI)) {
 836		if (gpa < 2 * PAGE_SIZE) {
 837			rc = set_validity_icpt(scb_s, 0x0043U);
 838			goto unpin;
 839		}
 840		/* 64 bytes cannot cross page boundaries */
 841		rc = pin_guest_page(vcpu->kvm, gpa, &hpa);
 842		if (rc) {
 843			rc = set_validity_icpt(scb_s, 0x0043U);
 844			goto unpin;
 845		}
 846		/* Validity 0x0044 will be checked by SIE */
 847		vsie_page->riccbd_gpa = gpa;
 848		scb_s->riccbd = hpa;
 849	}
 850	if (((scb_s->ecb & ECB_GS) && !(scb_s->ecd & ECD_HOSTREGMGMT)) ||
 851	    (scb_s->ecd & ECD_ETOKENF)) {
 852		unsigned long sdnxc;
 853
 854		gpa = READ_ONCE(scb_o->sdnxo) & ~0xfUL;
 855		sdnxc = READ_ONCE(scb_o->sdnxo) & 0xfUL;
 856		if (!gpa || gpa < 2 * PAGE_SIZE) {
 857			rc = set_validity_icpt(scb_s, 0x10b0U);
 858			goto unpin;
 859		}
 860		if (sdnxc < 6 || sdnxc > 12) {
 861			rc = set_validity_icpt(scb_s, 0x10b1U);
 862			goto unpin;
 863		}
 864		if (gpa & ((1 << sdnxc) - 1)) {
 865			rc = set_validity_icpt(scb_s, 0x10b2U);
 866			goto unpin;
 867		}
 868		/* Due to alignment rules (checked above) this cannot
 869		 * cross page boundaries
 870		 */
 871		rc = pin_guest_page(vcpu->kvm, gpa, &hpa);
 872		if (rc) {
 873			rc = set_validity_icpt(scb_s, 0x10b0U);
 874			goto unpin;
 875		}
 876		vsie_page->sdnx_gpa = gpa;
 877		scb_s->sdnxo = hpa | sdnxc;
 878	}
 879	return 0;
 880unpin:
 881	unpin_blocks(vcpu, vsie_page);
 882	return rc;
 883}
 884
 885/* unpin the scb provided by guest 2, marking it as dirty */
 886static void unpin_scb(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page,
 887		      gpa_t gpa)
 888{
 889	hpa_t hpa = virt_to_phys(vsie_page->scb_o);
 890
 891	if (hpa)
 892		unpin_guest_page(vcpu->kvm, gpa, hpa);
 893	vsie_page->scb_o = NULL;
 894}
 895
 896/*
 897 * Pin the scb at gpa provided by guest 2 at vsie_page->scb_o.
 898 *
 899 * Returns: - 0 if the scb was pinned.
 900 *          - > 0 if control has to be given to guest 2
 901 */
 902static int pin_scb(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page,
 903		   gpa_t gpa)
 904{
 905	hpa_t hpa;
 906	int rc;
 907
 908	rc = pin_guest_page(vcpu->kvm, gpa, &hpa);
 909	if (rc) {
 910		rc = kvm_s390_inject_program_int(vcpu, PGM_ADDRESSING);
 911		WARN_ON_ONCE(rc);
 912		return 1;
 913	}
 914	vsie_page->scb_o = phys_to_virt(hpa);
 915	return 0;
 916}
 917
 918/*
 919 * Inject a fault into guest 2.
 920 *
 921 * Returns: - > 0 if control has to be given to guest 2
 922 *            < 0 if an error occurred during injection.
 923 */
 924static int inject_fault(struct kvm_vcpu *vcpu, __u16 code, __u64 vaddr,
 925			bool write_flag)
 926{
 927	struct kvm_s390_pgm_info pgm = {
 928		.code = code,
 929		.trans_exc_code =
 930			/* 0-51: virtual address */
 931			(vaddr & 0xfffffffffffff000UL) |
 932			/* 52-53: store / fetch */
 933			(((unsigned int) !write_flag) + 1) << 10,
 934			/* 62-63: asce id (always primary == 0) */
 935		.exc_access_id = 0, /* always primary */
 936		.op_access_id = 0, /* not MVPG */
 937	};
 938	int rc;
 939
 940	if (code == PGM_PROTECTION)
 941		pgm.trans_exc_code |= 0x4UL;
 942
 943	rc = kvm_s390_inject_prog_irq(vcpu, &pgm);
 944	return rc ? rc : 1;
 945}
 946
 947/*
 948 * Handle a fault during vsie execution on a gmap shadow.
 949 *
 950 * Returns: - 0 if the fault was resolved
 951 *          - > 0 if control has to be given to guest 2
 952 *          - < 0 if an error occurred
 953 */
 954static int handle_fault(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page)
 955{
 956	int rc;
 957
 958	if ((current->thread.gmap_int_code & PGM_INT_CODE_MASK) == PGM_PROTECTION)
 959		/* we can directly forward all protection exceptions */
 960		return inject_fault(vcpu, PGM_PROTECTION,
 961				    current->thread.gmap_teid.addr * PAGE_SIZE, 1);
 962
 963	rc = kvm_s390_shadow_fault(vcpu, vsie_page->gmap,
 964				   current->thread.gmap_teid.addr * PAGE_SIZE, NULL);
 965	if (rc > 0) {
 966		rc = inject_fault(vcpu, rc,
 967				  current->thread.gmap_teid.addr * PAGE_SIZE,
 968				  kvm_s390_cur_gmap_fault_is_write());
 969		if (rc >= 0)
 970			vsie_page->fault_addr = current->thread.gmap_teid.addr * PAGE_SIZE;
 971	}
 972	return rc;
 973}
 974
 975/*
 976 * Retry the previous fault that required guest 2 intervention. This avoids
 977 * one superfluous SIE re-entry and direct exit.
 978 *
 979 * Will ignore any errors. The next SIE fault will do proper fault handling.
 980 */
 981static void handle_last_fault(struct kvm_vcpu *vcpu,
 982			      struct vsie_page *vsie_page)
 983{
 984	if (vsie_page->fault_addr)
 985		kvm_s390_shadow_fault(vcpu, vsie_page->gmap,
 986				      vsie_page->fault_addr, NULL);
 987	vsie_page->fault_addr = 0;
 988}
 989
 990static inline void clear_vsie_icpt(struct vsie_page *vsie_page)
 991{
 992	vsie_page->scb_s.icptcode = 0;
 993}
 994
 995/* rewind the psw and clear the vsie icpt, so we can retry execution */
 996static void retry_vsie_icpt(struct vsie_page *vsie_page)
 997{
 998	struct kvm_s390_sie_block *scb_s = &vsie_page->scb_s;
 999	int ilen = insn_length(scb_s->ipa >> 8);
1000
1001	/* take care of EXECUTE instructions */
1002	if (scb_s->icptstatus & 1) {
1003		ilen = (scb_s->icptstatus >> 4) & 0x6;
1004		if (!ilen)
1005			ilen = 4;
1006	}
1007	scb_s->gpsw.addr = __rewind_psw(scb_s->gpsw, ilen);
1008	clear_vsie_icpt(vsie_page);
1009}
1010
1011/*
1012 * Try to shadow + enable the guest 2 provided facility list.
1013 * Retry instruction execution if enabled for and provided by guest 2.
1014 *
1015 * Returns: - 0 if handled (retry or guest 2 icpt)
1016 *          - > 0 if control has to be given to guest 2
1017 */
1018static int handle_stfle(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page)
1019{
1020	struct kvm_s390_sie_block *scb_s = &vsie_page->scb_s;
1021	__u32 fac = READ_ONCE(vsie_page->scb_o->fac);
1022
1023	/*
1024	 * Alternate-STFLE-Interpretive-Execution facilities are not supported
1025	 * -> format-0 flcb
1026	 */
1027	if (fac && test_kvm_facility(vcpu->kvm, 7)) {
1028		retry_vsie_icpt(vsie_page);
1029		/*
1030		 * The facility list origin (FLO) is in bits 1 - 28 of the FLD
1031		 * so we need to mask here before reading.
1032		 */
1033		fac = fac & 0x7ffffff8U;
1034		/*
1035		 * format-0 -> size of nested guest's facility list == guest's size
1036		 * guest's size == host's size, since STFLE is interpretatively executed
1037		 * using a format-0 for the guest, too.
1038		 */
1039		if (read_guest_real(vcpu, fac, &vsie_page->fac,
1040				    stfle_size() * sizeof(u64)))
1041			return set_validity_icpt(scb_s, 0x1090U);
1042		scb_s->fac = (u32)virt_to_phys(&vsie_page->fac);
1043	}
1044	return 0;
1045}
1046
1047/*
1048 * Get a register for a nested guest.
1049 * @vcpu the vcpu of the guest
1050 * @vsie_page the vsie_page for the nested guest
1051 * @reg the register number, the upper 4 bits are ignored.
1052 * returns: the value of the register.
1053 */
1054static u64 vsie_get_register(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page, u8 reg)
1055{
1056	/* no need to validate the parameter and/or perform error handling */
1057	reg &= 0xf;
1058	switch (reg) {
1059	case 15:
1060		return vsie_page->scb_s.gg15;
1061	case 14:
1062		return vsie_page->scb_s.gg14;
1063	default:
1064		return vcpu->run->s.regs.gprs[reg];
1065	}
1066}
1067
1068static int vsie_handle_mvpg(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page)
1069{
1070	struct kvm_s390_sie_block *scb_s = &vsie_page->scb_s;
1071	unsigned long pei_dest, pei_src, src, dest, mask, prefix;
1072	u64 *pei_block = &vsie_page->scb_o->mcic;
1073	int edat, rc_dest, rc_src;
1074	union ctlreg0 cr0;
1075
1076	cr0.val = vcpu->arch.sie_block->gcr[0];
1077	edat = cr0.edat && test_kvm_facility(vcpu->kvm, 8);
1078	mask = _kvm_s390_logical_to_effective(&scb_s->gpsw, PAGE_MASK);
1079	prefix = scb_s->prefix << GUEST_PREFIX_SHIFT;
1080
1081	dest = vsie_get_register(vcpu, vsie_page, scb_s->ipb >> 20) & mask;
1082	dest = _kvm_s390_real_to_abs(prefix, dest) + scb_s->mso;
1083	src = vsie_get_register(vcpu, vsie_page, scb_s->ipb >> 16) & mask;
1084	src = _kvm_s390_real_to_abs(prefix, src) + scb_s->mso;
1085
1086	rc_dest = kvm_s390_shadow_fault(vcpu, vsie_page->gmap, dest, &pei_dest);
1087	rc_src = kvm_s390_shadow_fault(vcpu, vsie_page->gmap, src, &pei_src);
1088	/*
1089	 * Either everything went well, or something non-critical went wrong
1090	 * e.g. because of a race. In either case, simply retry.
1091	 */
1092	if (rc_dest == -EAGAIN || rc_src == -EAGAIN || (!rc_dest && !rc_src)) {
1093		retry_vsie_icpt(vsie_page);
1094		return -EAGAIN;
1095	}
1096	/* Something more serious went wrong, propagate the error */
1097	if (rc_dest < 0)
1098		return rc_dest;
1099	if (rc_src < 0)
1100		return rc_src;
1101
1102	/* The only possible suppressing exception: just deliver it */
1103	if (rc_dest == PGM_TRANSLATION_SPEC || rc_src == PGM_TRANSLATION_SPEC) {
1104		clear_vsie_icpt(vsie_page);
1105		rc_dest = kvm_s390_inject_program_int(vcpu, PGM_TRANSLATION_SPEC);
1106		WARN_ON_ONCE(rc_dest);
1107		return 1;
1108	}
1109
1110	/*
1111	 * Forward the PEI intercept to the guest if it was a page fault, or
1112	 * also for segment and region table faults if EDAT applies.
1113	 */
1114	if (edat) {
1115		rc_dest = rc_dest == PGM_ASCE_TYPE ? rc_dest : 0;
1116		rc_src = rc_src == PGM_ASCE_TYPE ? rc_src : 0;
1117	} else {
1118		rc_dest = rc_dest != PGM_PAGE_TRANSLATION ? rc_dest : 0;
1119		rc_src = rc_src != PGM_PAGE_TRANSLATION ? rc_src : 0;
1120	}
1121	if (!rc_dest && !rc_src) {
1122		pei_block[0] = pei_dest;
1123		pei_block[1] = pei_src;
1124		return 1;
1125	}
1126
1127	retry_vsie_icpt(vsie_page);
1128
1129	/*
1130	 * The host has edat, and the guest does not, or it was an ASCE type
1131	 * exception. The host needs to inject the appropriate DAT interrupts
1132	 * into the guest.
1133	 */
1134	if (rc_dest)
1135		return inject_fault(vcpu, rc_dest, dest, 1);
1136	return inject_fault(vcpu, rc_src, src, 0);
1137}
1138
1139/*
1140 * Run the vsie on a shadow scb and a shadow gmap, without any further
1141 * sanity checks, handling SIE faults.
1142 *
1143 * Returns: - 0 everything went fine
1144 *          - > 0 if control has to be given to guest 2
1145 *          - < 0 if an error occurred
1146 */
1147static int do_vsie_run(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page)
1148	__releases(vcpu->kvm->srcu)
1149	__acquires(vcpu->kvm->srcu)
1150{
1151	struct kvm_s390_sie_block *scb_s = &vsie_page->scb_s;
1152	struct kvm_s390_sie_block *scb_o = vsie_page->scb_o;
1153	int guest_bp_isolation;
1154	int rc = 0;
1155
1156	handle_last_fault(vcpu, vsie_page);
1157
1158	kvm_vcpu_srcu_read_unlock(vcpu);
1159
1160	/* save current guest state of bp isolation override */
1161	guest_bp_isolation = test_thread_flag(TIF_ISOLATE_BP_GUEST);
1162
1163	/*
1164	 * The guest is running with BPBC, so we have to force it on for our
1165	 * nested guest. This is done by enabling BPBC globally, so the BPBC
1166	 * control in the SCB (which the nested guest can modify) is simply
1167	 * ignored.
1168	 */
1169	if (test_kvm_facility(vcpu->kvm, 82) &&
1170	    vcpu->arch.sie_block->fpf & FPF_BPBC)
1171		set_thread_flag(TIF_ISOLATE_BP_GUEST);
1172
1173	/*
1174	 * Simulate a SIE entry of the VCPU (see sie64a), so VCPU blocking
1175	 * and VCPU requests also hinder the vSIE from running and lead
1176	 * to an immediate exit. kvm_s390_vsie_kick() has to be used to
1177	 * also kick the vSIE.
1178	 */
1179	vcpu->arch.sie_block->prog0c |= PROG_IN_SIE;
1180	current->thread.gmap_int_code = 0;
1181	barrier();
1182	if (!kvm_s390_vcpu_sie_inhibited(vcpu)) {
1183xfer_to_guest_mode_check:
1184		local_irq_disable();
1185		xfer_to_guest_mode_prepare();
1186		if (xfer_to_guest_mode_work_pending()) {
1187			local_irq_enable();
1188			rc = kvm_xfer_to_guest_mode_handle_work(vcpu);
1189			if (rc)
1190				goto skip_sie;
1191			goto xfer_to_guest_mode_check;
1192		}
1193		guest_timing_enter_irqoff();
1194		rc = kvm_s390_enter_exit_sie(scb_s, vcpu->run->s.regs.gprs, vsie_page->gmap->asce);
1195		guest_timing_exit_irqoff();
1196		local_irq_enable();
1197	}
1198
1199skip_sie:
1200	barrier();
1201	vcpu->arch.sie_block->prog0c &= ~PROG_IN_SIE;
1202
1203	/* restore guest state for bp isolation override */
1204	if (!guest_bp_isolation)
1205		clear_thread_flag(TIF_ISOLATE_BP_GUEST);
1206
1207	kvm_vcpu_srcu_read_lock(vcpu);
1208
1209	if (rc == -EINTR) {
1210		VCPU_EVENT(vcpu, 3, "%s", "machine check");
1211		kvm_s390_reinject_machine_check(vcpu, &vsie_page->mcck_info);
1212		return 0;
1213	}
1214
1215	if (rc > 0)
1216		rc = 0; /* we could still have an icpt */
1217	else if (current->thread.gmap_int_code)
1218		return handle_fault(vcpu, vsie_page);
1219
1220	switch (scb_s->icptcode) {
1221	case ICPT_INST:
1222		if (scb_s->ipa == 0xb2b0)
1223			rc = handle_stfle(vcpu, vsie_page);
1224		break;
1225	case ICPT_STOP:
1226		/* stop not requested by g2 - must have been a kick */
1227		if (!(atomic_read(&scb_o->cpuflags) & CPUSTAT_STOP_INT))
1228			clear_vsie_icpt(vsie_page);
1229		break;
1230	case ICPT_VALIDITY:
1231		if ((scb_s->ipa & 0xf000) != 0xf000)
1232			scb_s->ipa += 0x1000;
1233		break;
1234	case ICPT_PARTEXEC:
1235		if (scb_s->ipa == 0xb254)
1236			rc = vsie_handle_mvpg(vcpu, vsie_page);
1237		break;
1238	}
1239	return rc;
1240}
1241
1242static void release_gmap_shadow(struct vsie_page *vsie_page)
1243{
1244	if (vsie_page->gmap)
1245		gmap_put(vsie_page->gmap);
1246	WRITE_ONCE(vsie_page->gmap, NULL);
1247	prefix_unmapped(vsie_page);
1248}
1249
1250static int acquire_gmap_shadow(struct kvm_vcpu *vcpu,
1251			       struct vsie_page *vsie_page)
1252{
1253	unsigned long asce;
1254	union ctlreg0 cr0;
1255	struct gmap *gmap;
1256	int edat;
1257
1258	asce = vcpu->arch.sie_block->gcr[1];
1259	cr0.val = vcpu->arch.sie_block->gcr[0];
1260	edat = cr0.edat && test_kvm_facility(vcpu->kvm, 8);
1261	edat += edat && test_kvm_facility(vcpu->kvm, 78);
1262
1263	/*
1264	 * ASCE or EDAT could have changed since last icpt, or the gmap
1265	 * we're holding has been unshadowed. If the gmap is still valid,
1266	 * we can safely reuse it.
1267	 */
1268	if (vsie_page->gmap && gmap_shadow_valid(vsie_page->gmap, asce, edat)) {
1269		vcpu->kvm->stat.gmap_shadow_reuse++;
1270		return 0;
1271	}
1272
1273	/* release the old shadow - if any, and mark the prefix as unmapped */
1274	release_gmap_shadow(vsie_page);
1275	gmap = gmap_shadow(vcpu->arch.gmap, asce, edat);
1276	if (IS_ERR(gmap))
1277		return PTR_ERR(gmap);
1278	vcpu->kvm->stat.gmap_shadow_create++;
1279	WRITE_ONCE(vsie_page->gmap, gmap);
1280	return 0;
1281}
1282
1283/*
1284 * Register the shadow scb at the VCPU, e.g. for kicking out of vsie.
1285 */
1286static void register_shadow_scb(struct kvm_vcpu *vcpu,
1287				struct vsie_page *vsie_page)
1288{
1289	struct kvm_s390_sie_block *scb_s = &vsie_page->scb_s;
1290
1291	WRITE_ONCE(vcpu->arch.vsie_block, &vsie_page->scb_s);
1292	/*
1293	 * External calls have to lead to a kick of the vcpu and
1294	 * therefore the vsie -> Simulate Wait state.
1295	 */
1296	kvm_s390_set_cpuflags(vcpu, CPUSTAT_WAIT);
1297	/*
1298	 * We have to adjust the g3 epoch by the g2 epoch. The epoch will
1299	 * automatically be adjusted on tod clock changes via kvm_sync_clock.
1300	 */
1301	preempt_disable();
1302	scb_s->epoch += vcpu->kvm->arch.epoch;
1303
1304	if (scb_s->ecd & ECD_MEF) {
1305		scb_s->epdx += vcpu->kvm->arch.epdx;
1306		if (scb_s->epoch < vcpu->kvm->arch.epoch)
1307			scb_s->epdx += 1;
1308	}
1309
1310	preempt_enable();
1311}
1312
1313/*
1314 * Unregister a shadow scb from a VCPU.
1315 */
1316static void unregister_shadow_scb(struct kvm_vcpu *vcpu)
1317{
1318	kvm_s390_clear_cpuflags(vcpu, CPUSTAT_WAIT);
1319	WRITE_ONCE(vcpu->arch.vsie_block, NULL);
1320}
1321
1322/*
1323 * Run the vsie on a shadowed scb, managing the gmap shadow, handling
1324 * prefix pages and faults.
1325 *
1326 * Returns: - 0 if no errors occurred
1327 *          - > 0 if control has to be given to guest 2
1328 *          - -ENOMEM if out of memory
1329 */
1330static int vsie_run(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page)
1331{
1332	struct kvm_s390_sie_block *scb_s = &vsie_page->scb_s;
1333	int rc = 0;
1334
1335	while (1) {
1336		rc = acquire_gmap_shadow(vcpu, vsie_page);
1337		if (!rc)
1338			rc = map_prefix(vcpu, vsie_page);
1339		if (!rc) {
1340			update_intervention_requests(vsie_page);
1341			rc = do_vsie_run(vcpu, vsie_page);
1342		}
1343		atomic_andnot(PROG_BLOCK_SIE, &scb_s->prog20);
1344
1345		if (rc == -EAGAIN)
1346			rc = 0;
1347
1348		/*
1349		 * Exit the loop if the guest needs to process the intercept
1350		 */
1351		if (rc || scb_s->icptcode)
1352			break;
1353
1354		/*
1355		 * Exit the loop if the host needs to process an intercept,
1356		 * but rewind the PSW to re-enter SIE once that's completed
1357		 * instead of passing a "no action" intercept to the guest.
1358		 */
1359		if (kvm_s390_vcpu_has_irq(vcpu, 0) ||
1360		    kvm_s390_vcpu_sie_inhibited(vcpu)) {
1361			kvm_s390_rewind_psw(vcpu, 4);
1362			break;
1363		}
1364	}
1365
1366	if (rc == -EFAULT) {
1367		/*
1368		 * Addressing exceptions are always presentes as intercepts.
1369		 * As addressing exceptions are suppressing and our guest 3 PSW
1370		 * points at the responsible instruction, we have to
1371		 * forward the PSW and set the ilc. If we can't read guest 3
1372		 * instruction, we can use an arbitrary ilc. Let's always use
1373		 * ilen = 4 for now, so we can avoid reading in guest 3 virtual
1374		 * memory. (we could also fake the shadow so the hardware
1375		 * handles it).
1376		 */
1377		scb_s->icptcode = ICPT_PROGI;
1378		scb_s->iprcc = PGM_ADDRESSING;
1379		scb_s->pgmilc = 4;
1380		scb_s->gpsw.addr = __rewind_psw(scb_s->gpsw, 4);
1381		rc = 1;
1382	}
1383	return rc;
1384}
1385
1386/* Try getting a given vsie page, returning "true" on success. */
1387static inline bool try_get_vsie_page(struct vsie_page *vsie_page)
1388{
1389	if (test_bit(VSIE_PAGE_IN_USE, &vsie_page->flags))
1390		return false;
1391	return !test_and_set_bit(VSIE_PAGE_IN_USE, &vsie_page->flags);
1392}
1393
1394/* Put a vsie page acquired through get_vsie_page / try_get_vsie_page. */
1395static void put_vsie_page(struct vsie_page *vsie_page)
1396{
1397	clear_bit(VSIE_PAGE_IN_USE, &vsie_page->flags);
1398}
1399
1400/*
1401 * Get or create a vsie page for a scb address.
1402 *
1403 * Returns: - address of a vsie page (cached or new one)
1404 *          - NULL if the same scb address is already used by another VCPU
1405 *          - ERR_PTR(-ENOMEM) if out of memory
1406 */
1407static struct vsie_page *get_vsie_page(struct kvm *kvm, unsigned long addr)
1408{
1409	struct vsie_page *vsie_page;
1410	int nr_vcpus;
1411
1412	rcu_read_lock();
1413	vsie_page = radix_tree_lookup(&kvm->arch.vsie.addr_to_page, addr >> 9);
1414	rcu_read_unlock();
1415	if (vsie_page) {
1416		if (try_get_vsie_page(vsie_page)) {
1417			if (vsie_page->scb_gpa == addr)
1418				return vsie_page;
1419			/*
1420			 * We raced with someone reusing + putting this vsie
1421			 * page before we grabbed it.
1422			 */
1423			put_vsie_page(vsie_page);
1424		}
1425	}
1426
1427	/*
1428	 * We want at least #online_vcpus shadows, so every VCPU can execute
1429	 * the VSIE in parallel.
1430	 */
1431	nr_vcpus = atomic_read(&kvm->online_vcpus);
1432
1433	mutex_lock(&kvm->arch.vsie.mutex);
1434	if (kvm->arch.vsie.page_count < nr_vcpus) {
1435		vsie_page = (void *)__get_free_page(GFP_KERNEL_ACCOUNT | __GFP_ZERO | GFP_DMA);
1436		if (!vsie_page) {
1437			mutex_unlock(&kvm->arch.vsie.mutex);
1438			return ERR_PTR(-ENOMEM);
1439		}
1440		__set_bit(VSIE_PAGE_IN_USE, &vsie_page->flags);
1441		kvm->arch.vsie.pages[kvm->arch.vsie.page_count] = vsie_page;
1442		kvm->arch.vsie.page_count++;
1443	} else {
1444		/* reuse an existing entry that belongs to nobody */
1445		while (true) {
1446			vsie_page = kvm->arch.vsie.pages[kvm->arch.vsie.next];
1447			if (try_get_vsie_page(vsie_page))
1448				break;
1449			kvm->arch.vsie.next++;
1450			kvm->arch.vsie.next %= nr_vcpus;
1451		}
1452		if (vsie_page->scb_gpa != ULONG_MAX)
1453			radix_tree_delete(&kvm->arch.vsie.addr_to_page,
1454					  vsie_page->scb_gpa >> 9);
1455	}
1456	/* Mark it as invalid until it resides in the tree. */
1457	vsie_page->scb_gpa = ULONG_MAX;
1458
1459	/* Double use of the same address or allocation failure. */
1460	if (radix_tree_insert(&kvm->arch.vsie.addr_to_page, addr >> 9,
1461			      vsie_page)) {
1462		put_vsie_page(vsie_page);
1463		mutex_unlock(&kvm->arch.vsie.mutex);
1464		return NULL;
1465	}
1466	vsie_page->scb_gpa = addr;
1467	mutex_unlock(&kvm->arch.vsie.mutex);
1468
1469	memset(&vsie_page->scb_s, 0, sizeof(struct kvm_s390_sie_block));
1470	release_gmap_shadow(vsie_page);
1471	vsie_page->fault_addr = 0;
1472	vsie_page->scb_s.ihcpu = 0xffffU;
1473	return vsie_page;
1474}
1475
1476int kvm_s390_handle_vsie(struct kvm_vcpu *vcpu)
1477{
1478	struct vsie_page *vsie_page;
1479	unsigned long scb_addr;
1480	int rc;
1481
1482	vcpu->stat.instruction_sie++;
1483	if (!test_kvm_cpu_feat(vcpu->kvm, KVM_S390_VM_CPU_FEAT_SIEF2))
1484		return -EOPNOTSUPP;
1485	if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
1486		return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
1487
1488	BUILD_BUG_ON(sizeof(struct vsie_page) != PAGE_SIZE);
1489	scb_addr = kvm_s390_get_base_disp_s(vcpu, NULL);
1490
1491	/* 512 byte alignment */
1492	if (unlikely(scb_addr & 0x1ffUL))
1493		return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
1494
1495	if (kvm_s390_vcpu_has_irq(vcpu, 0) || kvm_s390_vcpu_sie_inhibited(vcpu)) {
1496		kvm_s390_rewind_psw(vcpu, 4);
1497		return 0;
1498	}
1499
1500	vsie_page = get_vsie_page(vcpu->kvm, scb_addr);
1501	if (IS_ERR(vsie_page))
1502		return PTR_ERR(vsie_page);
1503	else if (!vsie_page)
1504		/* double use of sie control block - simply do nothing */
1505		return 0;
1506
1507	rc = pin_scb(vcpu, vsie_page, scb_addr);
1508	if (rc)
1509		goto out_put;
1510	rc = shadow_scb(vcpu, vsie_page);
1511	if (rc)
1512		goto out_unpin_scb;
1513	rc = pin_blocks(vcpu, vsie_page);
1514	if (rc)
1515		goto out_unshadow;
1516	register_shadow_scb(vcpu, vsie_page);
1517	rc = vsie_run(vcpu, vsie_page);
1518	unregister_shadow_scb(vcpu);
1519	unpin_blocks(vcpu, vsie_page);
1520out_unshadow:
1521	unshadow_scb(vcpu, vsie_page);
1522out_unpin_scb:
1523	unpin_scb(vcpu, vsie_page, scb_addr);
1524out_put:
1525	put_vsie_page(vsie_page);
1526
1527	return rc < 0 ? rc : 0;
1528}
1529
1530/* Init the vsie data structures. To be called when a vm is initialized. */
1531void kvm_s390_vsie_init(struct kvm *kvm)
1532{
1533	mutex_init(&kvm->arch.vsie.mutex);
1534	INIT_RADIX_TREE(&kvm->arch.vsie.addr_to_page, GFP_KERNEL_ACCOUNT);
1535}
1536
1537/* Destroy the vsie data structures. To be called when a vm is destroyed. */
1538void kvm_s390_vsie_destroy(struct kvm *kvm)
1539{
1540	struct vsie_page *vsie_page;
1541	int i;
1542
1543	mutex_lock(&kvm->arch.vsie.mutex);
1544	for (i = 0; i < kvm->arch.vsie.page_count; i++) {
1545		vsie_page = kvm->arch.vsie.pages[i];
1546		kvm->arch.vsie.pages[i] = NULL;
1547		release_gmap_shadow(vsie_page);
1548		/* free the radix tree entry */
1549		if (vsie_page->scb_gpa != ULONG_MAX)
1550			radix_tree_delete(&kvm->arch.vsie.addr_to_page,
1551					  vsie_page->scb_gpa >> 9);
1552		free_page((unsigned long)vsie_page);
1553	}
1554	kvm->arch.vsie.page_count = 0;
1555	mutex_unlock(&kvm->arch.vsie.mutex);
1556}
1557
1558void kvm_s390_vsie_kick(struct kvm_vcpu *vcpu)
1559{
1560	struct kvm_s390_sie_block *scb = READ_ONCE(vcpu->arch.vsie_block);
1561
1562	/*
1563	 * Even if the VCPU lets go of the shadow sie block reference, it is
1564	 * still valid in the cache. So we can safely kick it.
1565	 */
1566	if (scb) {
1567		atomic_or(PROG_BLOCK_SIE, &scb->prog20);
1568		if (scb->prog0c & PROG_IN_SIE)
1569			atomic_or(CPUSTAT_STOP_INT, &scb->cpuflags);
1570	}
1571}