book3s_hv_nested.c

   1// SPDX-License-Identifier: GPL-2.0
   2/*
   3 * Copyright IBM Corporation, 2018
   4 * Authors Suraj Jitindar Singh <sjitindarsingh@gmail.com>
   5 *	   Paul Mackerras <paulus@ozlabs.org>
   6 *
   7 * Description: KVM functions specific to running nested KVM-HV guests
   8 * on Book3S processors (specifically POWER9 and later).
   9 */
  10
  11#include <linux/kernel.h>
  12#include <linux/kvm_host.h>
  13#include <linux/llist.h>
  14#include <linux/pgtable.h>
  15
  16#include <asm/kvm_ppc.h>
  17#include <asm/kvm_book3s.h>
  18#include <asm/mmu.h>
  19#include <asm/pgalloc.h>
  20#include <asm/pte-walk.h>
  21#include <asm/reg.h>
  22#include <asm/plpar_wrappers.h>
  23#include <asm/firmware.h>
  24
  25static struct patb_entry *pseries_partition_tb;
  26
  27static void kvmhv_update_ptbl_cache(struct kvm_nested_guest *gp);
  28static void kvmhv_free_memslot_nest_rmap(struct kvm_memory_slot *free);
  29
  30void kvmhv_save_hv_regs(struct kvm_vcpu *vcpu, struct hv_guest_state *hr)
  31{
  32	struct kvmppc_vcore *vc = vcpu->arch.vcore;
  33
  34	hr->pcr = vc->pcr | PCR_MASK;
  35	hr->dpdes = vcpu->arch.doorbell_request;
  36	hr->hfscr = vcpu->arch.hfscr;
  37	hr->tb_offset = vc->tb_offset;
  38	hr->dawr0 = vcpu->arch.dawr0;
  39	hr->dawrx0 = vcpu->arch.dawrx0;
  40	hr->ciabr = vcpu->arch.ciabr;
  41	hr->purr = vcpu->arch.purr;
  42	hr->spurr = vcpu->arch.spurr;
  43	hr->ic = vcpu->arch.ic;
  44	hr->vtb = vc->vtb;
  45	hr->srr0 = vcpu->arch.shregs.srr0;
  46	hr->srr1 = vcpu->arch.shregs.srr1;
  47	hr->sprg[0] = vcpu->arch.shregs.sprg0;
  48	hr->sprg[1] = vcpu->arch.shregs.sprg1;
  49	hr->sprg[2] = vcpu->arch.shregs.sprg2;
  50	hr->sprg[3] = vcpu->arch.shregs.sprg3;
  51	hr->pidr = vcpu->arch.pid;
  52	hr->cfar = vcpu->arch.cfar;
  53	hr->ppr = vcpu->arch.ppr;
  54	hr->dawr1 = vcpu->arch.dawr1;
  55	hr->dawrx1 = vcpu->arch.dawrx1;
  56}
  57
  58/* Use noinline_for_stack due to https://llvm.org/pr49610 */
  59static noinline_for_stack void byteswap_pt_regs(struct pt_regs *regs)
  60{
  61	unsigned long *addr = (unsigned long *) regs;
  62
  63	for (; addr < ((unsigned long *) (regs + 1)); addr++)
  64		*addr = swab64(*addr);
  65}
  66
  67static void byteswap_hv_regs(struct hv_guest_state *hr)
  68{
  69	hr->version = swab64(hr->version);
  70	hr->lpid = swab32(hr->lpid);
  71	hr->vcpu_token = swab32(hr->vcpu_token);
  72	hr->lpcr = swab64(hr->lpcr);
  73	hr->pcr = swab64(hr->pcr) | PCR_MASK;
  74	hr->amor = swab64(hr->amor);
  75	hr->dpdes = swab64(hr->dpdes);
  76	hr->hfscr = swab64(hr->hfscr);
  77	hr->tb_offset = swab64(hr->tb_offset);
  78	hr->dawr0 = swab64(hr->dawr0);
  79	hr->dawrx0 = swab64(hr->dawrx0);
  80	hr->ciabr = swab64(hr->ciabr);
  81	hr->hdec_expiry = swab64(hr->hdec_expiry);
  82	hr->purr = swab64(hr->purr);
  83	hr->spurr = swab64(hr->spurr);
  84	hr->ic = swab64(hr->ic);
  85	hr->vtb = swab64(hr->vtb);
  86	hr->hdar = swab64(hr->hdar);
  87	hr->hdsisr = swab64(hr->hdsisr);
  88	hr->heir = swab64(hr->heir);
  89	hr->asdr = swab64(hr->asdr);
  90	hr->srr0 = swab64(hr->srr0);
  91	hr->srr1 = swab64(hr->srr1);
  92	hr->sprg[0] = swab64(hr->sprg[0]);
  93	hr->sprg[1] = swab64(hr->sprg[1]);
  94	hr->sprg[2] = swab64(hr->sprg[2]);
  95	hr->sprg[3] = swab64(hr->sprg[3]);
  96	hr->pidr = swab64(hr->pidr);
  97	hr->cfar = swab64(hr->cfar);
  98	hr->ppr = swab64(hr->ppr);
  99	hr->dawr1 = swab64(hr->dawr1);
 100	hr->dawrx1 = swab64(hr->dawrx1);
 101}
 102
 103static void save_hv_return_state(struct kvm_vcpu *vcpu,
 104				 struct hv_guest_state *hr)
 105{
 106	struct kvmppc_vcore *vc = vcpu->arch.vcore;
 107
 108	hr->dpdes = vcpu->arch.doorbell_request;
 109	hr->purr = vcpu->arch.purr;
 110	hr->spurr = vcpu->arch.spurr;
 111	hr->ic = vcpu->arch.ic;
 112	hr->vtb = vc->vtb;
 113	hr->srr0 = vcpu->arch.shregs.srr0;
 114	hr->srr1 = vcpu->arch.shregs.srr1;
 115	hr->sprg[0] = vcpu->arch.shregs.sprg0;
 116	hr->sprg[1] = vcpu->arch.shregs.sprg1;
 117	hr->sprg[2] = vcpu->arch.shregs.sprg2;
 118	hr->sprg[3] = vcpu->arch.shregs.sprg3;
 119	hr->pidr = vcpu->arch.pid;
 120	hr->cfar = vcpu->arch.cfar;
 121	hr->ppr = vcpu->arch.ppr;
 122	switch (vcpu->arch.trap) {
 123	case BOOK3S_INTERRUPT_H_DATA_STORAGE:
 124		hr->hdar = vcpu->arch.fault_dar;
 125		hr->hdsisr = vcpu->arch.fault_dsisr;
 126		hr->asdr = vcpu->arch.fault_gpa;
 127		break;
 128	case BOOK3S_INTERRUPT_H_INST_STORAGE:
 129		hr->asdr = vcpu->arch.fault_gpa;
 130		break;
 131	case BOOK3S_INTERRUPT_H_FAC_UNAVAIL:
 132		hr->hfscr = ((~HFSCR_INTR_CAUSE & hr->hfscr) |
 133			     (HFSCR_INTR_CAUSE & vcpu->arch.hfscr));
 134		break;
 135	case BOOK3S_INTERRUPT_H_EMUL_ASSIST:
 136		hr->heir = vcpu->arch.emul_inst;
 137		break;
 138	}
 139}
 140
 141static void restore_hv_regs(struct kvm_vcpu *vcpu, const struct hv_guest_state *hr)
 142{
 143	struct kvmppc_vcore *vc = vcpu->arch.vcore;
 144
 145	vc->pcr = hr->pcr | PCR_MASK;
 146	vcpu->arch.doorbell_request = hr->dpdes;
 147	vcpu->arch.hfscr = hr->hfscr;
 148	vcpu->arch.dawr0 = hr->dawr0;
 149	vcpu->arch.dawrx0 = hr->dawrx0;
 150	vcpu->arch.ciabr = hr->ciabr;
 151	vcpu->arch.purr = hr->purr;
 152	vcpu->arch.spurr = hr->spurr;
 153	vcpu->arch.ic = hr->ic;
 154	vc->vtb = hr->vtb;
 155	vcpu->arch.shregs.srr0 = hr->srr0;
 156	vcpu->arch.shregs.srr1 = hr->srr1;
 157	vcpu->arch.shregs.sprg0 = hr->sprg[0];
 158	vcpu->arch.shregs.sprg1 = hr->sprg[1];
 159	vcpu->arch.shregs.sprg2 = hr->sprg[2];
 160	vcpu->arch.shregs.sprg3 = hr->sprg[3];
 161	vcpu->arch.pid = hr->pidr;
 162	vcpu->arch.cfar = hr->cfar;
 163	vcpu->arch.ppr = hr->ppr;
 164	vcpu->arch.dawr1 = hr->dawr1;
 165	vcpu->arch.dawrx1 = hr->dawrx1;
 166}
 167
 168void kvmhv_restore_hv_return_state(struct kvm_vcpu *vcpu,
 169				   struct hv_guest_state *hr)
 170{
 171	struct kvmppc_vcore *vc = vcpu->arch.vcore;
 172
 173	/*
 174	 * This L2 vCPU might have received a doorbell while H_ENTER_NESTED was being handled.
 175	 * Make sure we preserve the doorbell if it was either:
 176	 *   a) Sent after H_ENTER_NESTED was called on this vCPU (arch.doorbell_request would be 1)
 177	 *   b) Doorbell was not handled and L2 exited for some other reason (hr->dpdes would be 1)
 178	 */
 179	vcpu->arch.doorbell_request = vcpu->arch.doorbell_request | hr->dpdes;
 180	vcpu->arch.hfscr = hr->hfscr;
 181	vcpu->arch.purr = hr->purr;
 182	vcpu->arch.spurr = hr->spurr;
 183	vcpu->arch.ic = hr->ic;
 184	vc->vtb = hr->vtb;
 185	vcpu->arch.fault_dar = hr->hdar;
 186	vcpu->arch.fault_dsisr = hr->hdsisr;
 187	vcpu->arch.fault_gpa = hr->asdr;
 188	vcpu->arch.emul_inst = hr->heir;
 189	vcpu->arch.shregs.srr0 = hr->srr0;
 190	vcpu->arch.shregs.srr1 = hr->srr1;
 191	vcpu->arch.shregs.sprg0 = hr->sprg[0];
 192	vcpu->arch.shregs.sprg1 = hr->sprg[1];
 193	vcpu->arch.shregs.sprg2 = hr->sprg[2];
 194	vcpu->arch.shregs.sprg3 = hr->sprg[3];
 195	vcpu->arch.pid = hr->pidr;
 196	vcpu->arch.cfar = hr->cfar;
 197	vcpu->arch.ppr = hr->ppr;
 198}
 199
 200static void kvmhv_nested_mmio_needed(struct kvm_vcpu *vcpu, u64 regs_ptr)
 201{
 202	/* No need to reflect the page fault to L1, we've handled it */
 203	vcpu->arch.trap = 0;
 204
 205	/*
 206	 * Since the L2 gprs have already been written back into L1 memory when
 207	 * we complete the mmio, store the L1 memory location of the L2 gpr
 208	 * being loaded into by the mmio so that the loaded value can be
 209	 * written there in kvmppc_complete_mmio_load()
 210	 */
 211	if (((vcpu->arch.io_gpr & KVM_MMIO_REG_EXT_MASK) == KVM_MMIO_REG_GPR)
 212	    && (vcpu->mmio_is_write == 0)) {
 213		vcpu->arch.nested_io_gpr = (gpa_t) regs_ptr +
 214					   offsetof(struct pt_regs,
 215						    gpr[vcpu->arch.io_gpr]);
 216		vcpu->arch.io_gpr = KVM_MMIO_REG_NESTED_GPR;
 217	}
 218}
 219
 220static int kvmhv_read_guest_state_and_regs(struct kvm_vcpu *vcpu,
 221					   struct hv_guest_state *l2_hv,
 222					   struct pt_regs *l2_regs,
 223					   u64 hv_ptr, u64 regs_ptr)
 224{
 225	int size;
 226
 227	if (kvm_vcpu_read_guest(vcpu, hv_ptr, &l2_hv->version,
 228				sizeof(l2_hv->version)))
 229		return -1;
 230
 231	if (kvmppc_need_byteswap(vcpu))
 232		l2_hv->version = swab64(l2_hv->version);
 233
 234	size = hv_guest_state_size(l2_hv->version);
 235	if (size < 0)
 236		return -1;
 237
 238	return kvm_vcpu_read_guest(vcpu, hv_ptr, l2_hv, size) ||
 239		kvm_vcpu_read_guest(vcpu, regs_ptr, l2_regs,
 240				    sizeof(struct pt_regs));
 241}
 242
 243static int kvmhv_write_guest_state_and_regs(struct kvm_vcpu *vcpu,
 244					    struct hv_guest_state *l2_hv,
 245					    struct pt_regs *l2_regs,
 246					    u64 hv_ptr, u64 regs_ptr)
 247{
 248	int size;
 249
 250	size = hv_guest_state_size(l2_hv->version);
 251	if (size < 0)
 252		return -1;
 253
 254	return kvm_vcpu_write_guest(vcpu, hv_ptr, l2_hv, size) ||
 255		kvm_vcpu_write_guest(vcpu, regs_ptr, l2_regs,
 256				     sizeof(struct pt_regs));
 257}
 258
 259static void load_l2_hv_regs(struct kvm_vcpu *vcpu,
 260			    const struct hv_guest_state *l2_hv,
 261			    const struct hv_guest_state *l1_hv, u64 *lpcr)
 262{
 263	struct kvmppc_vcore *vc = vcpu->arch.vcore;
 264	u64 mask;
 265
 266	restore_hv_regs(vcpu, l2_hv);
 267
 268	/*
 269	 * Don't let L1 change LPCR bits for the L2 except these:
 270	 */
 271	mask = LPCR_DPFD | LPCR_ILE | LPCR_TC | LPCR_AIL | LPCR_LD | LPCR_MER;
 272
 273	/*
 274	 * Additional filtering is required depending on hardware
 275	 * and configuration.
 276	 */
 277	*lpcr = kvmppc_filter_lpcr_hv(vcpu->kvm,
 278				      (vc->lpcr & ~mask) | (*lpcr & mask));
 279
 280	/*
 281	 * Don't let L1 enable features for L2 which we don't allow for L1,
 282	 * but preserve the interrupt cause field.
 283	 */
 284	vcpu->arch.hfscr = l2_hv->hfscr & (HFSCR_INTR_CAUSE | vcpu->arch.hfscr_permitted);
 285
 286	/* Don't let data address watchpoint match in hypervisor state */
 287	vcpu->arch.dawrx0 = l2_hv->dawrx0 & ~DAWRX_HYP;
 288	vcpu->arch.dawrx1 = l2_hv->dawrx1 & ~DAWRX_HYP;
 289
 290	/* Don't let completed instruction address breakpt match in HV state */
 291	if ((l2_hv->ciabr & CIABR_PRIV) == CIABR_PRIV_HYPER)
 292		vcpu->arch.ciabr = l2_hv->ciabr & ~CIABR_PRIV;
 293}
 294
 295long kvmhv_enter_nested_guest(struct kvm_vcpu *vcpu)
 296{
 297	long int err, r;
 298	struct kvm_nested_guest *l2;
 299	struct pt_regs l2_regs, saved_l1_regs;
 300	struct hv_guest_state l2_hv = {0}, saved_l1_hv;
 301	struct kvmppc_vcore *vc = vcpu->arch.vcore;
 302	u64 hv_ptr, regs_ptr;
 303	u64 hdec_exp, lpcr;
 304	s64 delta_purr, delta_spurr, delta_ic, delta_vtb;
 305
 306	if (vcpu->kvm->arch.l1_ptcr == 0)
 307		return H_NOT_AVAILABLE;
 308
 309	if (MSR_TM_TRANSACTIONAL(vcpu->arch.shregs.msr))
 310		return H_BAD_MODE;
 311
 312	/* copy parameters in */
 313	hv_ptr = kvmppc_get_gpr(vcpu, 4);
 314	regs_ptr = kvmppc_get_gpr(vcpu, 5);
 315	kvm_vcpu_srcu_read_lock(vcpu);
 316	err = kvmhv_read_guest_state_and_regs(vcpu, &l2_hv, &l2_regs,
 317					      hv_ptr, regs_ptr);
 318	kvm_vcpu_srcu_read_unlock(vcpu);
 319	if (err)
 320		return H_PARAMETER;
 321
 322	if (kvmppc_need_byteswap(vcpu))
 323		byteswap_hv_regs(&l2_hv);
 324	if (l2_hv.version > HV_GUEST_STATE_VERSION)
 325		return H_P2;
 326
 327	if (kvmppc_need_byteswap(vcpu))
 328		byteswap_pt_regs(&l2_regs);
 329	if (l2_hv.vcpu_token >= NR_CPUS)
 330		return H_PARAMETER;
 331
 332	/*
 333	 * L1 must have set up a suspended state to enter the L2 in a
 334	 * transactional state, and only in that case. These have to be
 335	 * filtered out here to prevent causing a TM Bad Thing in the
 336	 * host HRFID. We could synthesize a TM Bad Thing back to the L1
 337	 * here but there doesn't seem like much point.
 338	 */
 339	if (MSR_TM_SUSPENDED(vcpu->arch.shregs.msr)) {
 340		if (!MSR_TM_ACTIVE(l2_regs.msr))
 341			return H_BAD_MODE;
 342	} else {
 343		if (l2_regs.msr & MSR_TS_MASK)
 344			return H_BAD_MODE;
 345		if (WARN_ON_ONCE(vcpu->arch.shregs.msr & MSR_TS_MASK))
 346			return H_BAD_MODE;
 347	}
 348
 349	/* translate lpid */
 350	l2 = kvmhv_get_nested(vcpu->kvm, l2_hv.lpid, true);
 351	if (!l2)
 352		return H_PARAMETER;
 353	if (!l2->l1_gr_to_hr) {
 354		mutex_lock(&l2->tlb_lock);
 355		kvmhv_update_ptbl_cache(l2);
 356		mutex_unlock(&l2->tlb_lock);
 357	}
 358
 359	/* save l1 values of things */
 360	vcpu->arch.regs.msr = vcpu->arch.shregs.msr;
 361	saved_l1_regs = vcpu->arch.regs;
 362	kvmhv_save_hv_regs(vcpu, &saved_l1_hv);
 363
 364	/* convert TB values/offsets to host (L0) values */
 365	hdec_exp = l2_hv.hdec_expiry - vc->tb_offset;
 366	vc->tb_offset += l2_hv.tb_offset;
 367	vcpu->arch.dec_expires += l2_hv.tb_offset;
 368
 369	/* set L1 state to L2 state */
 370	vcpu->arch.nested = l2;
 371	vcpu->arch.nested_vcpu_id = l2_hv.vcpu_token;
 372	vcpu->arch.nested_hfscr = l2_hv.hfscr;
 373	vcpu->arch.regs = l2_regs;
 374
 375	/* Guest must always run with ME enabled, HV disabled. */
 376	vcpu->arch.shregs.msr = (vcpu->arch.regs.msr | MSR_ME) & ~MSR_HV;
 377
 378	lpcr = l2_hv.lpcr;
 379	load_l2_hv_regs(vcpu, &l2_hv, &saved_l1_hv, &lpcr);
 380
 381	vcpu->arch.ret = RESUME_GUEST;
 382	vcpu->arch.trap = 0;
 383	do {
 384		r = kvmhv_run_single_vcpu(vcpu, hdec_exp, lpcr);
 385	} while (is_kvmppc_resume_guest(r));
 386
 387	/* save L2 state for return */
 388	l2_regs = vcpu->arch.regs;
 389	l2_regs.msr = vcpu->arch.shregs.msr;
 390	delta_purr = vcpu->arch.purr - l2_hv.purr;
 391	delta_spurr = vcpu->arch.spurr - l2_hv.spurr;
 392	delta_ic = vcpu->arch.ic - l2_hv.ic;
 393	delta_vtb = vc->vtb - l2_hv.vtb;
 394	save_hv_return_state(vcpu, &l2_hv);
 395
 396	/* restore L1 state */
 397	vcpu->arch.nested = NULL;
 398	vcpu->arch.regs = saved_l1_regs;
 399	vcpu->arch.shregs.msr = saved_l1_regs.msr & ~MSR_TS_MASK;
 400	/* set L1 MSR TS field according to L2 transaction state */
 401	if (l2_regs.msr & MSR_TS_MASK)
 402		vcpu->arch.shregs.msr |= MSR_TS_S;
 403	vc->tb_offset = saved_l1_hv.tb_offset;
 404	/* XXX: is this always the same delta as saved_l1_hv.tb_offset? */
 405	vcpu->arch.dec_expires -= l2_hv.tb_offset;
 406	restore_hv_regs(vcpu, &saved_l1_hv);
 407	vcpu->arch.purr += delta_purr;
 408	vcpu->arch.spurr += delta_spurr;
 409	vcpu->arch.ic += delta_ic;
 410	vc->vtb += delta_vtb;
 411
 412	kvmhv_put_nested(l2);
 413
 414	/* copy l2_hv_state and regs back to guest */
 415	if (kvmppc_need_byteswap(vcpu)) {
 416		byteswap_hv_regs(&l2_hv);
 417		byteswap_pt_regs(&l2_regs);
 418	}
 419	kvm_vcpu_srcu_read_lock(vcpu);
 420	err = kvmhv_write_guest_state_and_regs(vcpu, &l2_hv, &l2_regs,
 421					       hv_ptr, regs_ptr);
 422	kvm_vcpu_srcu_read_unlock(vcpu);
 423	if (err)
 424		return H_AUTHORITY;
 425
 426	if (r == -EINTR)
 427		return H_INTERRUPT;
 428
 429	if (vcpu->mmio_needed) {
 430		kvmhv_nested_mmio_needed(vcpu, regs_ptr);
 431		return H_TOO_HARD;
 432	}
 433
 434	return vcpu->arch.trap;
 435}
 436
 437unsigned long nested_capabilities;
 438
 439long kvmhv_nested_init(void)
 440{
 441	long int ptb_order;
 442	unsigned long ptcr, host_capabilities;
 443	long rc;
 444
 445	if (!kvmhv_on_pseries())
 446		return 0;
 447	if (!radix_enabled())
 448		return -ENODEV;
 449
 450	rc = plpar_guest_get_capabilities(0, &host_capabilities);
 451	if (rc == H_SUCCESS) {
 452		unsigned long capabilities = 0;
 453
 454		if (cpu_has_feature(CPU_FTR_P11_PVR))
 455			capabilities |= H_GUEST_CAP_POWER11;
 456		if (cpu_has_feature(CPU_FTR_ARCH_31))
 457			capabilities |= H_GUEST_CAP_POWER10;
 458		if (cpu_has_feature(CPU_FTR_ARCH_300))
 459			capabilities |= H_GUEST_CAP_POWER9;
 460
 461		nested_capabilities = capabilities & host_capabilities;
 462		rc = plpar_guest_set_capabilities(0, nested_capabilities);
 463		if (rc != H_SUCCESS) {
 464			pr_err("kvm-hv: Could not configure parent hypervisor capabilities (rc=%ld)",
 465			       rc);
 466			return -ENODEV;
 467		}
 468
 469		static_branch_enable(&__kvmhv_is_nestedv2);
 470		return 0;
 471	}
 472
 473	pr_info("kvm-hv: nestedv2 get capabilities hcall failed, falling back to nestedv1 (rc=%ld)\n",
 474		rc);
 475	/* Partition table entry is 1<<4 bytes in size, hence the 4. */
 476	ptb_order = KVM_MAX_NESTED_GUESTS_SHIFT + 4;
 477	/* Minimum partition table size is 1<<12 bytes */
 478	if (ptb_order < 12)
 479		ptb_order = 12;
 480	pseries_partition_tb = kmalloc(sizeof(struct patb_entry) << ptb_order,
 481				       GFP_KERNEL);
 482	if (!pseries_partition_tb) {
 483		pr_err("kvm-hv: failed to allocated nested partition table\n");
 484		return -ENOMEM;
 485	}
 486
 487	ptcr = __pa(pseries_partition_tb) | (ptb_order - 12);
 488	rc = plpar_hcall_norets(H_SET_PARTITION_TABLE, ptcr);
 489	if (rc != H_SUCCESS) {
 490		pr_err("kvm-hv: Parent hypervisor does not support nesting (rc=%ld)\n",
 491		       rc);
 492		kfree(pseries_partition_tb);
 493		pseries_partition_tb = NULL;
 494		return -ENODEV;
 495	}
 496
 497	return 0;
 498}
 499
 500void kvmhv_nested_exit(void)
 501{
 502	/*
 503	 * N.B. the kvmhv_on_pseries() test is there because it enables
 504	 * the compiler to remove the call to plpar_hcall_norets()
 505	 * when CONFIG_PPC_PSERIES=n.
 506	 */
 507	if (kvmhv_on_pseries() && pseries_partition_tb) {
 508		plpar_hcall_norets(H_SET_PARTITION_TABLE, 0);
 509		kfree(pseries_partition_tb);
 510		pseries_partition_tb = NULL;
 511	}
 512}
 513
 514void kvmhv_flush_lpid(u64 lpid)
 515{
 516	long rc;
 517
 518	if (!kvmhv_on_pseries()) {
 519		radix__flush_all_lpid(lpid);
 520		return;
 521	}
 522
 523	if (!firmware_has_feature(FW_FEATURE_RPT_INVALIDATE))
 524		rc = plpar_hcall_norets(H_TLB_INVALIDATE, H_TLBIE_P1_ENC(2, 0, 1),
 525					lpid, TLBIEL_INVAL_SET_LPID);
 526	else
 527		rc = pseries_rpt_invalidate(lpid, H_RPTI_TARGET_CMMU,
 528					    H_RPTI_TYPE_NESTED |
 529					    H_RPTI_TYPE_TLB | H_RPTI_TYPE_PWC |
 530					    H_RPTI_TYPE_PAT,
 531					    H_RPTI_PAGE_ALL, 0, -1UL);
 532	if (rc)
 533		pr_err("KVM: TLB LPID invalidation hcall failed, rc=%ld\n", rc);
 534}
 535
 536void kvmhv_set_ptbl_entry(u64 lpid, u64 dw0, u64 dw1)
 537{
 538	if (!kvmhv_on_pseries()) {
 539		mmu_partition_table_set_entry(lpid, dw0, dw1, true);
 540		return;
 541	}
 542
 543	if (kvmhv_is_nestedv1()) {
 544		pseries_partition_tb[lpid].patb0 = cpu_to_be64(dw0);
 545		pseries_partition_tb[lpid].patb1 = cpu_to_be64(dw1);
 546		/* L0 will do the necessary barriers */
 547		kvmhv_flush_lpid(lpid);
 548	}
 549
 550	if (kvmhv_is_nestedv2())
 551		kvmhv_nestedv2_set_ptbl_entry(lpid, dw0, dw1);
 552}
 553
 554static void kvmhv_set_nested_ptbl(struct kvm_nested_guest *gp)
 555{
 556	unsigned long dw0;
 557
 558	dw0 = PATB_HR | radix__get_tree_size() |
 559		__pa(gp->shadow_pgtable) | RADIX_PGD_INDEX_SIZE;
 560	kvmhv_set_ptbl_entry(gp->shadow_lpid, dw0, gp->process_table);
 561}
 562
 563/*
 564 * Handle the H_SET_PARTITION_TABLE hcall.
 565 * r4 = guest real address of partition table + log_2(size) - 12
 566 * (formatted as for the PTCR).
 567 */
 568long kvmhv_set_partition_table(struct kvm_vcpu *vcpu)
 569{
 570	struct kvm *kvm = vcpu->kvm;
 571	unsigned long ptcr = kvmppc_get_gpr(vcpu, 4);
 572	int srcu_idx;
 573	long ret = H_SUCCESS;
 574
 575	srcu_idx = srcu_read_lock(&kvm->srcu);
 576	/* Check partition size and base address. */
 577	if ((ptcr & PRTS_MASK) + 12 - 4 > KVM_MAX_NESTED_GUESTS_SHIFT ||
 578	    !kvm_is_visible_gfn(vcpu->kvm, (ptcr & PRTB_MASK) >> PAGE_SHIFT))
 579		ret = H_PARAMETER;
 580	srcu_read_unlock(&kvm->srcu, srcu_idx);
 581	if (ret == H_SUCCESS)
 582		kvm->arch.l1_ptcr = ptcr;
 583
 584	return ret;
 585}
 586
 587/*
 588 * Handle the H_COPY_TOFROM_GUEST hcall.
 589 * r4 = L1 lpid of nested guest
 590 * r5 = pid
 591 * r6 = eaddr to access
 592 * r7 = to buffer (L1 gpa)
 593 * r8 = from buffer (L1 gpa)
 594 * r9 = n bytes to copy
 595 */
 596long kvmhv_copy_tofrom_guest_nested(struct kvm_vcpu *vcpu)
 597{
 598	struct kvm_nested_guest *gp;
 599	int l1_lpid = kvmppc_get_gpr(vcpu, 4);
 600	int pid = kvmppc_get_gpr(vcpu, 5);
 601	gva_t eaddr = kvmppc_get_gpr(vcpu, 6);
 602	gpa_t gp_to = (gpa_t) kvmppc_get_gpr(vcpu, 7);
 603	gpa_t gp_from = (gpa_t) kvmppc_get_gpr(vcpu, 8);
 604	void *buf;
 605	unsigned long n = kvmppc_get_gpr(vcpu, 9);
 606	bool is_load = !!gp_to;
 607	long rc;
 608
 609	if (gp_to && gp_from) /* One must be NULL to determine the direction */
 610		return H_PARAMETER;
 611
 612	if (eaddr & (0xFFFUL << 52))
 613		return H_PARAMETER;
 614
 615	buf = kzalloc(n, GFP_KERNEL | __GFP_NOWARN);
 616	if (!buf)
 617		return H_NO_MEM;
 618
 619	gp = kvmhv_get_nested(vcpu->kvm, l1_lpid, false);
 620	if (!gp) {
 621		rc = H_PARAMETER;
 622		goto out_free;
 623	}
 624
 625	mutex_lock(&gp->tlb_lock);
 626
 627	if (is_load) {
 628		/* Load from the nested guest into our buffer */
 629		rc = __kvmhv_copy_tofrom_guest_radix(gp->shadow_lpid, pid,
 630						     eaddr, buf, NULL, n);
 631		if (rc)
 632			goto not_found;
 633
 634		/* Write what was loaded into our buffer back to the L1 guest */
 635		kvm_vcpu_srcu_read_lock(vcpu);
 636		rc = kvm_vcpu_write_guest(vcpu, gp_to, buf, n);
 637		kvm_vcpu_srcu_read_unlock(vcpu);
 638		if (rc)
 639			goto not_found;
 640	} else {
 641		/* Load the data to be stored from the L1 guest into our buf */
 642		kvm_vcpu_srcu_read_lock(vcpu);
 643		rc = kvm_vcpu_read_guest(vcpu, gp_from, buf, n);
 644		kvm_vcpu_srcu_read_unlock(vcpu);
 645		if (rc)
 646			goto not_found;
 647
 648		/* Store from our buffer into the nested guest */
 649		rc = __kvmhv_copy_tofrom_guest_radix(gp->shadow_lpid, pid,
 650						     eaddr, NULL, buf, n);
 651		if (rc)
 652			goto not_found;
 653	}
 654
 655out_unlock:
 656	mutex_unlock(&gp->tlb_lock);
 657	kvmhv_put_nested(gp);
 658out_free:
 659	kfree(buf);
 660	return rc;
 661not_found:
 662	rc = H_NOT_FOUND;
 663	goto out_unlock;
 664}
 665
 666/*
 667 * Reload the partition table entry for a guest.
 668 * Caller must hold gp->tlb_lock.
 669 */
 670static void kvmhv_update_ptbl_cache(struct kvm_nested_guest *gp)
 671{
 672	int ret;
 673	struct patb_entry ptbl_entry;
 674	unsigned long ptbl_addr;
 675	struct kvm *kvm = gp->l1_host;
 676
 677	ret = -EFAULT;
 678	ptbl_addr = (kvm->arch.l1_ptcr & PRTB_MASK) + (gp->l1_lpid << 4);
 679	if (gp->l1_lpid < (1ul << ((kvm->arch.l1_ptcr & PRTS_MASK) + 12 - 4))) {
 680		int srcu_idx = srcu_read_lock(&kvm->srcu);
 681		ret = kvm_read_guest(kvm, ptbl_addr,
 682				     &ptbl_entry, sizeof(ptbl_entry));
 683		srcu_read_unlock(&kvm->srcu, srcu_idx);
 684	}
 685	if (ret) {
 686		gp->l1_gr_to_hr = 0;
 687		gp->process_table = 0;
 688	} else {
 689		gp->l1_gr_to_hr = be64_to_cpu(ptbl_entry.patb0);
 690		gp->process_table = be64_to_cpu(ptbl_entry.patb1);
 691	}
 692	kvmhv_set_nested_ptbl(gp);
 693}
 694
 695void kvmhv_vm_nested_init(struct kvm *kvm)
 696{
 697	idr_init(&kvm->arch.kvm_nested_guest_idr);
 698}
 699
 700static struct kvm_nested_guest *__find_nested(struct kvm *kvm, int lpid)
 701{
 702	return idr_find(&kvm->arch.kvm_nested_guest_idr, lpid);
 703}
 704
 705static bool __prealloc_nested(struct kvm *kvm, int lpid)
 706{
 707	if (idr_alloc(&kvm->arch.kvm_nested_guest_idr,
 708				NULL, lpid, lpid + 1, GFP_KERNEL) != lpid)
 709		return false;
 710	return true;
 711}
 712
 713static void __add_nested(struct kvm *kvm, int lpid, struct kvm_nested_guest *gp)
 714{
 715	if (idr_replace(&kvm->arch.kvm_nested_guest_idr, gp, lpid))
 716		WARN_ON(1);
 717}
 718
 719static void __remove_nested(struct kvm *kvm, int lpid)
 720{
 721	idr_remove(&kvm->arch.kvm_nested_guest_idr, lpid);
 722}
 723
 724static struct kvm_nested_guest *kvmhv_alloc_nested(struct kvm *kvm, unsigned int lpid)
 725{
 726	struct kvm_nested_guest *gp;
 727	long shadow_lpid;
 728
 729	gp = kzalloc(sizeof(*gp), GFP_KERNEL);
 730	if (!gp)
 731		return NULL;
 732	gp->l1_host = kvm;
 733	gp->l1_lpid = lpid;
 734	mutex_init(&gp->tlb_lock);
 735	gp->shadow_pgtable = pgd_alloc(kvm->mm);
 736	if (!gp->shadow_pgtable)
 737		goto out_free;
 738	shadow_lpid = kvmppc_alloc_lpid();
 739	if (shadow_lpid < 0)
 740		goto out_free2;
 741	gp->shadow_lpid = shadow_lpid;
 742	gp->radix = 1;
 743
 744	memset(gp->prev_cpu, -1, sizeof(gp->prev_cpu));
 745
 746	return gp;
 747
 748 out_free2:
 749	pgd_free(kvm->mm, gp->shadow_pgtable);
 750 out_free:
 751	kfree(gp);
 752	return NULL;
 753}
 754
 755/*
 756 * Free up any resources allocated for a nested guest.
 757 */
 758static void kvmhv_release_nested(struct kvm_nested_guest *gp)
 759{
 760	struct kvm *kvm = gp->l1_host;
 761
 762	if (gp->shadow_pgtable) {
 763		/*
 764		 * No vcpu is using this struct and no call to
 765		 * kvmhv_get_nested can find this struct,
 766		 * so we don't need to hold kvm->mmu_lock.
 767		 */
 768		kvmppc_free_pgtable_radix(kvm, gp->shadow_pgtable,
 769					  gp->shadow_lpid);
 770		pgd_free(kvm->mm, gp->shadow_pgtable);
 771	}
 772	kvmhv_set_ptbl_entry(gp->shadow_lpid, 0, 0);
 773	kvmppc_free_lpid(gp->shadow_lpid);
 774	kfree(gp);
 775}
 776
 777static void kvmhv_remove_nested(struct kvm_nested_guest *gp)
 778{
 779	struct kvm *kvm = gp->l1_host;
 780	int lpid = gp->l1_lpid;
 781	long ref;
 782
 783	spin_lock(&kvm->mmu_lock);
 784	if (gp == __find_nested(kvm, lpid)) {
 785		__remove_nested(kvm, lpid);
 786		--gp->refcnt;
 787	}
 788	ref = gp->refcnt;
 789	spin_unlock(&kvm->mmu_lock);
 790	if (ref == 0)
 791		kvmhv_release_nested(gp);
 792}
 793
 794/*
 795 * Free up all nested resources allocated for this guest.
 796 * This is called with no vcpus of the guest running, when
 797 * switching the guest to HPT mode or when destroying the
 798 * guest.
 799 */
 800void kvmhv_release_all_nested(struct kvm *kvm)
 801{
 802	int lpid;
 803	struct kvm_nested_guest *gp;
 804	struct kvm_nested_guest *freelist = NULL;
 805	struct kvm_memory_slot *memslot;
 806	int srcu_idx, bkt;
 807
 808	spin_lock(&kvm->mmu_lock);
 809	idr_for_each_entry(&kvm->arch.kvm_nested_guest_idr, gp, lpid) {
 810		__remove_nested(kvm, lpid);
 811		if (--gp->refcnt == 0) {
 812			gp->next = freelist;
 813			freelist = gp;
 814		}
 815	}
 816	idr_destroy(&kvm->arch.kvm_nested_guest_idr);
 817	/* idr is empty and may be reused at this point */
 818	spin_unlock(&kvm->mmu_lock);
 819	while ((gp = freelist) != NULL) {
 820		freelist = gp->next;
 821		kvmhv_release_nested(gp);
 822	}
 823
 824	srcu_idx = srcu_read_lock(&kvm->srcu);
 825	kvm_for_each_memslot(memslot, bkt, kvm_memslots(kvm))
 826		kvmhv_free_memslot_nest_rmap(memslot);
 827	srcu_read_unlock(&kvm->srcu, srcu_idx);
 828}
 829
 830/* caller must hold gp->tlb_lock */
 831static void kvmhv_flush_nested(struct kvm_nested_guest *gp)
 832{
 833	struct kvm *kvm = gp->l1_host;
 834
 835	spin_lock(&kvm->mmu_lock);
 836	kvmppc_free_pgtable_radix(kvm, gp->shadow_pgtable, gp->shadow_lpid);
 837	spin_unlock(&kvm->mmu_lock);
 838	kvmhv_flush_lpid(gp->shadow_lpid);
 839	kvmhv_update_ptbl_cache(gp);
 840	if (gp->l1_gr_to_hr == 0)
 841		kvmhv_remove_nested(gp);
 842}
 843
 844struct kvm_nested_guest *kvmhv_get_nested(struct kvm *kvm, int l1_lpid,
 845					  bool create)
 846{
 847	struct kvm_nested_guest *gp, *newgp;
 848
 849	if (l1_lpid >= (1ul << ((kvm->arch.l1_ptcr & PRTS_MASK) + 12 - 4)))
 850		return NULL;
 851
 852	spin_lock(&kvm->mmu_lock);
 853	gp = __find_nested(kvm, l1_lpid);
 854	if (gp)
 855		++gp->refcnt;
 856	spin_unlock(&kvm->mmu_lock);
 857
 858	if (gp || !create)
 859		return gp;
 860
 861	newgp = kvmhv_alloc_nested(kvm, l1_lpid);
 862	if (!newgp)
 863		return NULL;
 864
 865	if (!__prealloc_nested(kvm, l1_lpid)) {
 866		kvmhv_release_nested(newgp);
 867		return NULL;
 868	}
 869
 870	spin_lock(&kvm->mmu_lock);
 871	gp = __find_nested(kvm, l1_lpid);
 872	if (!gp) {
 873		__add_nested(kvm, l1_lpid, newgp);
 874		++newgp->refcnt;
 875		gp = newgp;
 876		newgp = NULL;
 877	}
 878	++gp->refcnt;
 879	spin_unlock(&kvm->mmu_lock);
 880
 881	if (newgp)
 882		kvmhv_release_nested(newgp);
 883
 884	return gp;
 885}
 886
 887void kvmhv_put_nested(struct kvm_nested_guest *gp)
 888{
 889	struct kvm *kvm = gp->l1_host;
 890	long ref;
 891
 892	spin_lock(&kvm->mmu_lock);
 893	ref = --gp->refcnt;
 894	spin_unlock(&kvm->mmu_lock);
 895	if (ref == 0)
 896		kvmhv_release_nested(gp);
 897}
 898
 899pte_t *find_kvm_nested_guest_pte(struct kvm *kvm, unsigned long lpid,
 900				 unsigned long ea, unsigned *hshift)
 901{
 902	struct kvm_nested_guest *gp;
 903	pte_t *pte;
 904
 905	gp = __find_nested(kvm, lpid);
 906	if (!gp)
 907		return NULL;
 908
 909	VM_WARN(!spin_is_locked(&kvm->mmu_lock),
 910		"%s called with kvm mmu_lock not held \n", __func__);
 911	pte = __find_linux_pte(gp->shadow_pgtable, ea, NULL, hshift);
 912
 913	return pte;
 914}
 915
 916static inline bool kvmhv_n_rmap_is_equal(u64 rmap_1, u64 rmap_2)
 917{
 918	return !((rmap_1 ^ rmap_2) & (RMAP_NESTED_LPID_MASK |
 919				       RMAP_NESTED_GPA_MASK));
 920}
 921
 922void kvmhv_insert_nest_rmap(struct kvm *kvm, unsigned long *rmapp,
 923			    struct rmap_nested **n_rmap)
 924{
 925	struct llist_node *entry = ((struct llist_head *) rmapp)->first;
 926	struct rmap_nested *cursor;
 927	u64 rmap, new_rmap = (*n_rmap)->rmap;
 928
 929	/* Are there any existing entries? */
 930	if (!(*rmapp)) {
 931		/* No -> use the rmap as a single entry */
 932		*rmapp = new_rmap | RMAP_NESTED_IS_SINGLE_ENTRY;
 933		return;
 934	}
 935
 936	/* Do any entries match what we're trying to insert? */
 937	for_each_nest_rmap_safe(cursor, entry, &rmap) {
 938		if (kvmhv_n_rmap_is_equal(rmap, new_rmap))
 939			return;
 940	}
 941
 942	/* Do we need to create a list or just add the new entry? */
 943	rmap = *rmapp;
 944	if (rmap & RMAP_NESTED_IS_SINGLE_ENTRY) /* Not previously a list */
 945		*rmapp = 0UL;
 946	llist_add(&((*n_rmap)->list), (struct llist_head *) rmapp);
 947	if (rmap & RMAP_NESTED_IS_SINGLE_ENTRY) /* Not previously a list */
 948		(*n_rmap)->list.next = (struct llist_node *) rmap;
 949
 950	/* Set NULL so not freed by caller */
 951	*n_rmap = NULL;
 952}
 953
 954static void kvmhv_update_nest_rmap_rc(struct kvm *kvm, u64 n_rmap,
 955				      unsigned long clr, unsigned long set,
 956				      unsigned long hpa, unsigned long mask)
 957{
 958	unsigned long gpa;
 959	unsigned int shift, lpid;
 960	pte_t *ptep;
 961
 962	gpa = n_rmap & RMAP_NESTED_GPA_MASK;
 963	lpid = (n_rmap & RMAP_NESTED_LPID_MASK) >> RMAP_NESTED_LPID_SHIFT;
 964
 965	/* Find the pte */
 966	ptep = find_kvm_nested_guest_pte(kvm, lpid, gpa, &shift);
 967	/*
 968	 * If the pte is present and the pfn is still the same, update the pte.
 969	 * If the pfn has changed then this is a stale rmap entry, the nested
 970	 * gpa actually points somewhere else now, and there is nothing to do.
 971	 * XXX A future optimisation would be to remove the rmap entry here.
 972	 */
 973	if (ptep && pte_present(*ptep) && ((pte_val(*ptep) & mask) == hpa)) {
 974		__radix_pte_update(ptep, clr, set);
 975		kvmppc_radix_tlbie_page(kvm, gpa, shift, lpid);
 976	}
 977}
 978
 979/*
 980 * For a given list of rmap entries, update the rc bits in all ptes in shadow
 981 * page tables for nested guests which are referenced by the rmap list.
 982 */
 983void kvmhv_update_nest_rmap_rc_list(struct kvm *kvm, unsigned long *rmapp,
 984				    unsigned long clr, unsigned long set,
 985				    unsigned long hpa, unsigned long nbytes)
 986{
 987	struct llist_node *entry = ((struct llist_head *) rmapp)->first;
 988	struct rmap_nested *cursor;
 989	unsigned long rmap, mask;
 990
 991	if ((clr | set) & ~(_PAGE_DIRTY | _PAGE_ACCESSED))
 992		return;
 993
 994	mask = PTE_RPN_MASK & ~(nbytes - 1);
 995	hpa &= mask;
 996
 997	for_each_nest_rmap_safe(cursor, entry, &rmap)
 998		kvmhv_update_nest_rmap_rc(kvm, rmap, clr, set, hpa, mask);
 999}
1000
1001static void kvmhv_remove_nest_rmap(struct kvm *kvm, u64 n_rmap,
1002				   unsigned long hpa, unsigned long mask)
1003{
1004	struct kvm_nested_guest *gp;
1005	unsigned long gpa;
1006	unsigned int shift, lpid;
1007	pte_t *ptep;
1008
1009	gpa = n_rmap & RMAP_NESTED_GPA_MASK;
1010	lpid = (n_rmap & RMAP_NESTED_LPID_MASK) >> RMAP_NESTED_LPID_SHIFT;
1011	gp = __find_nested(kvm, lpid);
1012	if (!gp)
1013		return;
1014
1015	/* Find and invalidate the pte */
1016	ptep = find_kvm_nested_guest_pte(kvm, lpid, gpa, &shift);
1017	/* Don't spuriously invalidate ptes if the pfn has changed */
1018	if (ptep && pte_present(*ptep) && ((pte_val(*ptep) & mask) == hpa))
1019		kvmppc_unmap_pte(kvm, ptep, gpa, shift, NULL, gp->shadow_lpid);
1020}
1021
1022static void kvmhv_remove_nest_rmap_list(struct kvm *kvm, unsigned long *rmapp,
1023					unsigned long hpa, unsigned long mask)
1024{
1025	struct llist_node *entry = llist_del_all((struct llist_head *) rmapp);
1026	struct rmap_nested *cursor;
1027	unsigned long rmap;
1028
1029	for_each_nest_rmap_safe(cursor, entry, &rmap) {
1030		kvmhv_remove_nest_rmap(kvm, rmap, hpa, mask);
1031		kfree(cursor);
1032	}
1033}
1034
1035/* called with kvm->mmu_lock held */
1036void kvmhv_remove_nest_rmap_range(struct kvm *kvm,
1037				  const struct kvm_memory_slot *memslot,
1038				  unsigned long gpa, unsigned long hpa,
1039				  unsigned long nbytes)
1040{
1041	unsigned long gfn, end_gfn;
1042	unsigned long addr_mask;
1043
1044	if (!memslot)
1045		return;
1046	gfn = (gpa >> PAGE_SHIFT) - memslot->base_gfn;
1047	end_gfn = gfn + (nbytes >> PAGE_SHIFT);
1048
1049	addr_mask = PTE_RPN_MASK & ~(nbytes - 1);
1050	hpa &= addr_mask;
1051
1052	for (; gfn < end_gfn; gfn++) {
1053		unsigned long *rmap = &memslot->arch.rmap[gfn];
1054		kvmhv_remove_nest_rmap_list(kvm, rmap, hpa, addr_mask);
1055	}
1056}
1057
1058static void kvmhv_free_memslot_nest_rmap(struct kvm_memory_slot *free)
1059{
1060	unsigned long page;
1061
1062	for (page = 0; page < free->npages; page++) {
1063		unsigned long rmap, *rmapp = &free->arch.rmap[page];
1064		struct rmap_nested *cursor;
1065		struct llist_node *entry;
1066
1067		entry = llist_del_all((struct llist_head *) rmapp);
1068		for_each_nest_rmap_safe(cursor, entry, &rmap)
1069			kfree(cursor);
1070	}
1071}
1072
1073static bool kvmhv_invalidate_shadow_pte(struct kvm_vcpu *vcpu,
1074					struct kvm_nested_guest *gp,
1075					long gpa, int *shift_ret)
1076{
1077	struct kvm *kvm = vcpu->kvm;
1078	bool ret = false;
1079	pte_t *ptep;
1080	int shift;
1081
1082	spin_lock(&kvm->mmu_lock);
1083	ptep = find_kvm_nested_guest_pte(kvm, gp->l1_lpid, gpa, &shift);
1084	if (!shift)
1085		shift = PAGE_SHIFT;
1086	if (ptep && pte_present(*ptep)) {
1087		kvmppc_unmap_pte(kvm, ptep, gpa, shift, NULL, gp->shadow_lpid);
1088		ret = true;
1089	}
1090	spin_unlock(&kvm->mmu_lock);
1091
1092	if (shift_ret)
1093		*shift_ret = shift;
1094	return ret;
1095}
1096
1097static inline int get_ric(unsigned int instr)
1098{
1099	return (instr >> 18) & 0x3;
1100}
1101
1102static inline int get_prs(unsigned int instr)
1103{
1104	return (instr >> 17) & 0x1;
1105}
1106
1107static inline int get_r(unsigned int instr)
1108{
1109	return (instr >> 16) & 0x1;
1110}
1111
1112static inline int get_lpid(unsigned long r_val)
1113{
1114	return r_val & 0xffffffff;
1115}
1116
1117static inline int get_is(unsigned long r_val)
1118{
1119	return (r_val >> 10) & 0x3;
1120}
1121
1122static inline int get_ap(unsigned long r_val)
1123{
1124	return (r_val >> 5) & 0x7;
1125}
1126
1127static inline long get_epn(unsigned long r_val)
1128{
1129	return r_val >> 12;
1130}
1131
1132static int kvmhv_emulate_tlbie_tlb_addr(struct kvm_vcpu *vcpu, int lpid,
1133					int ap, long epn)
1134{
1135	struct kvm *kvm = vcpu->kvm;
1136	struct kvm_nested_guest *gp;
1137	long npages;
1138	int shift, shadow_shift;
1139	unsigned long addr;
1140
1141	shift = ap_to_shift(ap);
1142	addr = epn << 12;
1143	if (shift < 0)
1144		/* Invalid ap encoding */
1145		return -EINVAL;
1146
1147	addr &= ~((1UL << shift) - 1);
1148	npages = 1UL << (shift - PAGE_SHIFT);
1149
1150	gp = kvmhv_get_nested(kvm, lpid, false);
1151	if (!gp) /* No such guest -> nothing to do */
1152		return 0;
1153	mutex_lock(&gp->tlb_lock);
1154
1155	/* There may be more than one host page backing this single guest pte */
1156	do {
1157		kvmhv_invalidate_shadow_pte(vcpu, gp, addr, &shadow_shift);
1158
1159		npages -= 1UL << (shadow_shift - PAGE_SHIFT);
1160		addr += 1UL << shadow_shift;
1161	} while (npages > 0);
1162
1163	mutex_unlock(&gp->tlb_lock);
1164	kvmhv_put_nested(gp);
1165	return 0;
1166}
1167
1168static void kvmhv_emulate_tlbie_lpid(struct kvm_vcpu *vcpu,
1169				     struct kvm_nested_guest *gp, int ric)
1170{
1171	struct kvm *kvm = vcpu->kvm;
1172
1173	mutex_lock(&gp->tlb_lock);
1174	switch (ric) {
1175	case 0:
1176		/* Invalidate TLB */
1177		spin_lock(&kvm->mmu_lock);
1178		kvmppc_free_pgtable_radix(kvm, gp->shadow_pgtable,
1179					  gp->shadow_lpid);
1180		kvmhv_flush_lpid(gp->shadow_lpid);
1181		spin_unlock(&kvm->mmu_lock);
1182		break;
1183	case 1:
1184		/*
1185		 * Invalidate PWC
1186		 * We don't cache this -> nothing to do
1187		 */
1188		break;
1189	case 2:
1190		/* Invalidate TLB, PWC and caching of partition table entries */
1191		kvmhv_flush_nested(gp);
1192		break;
1193	default:
1194		break;
1195	}
1196	mutex_unlock(&gp->tlb_lock);
1197}
1198
1199static void kvmhv_emulate_tlbie_all_lpid(struct kvm_vcpu *vcpu, int ric)
1200{
1201	struct kvm *kvm = vcpu->kvm;
1202	struct kvm_nested_guest *gp;
1203	int lpid;
1204
1205	spin_lock(&kvm->mmu_lock);
1206	idr_for_each_entry(&kvm->arch.kvm_nested_guest_idr, gp, lpid) {
1207		spin_unlock(&kvm->mmu_lock);
1208		kvmhv_emulate_tlbie_lpid(vcpu, gp, ric);
1209		spin_lock(&kvm->mmu_lock);
1210	}
1211	spin_unlock(&kvm->mmu_lock);
1212}
1213
1214static int kvmhv_emulate_priv_tlbie(struct kvm_vcpu *vcpu, unsigned int instr,
1215				    unsigned long rsval, unsigned long rbval)
1216{
1217	struct kvm *kvm = vcpu->kvm;
1218	struct kvm_nested_guest *gp;
1219	int r, ric, prs, is, ap;
1220	int lpid;
1221	long epn;
1222	int ret = 0;
1223
1224	ric = get_ric(instr);
1225	prs = get_prs(instr);
1226	r = get_r(instr);
1227	lpid = get_lpid(rsval);
1228	is = get_is(rbval);
1229
1230	/*
1231	 * These cases are invalid and are not handled:
1232	 * r   != 1 -> Only radix supported
1233	 * prs == 1 -> Not HV privileged
1234	 * ric == 3 -> No cluster bombs for radix
1235	 * is  == 1 -> Partition scoped translations not associated with pid
1236	 * (!is) && (ric == 1 || ric == 2) -> Not supported by ISA
1237	 */
1238	if ((!r) || (prs) || (ric == 3) || (is == 1) ||
1239	    ((!is) && (ric == 1 || ric == 2)))
1240		return -EINVAL;
1241
1242	switch (is) {
1243	case 0:
1244		/*
1245		 * We know ric == 0
1246		 * Invalidate TLB for a given target address
1247		 */
1248		epn = get_epn(rbval);
1249		ap = get_ap(rbval);
1250		ret = kvmhv_emulate_tlbie_tlb_addr(vcpu, lpid, ap, epn);
1251		break;
1252	case 2:
1253		/* Invalidate matching LPID */
1254		gp = kvmhv_get_nested(kvm, lpid, false);
1255		if (gp) {
1256			kvmhv_emulate_tlbie_lpid(vcpu, gp, ric);
1257			kvmhv_put_nested(gp);
1258		}
1259		break;
1260	case 3:
1261		/* Invalidate ALL LPIDs */
1262		kvmhv_emulate_tlbie_all_lpid(vcpu, ric);
1263		break;
1264	default:
1265		ret = -EINVAL;
1266		break;
1267	}
1268
1269	return ret;
1270}
1271
1272/*
1273 * This handles the H_TLB_INVALIDATE hcall.
1274 * Parameters are (r4) tlbie instruction code, (r5) rS contents,
1275 * (r6) rB contents.
1276 */
1277long kvmhv_do_nested_tlbie(struct kvm_vcpu *vcpu)
1278{
1279	int ret;
1280
1281	ret = kvmhv_emulate_priv_tlbie(vcpu, kvmppc_get_gpr(vcpu, 4),
1282			kvmppc_get_gpr(vcpu, 5), kvmppc_get_gpr(vcpu, 6));
1283	if (ret)
1284		return H_PARAMETER;
1285	return H_SUCCESS;
1286}
1287
1288static long do_tlb_invalidate_nested_all(struct kvm_vcpu *vcpu,
1289					 unsigned long lpid, unsigned long ric)
1290{
1291	struct kvm *kvm = vcpu->kvm;
1292	struct kvm_nested_guest *gp;
1293
1294	gp = kvmhv_get_nested(kvm, lpid, false);
1295	if (gp) {
1296		kvmhv_emulate_tlbie_lpid(vcpu, gp, ric);
1297		kvmhv_put_nested(gp);
1298	}
1299	return H_SUCCESS;
1300}
1301
1302/*
1303 * Number of pages above which we invalidate the entire LPID rather than
1304 * flush individual pages.
1305 */
1306static unsigned long tlb_range_flush_page_ceiling __read_mostly = 33;
1307
1308static long do_tlb_invalidate_nested_tlb(struct kvm_vcpu *vcpu,
1309					 unsigned long lpid,
1310					 unsigned long pg_sizes,
1311					 unsigned long start,
1312					 unsigned long end)
1313{
1314	int ret = H_P4;
1315	unsigned long addr, nr_pages;
1316	struct mmu_psize_def *def;
1317	unsigned long psize, ap, page_size;
1318	bool flush_lpid;
1319
1320	for (psize = 0; psize < MMU_PAGE_COUNT; psize++) {
1321		def = &mmu_psize_defs[psize];
1322		if (!(pg_sizes & def->h_rpt_pgsize))
1323			continue;
1324
1325		nr_pages = (end - start) >> def->shift;
1326		flush_lpid = nr_pages > tlb_range_flush_page_ceiling;
1327		if (flush_lpid)
1328			return do_tlb_invalidate_nested_all(vcpu, lpid,
1329							RIC_FLUSH_TLB);
1330		addr = start;
1331		ap = mmu_get_ap(psize);
1332		page_size = 1UL << def->shift;
1333		do {
1334			ret = kvmhv_emulate_tlbie_tlb_addr(vcpu, lpid, ap,
1335						   get_epn(addr));
1336			if (ret)
1337				return H_P4;
1338			addr += page_size;
1339		} while (addr < end);
1340	}
1341	return ret;
1342}
1343
1344/*
1345 * Performs partition-scoped invalidations for nested guests
1346 * as part of H_RPT_INVALIDATE hcall.
1347 */
1348long do_h_rpt_invalidate_pat(struct kvm_vcpu *vcpu, unsigned long lpid,
1349			     unsigned long type, unsigned long pg_sizes,
1350			     unsigned long start, unsigned long end)
1351{
1352	/*
1353	 * If L2 lpid isn't valid, we need to return H_PARAMETER.
1354	 *
1355	 * However, nested KVM issues a L2 lpid flush call when creating
1356	 * partition table entries for L2. This happens even before the
1357	 * corresponding shadow lpid is created in HV which happens in
1358	 * H_ENTER_NESTED call. Since we can't differentiate this case from
1359	 * the invalid case, we ignore such flush requests and return success.
1360	 */
1361	if (!__find_nested(vcpu->kvm, lpid))
1362		return H_SUCCESS;
1363
1364	/*
1365	 * A flush all request can be handled by a full lpid flush only.
1366	 */
1367	if ((type & H_RPTI_TYPE_NESTED_ALL) == H_RPTI_TYPE_NESTED_ALL)
1368		return do_tlb_invalidate_nested_all(vcpu, lpid, RIC_FLUSH_ALL);
1369
1370	/*
1371	 * We don't need to handle a PWC flush like process table here,
1372	 * because intermediate partition scoped table in nested guest doesn't
1373	 * really have PWC. Only level we have PWC is in L0 and for nested
1374	 * invalidate at L0 we always do kvm_flush_lpid() which does
1375	 * radix__flush_all_lpid(). For range invalidate at any level, we
1376	 * are not removing the higher level page tables and hence there is
1377	 * no PWC invalidate needed.
1378	 *
1379	 * if (type & H_RPTI_TYPE_PWC) {
1380	 *	ret = do_tlb_invalidate_nested_all(vcpu, lpid, RIC_FLUSH_PWC);
1381	 *	if (ret)
1382	 *		return H_P4;
1383	 * }
1384	 */
1385
1386	if (start == 0 && end == -1)
1387		return do_tlb_invalidate_nested_all(vcpu, lpid, RIC_FLUSH_TLB);
1388
1389	if (type & H_RPTI_TYPE_TLB)
1390		return do_tlb_invalidate_nested_tlb(vcpu, lpid, pg_sizes,
1391						    start, end);
1392	return H_SUCCESS;
1393}
1394
1395/* Used to convert a nested guest real address to a L1 guest real address */
1396static int kvmhv_translate_addr_nested(struct kvm_vcpu *vcpu,
1397				       struct kvm_nested_guest *gp,
1398				       unsigned long n_gpa, unsigned long dsisr,
1399				       struct kvmppc_pte *gpte_p)
1400{
1401	u64 fault_addr, flags = dsisr & DSISR_ISSTORE;
1402	int ret;
1403
1404	ret = kvmppc_mmu_walk_radix_tree(vcpu, n_gpa, gpte_p, gp->l1_gr_to_hr,
1405					 &fault_addr);
1406
1407	if (ret) {
1408		/* We didn't find a pte */
1409		if (ret == -EINVAL) {
1410			/* Unsupported mmu config */
1411			flags |= DSISR_UNSUPP_MMU;
1412		} else if (ret == -ENOENT) {
1413			/* No translation found */
1414			flags |= DSISR_NOHPTE;
1415		} else if (ret == -EFAULT) {
1416			/* Couldn't access L1 real address */
1417			flags |= DSISR_PRTABLE_FAULT;
1418			vcpu->arch.fault_gpa = fault_addr;
1419		} else {
1420			/* Unknown error */
1421			return ret;
1422		}
1423		goto forward_to_l1;
1424	} else {
1425		/* We found a pte -> check permissions */
1426		if (dsisr & DSISR_ISSTORE) {
1427			/* Can we write? */
1428			if (!gpte_p->may_write) {
1429				flags |= DSISR_PROTFAULT;
1430				goto forward_to_l1;
1431			}
1432		} else if (vcpu->arch.trap == BOOK3S_INTERRUPT_H_INST_STORAGE) {
1433			/* Can we execute? */
1434			if (!gpte_p->may_execute) {
1435				flags |= SRR1_ISI_N_G_OR_CIP;
1436				goto forward_to_l1;
1437			}
1438		} else {
1439			/* Can we read? */
1440			if (!gpte_p->may_read && !gpte_p->may_write) {
1441				flags |= DSISR_PROTFAULT;
1442				goto forward_to_l1;
1443			}
1444		}
1445	}
1446
1447	return 0;
1448
1449forward_to_l1:
1450	vcpu->arch.fault_dsisr = flags;
1451	if (vcpu->arch.trap == BOOK3S_INTERRUPT_H_INST_STORAGE) {
1452		vcpu->arch.shregs.msr &= SRR1_MSR_BITS;
1453		vcpu->arch.shregs.msr |= flags;
1454	}
1455	return RESUME_HOST;
1456}
1457
1458static long kvmhv_handle_nested_set_rc(struct kvm_vcpu *vcpu,
1459				       struct kvm_nested_guest *gp,
1460				       unsigned long n_gpa,
1461				       struct kvmppc_pte gpte,
1462				       unsigned long dsisr)
1463{
1464	struct kvm *kvm = vcpu->kvm;
1465	bool writing = !!(dsisr & DSISR_ISSTORE);
1466	u64 pgflags;
1467	long ret;
1468
1469	/* Are the rc bits set in the L1 partition scoped pte? */
1470	pgflags = _PAGE_ACCESSED;
1471	if (writing)
1472		pgflags |= _PAGE_DIRTY;
1473	if (pgflags & ~gpte.rc)
1474		return RESUME_HOST;
1475
1476	spin_lock(&kvm->mmu_lock);
1477	/* Set the rc bit in the pte of our (L0) pgtable for the L1 guest */
1478	ret = kvmppc_hv_handle_set_rc(kvm, false, writing,
1479				      gpte.raddr, kvm->arch.lpid);
1480	if (!ret) {
1481		ret = -EINVAL;
1482		goto out_unlock;
1483	}
1484
1485	/* Set the rc bit in the pte of the shadow_pgtable for the nest guest */
1486	ret = kvmppc_hv_handle_set_rc(kvm, true, writing,
1487				      n_gpa, gp->l1_lpid);
1488	if (!ret)
1489		ret = -EINVAL;
1490	else
1491		ret = 0;
1492
1493out_unlock:
1494	spin_unlock(&kvm->mmu_lock);
1495	return ret;
1496}
1497
1498static inline int kvmppc_radix_level_to_shift(int level)
1499{
1500	switch (level) {
1501	case 2:
1502		return PUD_SHIFT;
1503	case 1:
1504		return PMD_SHIFT;
1505	default:
1506		return PAGE_SHIFT;
1507	}
1508}
1509
1510static inline int kvmppc_radix_shift_to_level(int shift)
1511{
1512	if (shift == PUD_SHIFT)
1513		return 2;
1514	if (shift == PMD_SHIFT)
1515		return 1;
1516	if (shift == PAGE_SHIFT)
1517		return 0;
1518	WARN_ON_ONCE(1);
1519	return 0;
1520}
1521
1522/* called with gp->tlb_lock held */
1523static long int __kvmhv_nested_page_fault(struct kvm_vcpu *vcpu,
1524					  struct kvm_nested_guest *gp)
1525{
1526	struct kvm *kvm = vcpu->kvm;
1527	struct kvm_memory_slot *memslot;
1528	struct rmap_nested *n_rmap;
1529	struct kvmppc_pte gpte;
1530	pte_t pte, *pte_p;
1531	unsigned long mmu_seq;
1532	unsigned long dsisr = vcpu->arch.fault_dsisr;
1533	unsigned long ea = vcpu->arch.fault_dar;
1534	unsigned long *rmapp;
1535	unsigned long n_gpa, gpa, gfn, perm = 0UL;
1536	unsigned int shift, l1_shift, level;
1537	bool writing = !!(dsisr & DSISR_ISSTORE);
1538	long int ret;
1539
1540	if (!gp->l1_gr_to_hr) {
1541		kvmhv_update_ptbl_cache(gp);
1542		if (!gp->l1_gr_to_hr)
1543			return RESUME_HOST;
1544	}
1545
1546	/* Convert the nested guest real address into a L1 guest real address */
1547
1548	n_gpa = vcpu->arch.fault_gpa & ~0xF000000000000FFFULL;
1549	if (!(dsisr & DSISR_PRTABLE_FAULT))
1550		n_gpa |= ea & 0xFFF;
1551	ret = kvmhv_translate_addr_nested(vcpu, gp, n_gpa, dsisr, &gpte);
1552
1553	/*
1554	 * If the hardware found a translation but we don't now have a usable
1555	 * translation in the l1 partition-scoped tree, remove the shadow pte
1556	 * and let the guest retry.
1557	 */
1558	if (ret == RESUME_HOST &&
1559	    (dsisr & (DSISR_PROTFAULT | DSISR_BADACCESS | DSISR_NOEXEC_OR_G |
1560		      DSISR_BAD_COPYPASTE)))
1561		goto inval;
1562	if (ret)
1563		return ret;
1564
1565	/* Failed to set the reference/change bits */
1566	if (dsisr & DSISR_SET_RC) {
1567		ret = kvmhv_handle_nested_set_rc(vcpu, gp, n_gpa, gpte, dsisr);
1568		if (ret == RESUME_HOST)
1569			return ret;
1570		if (ret)
1571			goto inval;
1572		dsisr &= ~DSISR_SET_RC;
1573		if (!(dsisr & (DSISR_BAD_FAULT_64S | DSISR_NOHPTE |
1574			       DSISR_PROTFAULT)))
1575			return RESUME_GUEST;
1576	}
1577
1578	/*
1579	 * We took an HISI or HDSI while we were running a nested guest which
1580	 * means we have no partition scoped translation for that. This means
1581	 * we need to insert a pte for the mapping into our shadow_pgtable.
1582	 */
1583
1584	l1_shift = gpte.page_shift;
1585	if (l1_shift < PAGE_SHIFT) {
1586		/* We don't support l1 using a page size smaller than our own */
1587		pr_err("KVM: L1 guest page shift (%d) less than our own (%d)\n",
1588			l1_shift, PAGE_SHIFT);
1589		return -EINVAL;
1590	}
1591	gpa = gpte.raddr;
1592	gfn = gpa >> PAGE_SHIFT;
1593
1594	/* 1. Get the corresponding host memslot */
1595
1596	memslot = gfn_to_memslot(kvm, gfn);
1597	if (!memslot || (memslot->flags & KVM_MEMSLOT_INVALID)) {
1598		if (dsisr & (DSISR_PRTABLE_FAULT | DSISR_BADACCESS)) {
1599			/* unusual error -> reflect to the guest as a DSI */
1600			kvmppc_core_queue_data_storage(vcpu,
1601					kvmppc_get_msr(vcpu) & SRR1_PREFIXED,
1602					ea, dsisr);
1603			return RESUME_GUEST;
1604		}
1605
1606		/* passthrough of emulated MMIO case */
1607		return kvmppc_hv_emulate_mmio(vcpu, gpa, ea, writing);
1608	}
1609	if (memslot->flags & KVM_MEM_READONLY) {
1610		if (writing) {
1611			/* Give the guest a DSI */
1612			kvmppc_core_queue_data_storage(vcpu,
1613					kvmppc_get_msr(vcpu) & SRR1_PREFIXED,
1614					ea, DSISR_ISSTORE | DSISR_PROTFAULT);
1615			return RESUME_GUEST;
1616		}
1617	}
1618
1619	/* 2. Find the host pte for this L1 guest real address */
1620
1621	/* Used to check for invalidations in progress */
1622	mmu_seq = kvm->mmu_invalidate_seq;
1623	smp_rmb();
1624
1625	/* See if can find translation in our partition scoped tables for L1 */
1626	pte = __pte(0);
1627	spin_lock(&kvm->mmu_lock);
1628	pte_p = find_kvm_secondary_pte(kvm, gpa, &shift);
1629	if (!shift)
1630		shift = PAGE_SHIFT;
1631	if (pte_p)
1632		pte = *pte_p;
1633	spin_unlock(&kvm->mmu_lock);
1634
1635	if (!pte_present(pte) || (writing && !(pte_val(pte) & _PAGE_WRITE))) {
1636		/* No suitable pte found -> try to insert a mapping */
1637		ret = kvmppc_book3s_instantiate_page(vcpu, gpa, memslot,
1638					writing, &pte, &level);
1639		if (ret == -EAGAIN)
1640			return RESUME_GUEST;
1641		else if (ret)
1642			return ret;
1643		shift = kvmppc_radix_level_to_shift(level);
1644	}
1645	/* Align gfn to the start of the page */
1646	gfn = (gpa & ~((1UL << shift) - 1)) >> PAGE_SHIFT;
1647
1648	/* 3. Compute the pte we need to insert for nest_gpa -> host r_addr */
1649
1650	/* The permissions is the combination of the host and l1 guest ptes */
1651	perm |= gpte.may_read ? 0UL : _PAGE_READ;
1652	perm |= gpte.may_write ? 0UL : _PAGE_WRITE;
1653	perm |= gpte.may_execute ? 0UL : _PAGE_EXEC;
1654	/* Only set accessed/dirty (rc) bits if set in host and l1 guest ptes */
1655	perm |= (gpte.rc & _PAGE_ACCESSED) ? 0UL : _PAGE_ACCESSED;
1656	perm |= ((gpte.rc & _PAGE_DIRTY) && writing) ? 0UL : _PAGE_DIRTY;
1657	pte = __pte(pte_val(pte) & ~perm);
1658
1659	/* What size pte can we insert? */
1660	if (shift > l1_shift) {
1661		u64 mask;
1662		unsigned int actual_shift = PAGE_SHIFT;
1663		if (PMD_SHIFT < l1_shift)
1664			actual_shift = PMD_SHIFT;
1665		mask = (1UL << shift) - (1UL << actual_shift);
1666		pte = __pte(pte_val(pte) | (gpa & mask));
1667		shift = actual_shift;
1668	}
1669	level = kvmppc_radix_shift_to_level(shift);
1670	n_gpa &= ~((1UL << shift) - 1);
1671
1672	/* 4. Insert the pte into our shadow_pgtable */
1673
1674	n_rmap = kzalloc(sizeof(*n_rmap), GFP_KERNEL);
1675	if (!n_rmap)
1676		return RESUME_GUEST; /* Let the guest try again */
1677	n_rmap->rmap = (n_gpa & RMAP_NESTED_GPA_MASK) |
1678		(((unsigned long) gp->l1_lpid) << RMAP_NESTED_LPID_SHIFT);
1679	rmapp = &memslot->arch.rmap[gfn - memslot->base_gfn];
1680	ret = kvmppc_create_pte(kvm, gp->shadow_pgtable, pte, n_gpa, level,
1681				mmu_seq, gp->shadow_lpid, rmapp, &n_rmap);
1682	kfree(n_rmap);
1683	if (ret == -EAGAIN)
1684		ret = RESUME_GUEST;	/* Let the guest try again */
1685
1686	return ret;
1687
1688 inval:
1689	kvmhv_invalidate_shadow_pte(vcpu, gp, n_gpa, NULL);
1690	return RESUME_GUEST;
1691}
1692
1693long int kvmhv_nested_page_fault(struct kvm_vcpu *vcpu)
1694{
1695	struct kvm_nested_guest *gp = vcpu->arch.nested;
1696	long int ret;
1697
1698	mutex_lock(&gp->tlb_lock);
1699	ret = __kvmhv_nested_page_fault(vcpu, gp);
1700	mutex_unlock(&gp->tlb_lock);
1701	return ret;
1702}
1703
1704int kvmhv_nested_next_lpid(struct kvm *kvm, int lpid)
1705{
1706	int ret = lpid + 1;
1707
1708	spin_lock(&kvm->mmu_lock);
1709	if (!idr_get_next(&kvm->arch.kvm_nested_guest_idr, &ret))
1710		ret = -1;
1711	spin_unlock(&kvm->mmu_lock);
1712
1713	return ret;
1714}