arch/powerpc/kvm/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
  15 #include <asm/kvm_ppc.h>
  16 #include <asm/kvm_book3s.h>
  17 #include <asm/mmu.h>
  18 #include <asm/pgtable.h>
  19 #include <asm/pgalloc.h>
  20 #include <asm/pte-walk.h>
  21 #include <asm/reg.h>
  22
  23 static struct patb_entry *pseries_partition_tb;
  24
  25 static void kvmhv_update_ptbl_cache(struct kvm_nested_guest *gp);
  26 static void kvmhv_free_memslot_nest_rmap(struct kvm_memory_slot *free);
  27
  28 void kvmhv_save_hv_regs(struct kvm_vcpu *vcpu, struct hv_guest_state *hr)
  29 {
  30         struct kvmppc_vcore *vc = vcpu->arch.vcore;
  31
  32         hr->pcr = vc->pcr | PCR_MASK;
  33         hr->dpdes = vc->dpdes;
  34         hr->hfscr = vcpu->arch.hfscr;
  35         hr->tb_offset = vc->tb_offset;
  36         hr->dawr0 = vcpu->arch.dawr;
  37         hr->dawrx0 = vcpu->arch.dawrx;
  38         hr->ciabr = vcpu->arch.ciabr;
  39         hr->purr = vcpu->arch.purr;
  40         hr->spurr = vcpu->arch.spurr;
  41         hr->ic = vcpu->arch.ic;
  42         hr->vtb = vc->vtb;
  43         hr->srr0 = vcpu->arch.shregs.srr0;
  44         hr->srr1 = vcpu->arch.shregs.srr1;
  45         hr->sprg[0] = vcpu->arch.shregs.sprg0;
  46         hr->sprg[1] = vcpu->arch.shregs.sprg1;
  47         hr->sprg[2] = vcpu->arch.shregs.sprg2;
  48         hr->sprg[3] = vcpu->arch.shregs.sprg3;
  49         hr->pidr = vcpu->arch.pid;
  50         hr->cfar = vcpu->arch.cfar;
  51         hr->ppr = vcpu->arch.ppr;
  52 }
  53
  54 /* Use noinline_for_stack due to https://bugs.llvm.org/show_bug.cgi?id=49610 */
  55 static noinline_for_stack void byteswap_pt_regs(struct pt_regs *regs)
  56 {
  57         unsigned long *addr = (unsigned long *) regs;
  58
  59         for (; addr < ((unsigned long *) (regs + 1)); addr++)
  60                 *addr = swab64(*addr);
  61 }
  62
  63 static void byteswap_hv_regs(struct hv_guest_state *hr)
  64 {
  65         hr->version = swab64(hr->version);
  66         hr->lpid = swab32(hr->lpid);
  67         hr->vcpu_token = swab32(hr->vcpu_token);
  68         hr->lpcr = swab64(hr->lpcr);
  69         hr->pcr = swab64(hr->pcr) | PCR_MASK;
  70         hr->amor = swab64(hr->amor);
  71         hr->dpdes = swab64(hr->dpdes);
  72         hr->hfscr = swab64(hr->hfscr);
  73         hr->tb_offset = swab64(hr->tb_offset);
  74         hr->dawr0 = swab64(hr->dawr0);
  75         hr->dawrx0 = swab64(hr->dawrx0);
  76         hr->ciabr = swab64(hr->ciabr);
  77         hr->hdec_expiry = swab64(hr->hdec_expiry);
  78         hr->purr = swab64(hr->purr);
  79         hr->spurr = swab64(hr->spurr);
  80         hr->ic = swab64(hr->ic);
  81         hr->vtb = swab64(hr->vtb);
  82         hr->hdar = swab64(hr->hdar);
  83         hr->hdsisr = swab64(hr->hdsisr);
  84         hr->heir = swab64(hr->heir);
  85         hr->asdr = swab64(hr->asdr);
  86         hr->srr0 = swab64(hr->srr0);
  87         hr->srr1 = swab64(hr->srr1);
  88         hr->sprg[0] = swab64(hr->sprg[0]);
  89         hr->sprg[1] = swab64(hr->sprg[1]);
  90         hr->sprg[2] = swab64(hr->sprg[2]);
  91         hr->sprg[3] = swab64(hr->sprg[3]);
  92         hr->pidr = swab64(hr->pidr);
  93         hr->cfar = swab64(hr->cfar);
  94         hr->ppr = swab64(hr->ppr);
  95 }
  96
  97 static void save_hv_return_state(struct kvm_vcpu *vcpu, int trap,
  98                                  struct hv_guest_state *hr)
  99 {
 100         struct kvmppc_vcore *vc = vcpu->arch.vcore;
 101
 102         hr->dpdes = vc->dpdes;
 103         hr->hfscr = vcpu->arch.hfscr;
 104         hr->purr = vcpu->arch.purr;
 105         hr->spurr = vcpu->arch.spurr;
 106         hr->ic = vcpu->arch.ic;
 107         hr->vtb = vc->vtb;
 108         hr->srr0 = vcpu->arch.shregs.srr0;
 109         hr->srr1 = vcpu->arch.shregs.srr1;
 110         hr->sprg[0] = vcpu->arch.shregs.sprg0;
 111         hr->sprg[1] = vcpu->arch.shregs.sprg1;
 112         hr->sprg[2] = vcpu->arch.shregs.sprg2;
 113         hr->sprg[3] = vcpu->arch.shregs.sprg3;
 114         hr->pidr = vcpu->arch.pid;
 115         hr->cfar = vcpu->arch.cfar;
 116         hr->ppr = vcpu->arch.ppr;
 117         switch (trap) {
 118         case BOOK3S_INTERRUPT_H_DATA_STORAGE:
 119                 hr->hdar = vcpu->arch.fault_dar;
 120                 hr->hdsisr = vcpu->arch.fault_dsisr;
 121                 hr->asdr = vcpu->arch.fault_gpa;
 122                 break;
 123         case BOOK3S_INTERRUPT_H_INST_STORAGE:
 124                 hr->asdr = vcpu->arch.fault_gpa;
 125                 break;
 126         case BOOK3S_INTERRUPT_H_EMUL_ASSIST:
 127                 hr->heir = vcpu->arch.emul_inst;
 128                 break;
 129         }
 130 }
 131
 132 static void sanitise_hv_regs(struct kvm_vcpu *vcpu, struct hv_guest_state *hr)
 133 {
 134         /*
 135          * Don't let L1 enable features for L2 which we've disabled for L1,
 136          * but preserve the interrupt cause field.
 137          */
 138         hr->hfscr &= (HFSCR_INTR_CAUSE | vcpu->arch.hfscr);
 139
 140         /* Don't let data address watchpoint match in hypervisor state */
 141         hr->dawrx0 &= ~DAWRX_HYP;
 142
 143         /* Don't let completed instruction address breakpt match in HV state */
 144         if ((hr->ciabr & CIABR_PRIV) == CIABR_PRIV_HYPER)
 145                 hr->ciabr &= ~CIABR_PRIV;
 146 }
 147
 148 static void restore_hv_regs(struct kvm_vcpu *vcpu, struct hv_guest_state *hr)
 149 {
 150         struct kvmppc_vcore *vc = vcpu->arch.vcore;
 151
 152         vc->pcr = hr->pcr | PCR_MASK;
 153         vc->dpdes = hr->dpdes;
 154         vcpu->arch.hfscr = hr->hfscr;
 155         vcpu->arch.dawr = hr->dawr0;
 156         vcpu->arch.dawrx = hr->dawrx0;
 157         vcpu->arch.ciabr = hr->ciabr;
 158         vcpu->arch.purr = hr->purr;
 159         vcpu->arch.spurr = hr->spurr;
 160         vcpu->arch.ic = hr->ic;
 161         vc->vtb = hr->vtb;
 162         vcpu->arch.shregs.srr0 = hr->srr0;
 163         vcpu->arch.shregs.srr1 = hr->srr1;
 164         vcpu->arch.shregs.sprg0 = hr->sprg[0];
 165         vcpu->arch.shregs.sprg1 = hr->sprg[1];
 166         vcpu->arch.shregs.sprg2 = hr->sprg[2];
 167         vcpu->arch.shregs.sprg3 = hr->sprg[3];
 168         vcpu->arch.pid = hr->pidr;
 169         vcpu->arch.cfar = hr->cfar;
 170         vcpu->arch.ppr = hr->ppr;
 171 }
 172
 173 void kvmhv_restore_hv_return_state(struct kvm_vcpu *vcpu,
 174                                    struct hv_guest_state *hr)
 175 {
 176         struct kvmppc_vcore *vc = vcpu->arch.vcore;
 177
 178         vc->dpdes = hr->dpdes;
 179         vcpu->arch.hfscr = hr->hfscr;
 180         vcpu->arch.purr = hr->purr;
 181         vcpu->arch.spurr = hr->spurr;
 182         vcpu->arch.ic = hr->ic;
 183         vc->vtb = hr->vtb;
 184         vcpu->arch.fault_dar = hr->hdar;
 185         vcpu->arch.fault_dsisr = hr->hdsisr;
 186         vcpu->arch.fault_gpa = hr->asdr;
 187         vcpu->arch.emul_inst = hr->heir;
 188         vcpu->arch.shregs.srr0 = hr->srr0;
 189         vcpu->arch.shregs.srr1 = hr->srr1;
 190         vcpu->arch.shregs.sprg0 = hr->sprg[0];
 191         vcpu->arch.shregs.sprg1 = hr->sprg[1];
 192         vcpu->arch.shregs.sprg2 = hr->sprg[2];
 193         vcpu->arch.shregs.sprg3 = hr->sprg[3];
 194         vcpu->arch.pid = hr->pidr;
 195         vcpu->arch.cfar = hr->cfar;
 196         vcpu->arch.ppr = hr->ppr;
 197 }
 198
 199 static void kvmhv_nested_mmio_needed(struct kvm_vcpu *vcpu, u64 regs_ptr)
 200 {
 201         /* No need to reflect the page fault to L1, we've handled it */
 202         vcpu->arch.trap = 0;
 203
 204         /*
 205          * Since the L2 gprs have already been written back into L1 memory when
 206          * we complete the mmio, store the L1 memory location of the L2 gpr
 207          * being loaded into by the mmio so that the loaded value can be
 208          * written there in kvmppc_complete_mmio_load()
 209          */
 210         if (((vcpu->arch.io_gpr & KVM_MMIO_REG_EXT_MASK) == KVM_MMIO_REG_GPR)
 211             && (vcpu->mmio_is_write == 0)) {
 212                 vcpu->arch.nested_io_gpr = (gpa_t) regs_ptr +
 213                                            offsetof(struct pt_regs,
 214                                                     gpr[vcpu->arch.io_gpr]);
 215                 vcpu->arch.io_gpr = KVM_MMIO_REG_NESTED_GPR;
 216         }
 217 }
 218
 219 long kvmhv_enter_nested_guest(struct kvm_vcpu *vcpu)
 220 {
 221         long int err, r;
 222         struct kvm_nested_guest *l2;
 223         struct pt_regs l2_regs, saved_l1_regs;
 224         struct hv_guest_state l2_hv, saved_l1_hv;
 225         struct kvmppc_vcore *vc = vcpu->arch.vcore;
 226         u64 hv_ptr, regs_ptr;
 227         u64 hdec_exp;
 228         s64 delta_purr, delta_spurr, delta_ic, delta_vtb;
 229         u64 mask;
 230         unsigned long lpcr;
 231
 232         if (vcpu->kvm->arch.l1_ptcr == 0)
 233                 return H_NOT_AVAILABLE;
 234
 235         if (MSR_TM_TRANSACTIONAL(vcpu->arch.shregs.msr))
 236                 return H_BAD_MODE;
 237
 238         /* copy parameters in */
 239         hv_ptr = kvmppc_get_gpr(vcpu, 4);
 240         err = kvm_vcpu_read_guest(vcpu, hv_ptr, &l2_hv,
 241                                   sizeof(struct hv_guest_state));
 242         if (err)
 243                 return H_PARAMETER;
 244         if (kvmppc_need_byteswap(vcpu))
 245                 byteswap_hv_regs(&l2_hv);
 246         if (l2_hv.version != HV_GUEST_STATE_VERSION)
 247                 return H_P2;
 248
 249         regs_ptr = kvmppc_get_gpr(vcpu, 5);
 250         err = kvm_vcpu_read_guest(vcpu, regs_ptr, &l2_regs,
 251                                   sizeof(struct pt_regs));
 252         if (err)
 253                 return H_PARAMETER;
 254         if (kvmppc_need_byteswap(vcpu))
 255                 byteswap_pt_regs(&l2_regs);
 256         if (l2_hv.vcpu_token >= NR_CPUS)
 257                 return H_PARAMETER;
 258
 259         /*
 260          * L1 must have set up a suspended state to enter the L2 in a
 261          * transactional state, and only in that case. These have to be
 262          * filtered out here to prevent causing a TM Bad Thing in the
 263          * host HRFID. We could synthesize a TM Bad Thing back to the L1
 264          * here but there doesn't seem like much point.
 265          */
 266         if (MSR_TM_SUSPENDED(vcpu->arch.shregs.msr)) {
 267                 if (!MSR_TM_ACTIVE(l2_regs.msr))
 268                         return H_BAD_MODE;
 269         } else {
 270                 if (l2_regs.msr & MSR_TS_MASK)
 271                         return H_BAD_MODE;
 272                 if (WARN_ON_ONCE(vcpu->arch.shregs.msr & MSR_TS_MASK))
 273                         return H_BAD_MODE;
 274         }
 275
 276         /* translate lpid */
 277         l2 = kvmhv_get_nested(vcpu->kvm, l2_hv.lpid, true);
 278         if (!l2)
 279                 return H_PARAMETER;
 280         if (!l2->l1_gr_to_hr) {
 281                 mutex_lock(&l2->tlb_lock);
 282                 kvmhv_update_ptbl_cache(l2);
 283                 mutex_unlock(&l2->tlb_lock);
 284         }
 285
 286         /* save l1 values of things */
 287         vcpu->arch.regs.msr = vcpu->arch.shregs.msr;
 288         saved_l1_regs = vcpu->arch.regs;
 289         kvmhv_save_hv_regs(vcpu, &saved_l1_hv);
 290
 291         /* convert TB values/offsets to host (L0) values */
 292         hdec_exp = l2_hv.hdec_expiry - vc->tb_offset;
 293         vc->tb_offset += l2_hv.tb_offset;
 294
 295         /* set L1 state to L2 state */
 296         vcpu->arch.nested = l2;
 297         vcpu->arch.nested_vcpu_id = l2_hv.vcpu_token;
 298         vcpu->arch.regs = l2_regs;
 299         vcpu->arch.shregs.msr = vcpu->arch.regs.msr;
 300         mask = LPCR_DPFD | LPCR_ILE | LPCR_TC | LPCR_AIL | LPCR_LD |
 301                 LPCR_LPES | LPCR_MER;
 302         lpcr = (vc->lpcr & ~mask) | (l2_hv.lpcr & mask);
 303         sanitise_hv_regs(vcpu, &l2_hv);
 304         restore_hv_regs(vcpu, &l2_hv);
 305
 306         vcpu->arch.ret = RESUME_GUEST;
 307         vcpu->arch.trap = 0;
 308         do {
 309                 if (mftb() >= hdec_exp) {
 310                         vcpu->arch.trap = BOOK3S_INTERRUPT_HV_DECREMENTER;
 311                         r = RESUME_HOST;
 312                         break;
 313                 }
 314                 r = kvmhv_run_single_vcpu(vcpu->arch.kvm_run, vcpu, hdec_exp,
 315                                           lpcr);
 316         } while (is_kvmppc_resume_guest(r));
 317
 318         /* save L2 state for return */
 319         l2_regs = vcpu->arch.regs;
 320         l2_regs.msr = vcpu->arch.shregs.msr;
 321         delta_purr = vcpu->arch.purr - l2_hv.purr;
 322         delta_spurr = vcpu->arch.spurr - l2_hv.spurr;
 323         delta_ic = vcpu->arch.ic - l2_hv.ic;
 324         delta_vtb = vc->vtb - l2_hv.vtb;
 325         save_hv_return_state(vcpu, vcpu->arch.trap, &l2_hv);
 326
 327         /* restore L1 state */
 328         vcpu->arch.nested = NULL;
 329         vcpu->arch.regs = saved_l1_regs;
 330         vcpu->arch.shregs.msr = saved_l1_regs.msr & ~MSR_TS_MASK;
 331         /* set L1 MSR TS field according to L2 transaction state */
 332         if (l2_regs.msr & MSR_TS_MASK)
 333                 vcpu->arch.shregs.msr |= MSR_TS_S;
 334         vc->tb_offset = saved_l1_hv.tb_offset;
 335         restore_hv_regs(vcpu, &saved_l1_hv);
 336         vcpu->arch.purr += delta_purr;
 337         vcpu->arch.spurr += delta_spurr;
 338         vcpu->arch.ic += delta_ic;
 339         vc->vtb += delta_vtb;
 340
 341         kvmhv_put_nested(l2);
 342
 343         /* copy l2_hv_state and regs back to guest */
 344         if (kvmppc_need_byteswap(vcpu)) {
 345                 byteswap_hv_regs(&l2_hv);
 346                 byteswap_pt_regs(&l2_regs);
 347         }
 348         err = kvm_vcpu_write_guest(vcpu, hv_ptr, &l2_hv,
 349                                    sizeof(struct hv_guest_state));
 350         if (err)
 351                 return H_AUTHORITY;
 352         err = kvm_vcpu_write_guest(vcpu, regs_ptr, &l2_regs,
 353                                    sizeof(struct pt_regs));
 354         if (err)
 355                 return H_AUTHORITY;
 356
 357         if (r == -EINTR)
 358                 return H_INTERRUPT;
 359
 360         if (vcpu->mmio_needed) {
 361                 kvmhv_nested_mmio_needed(vcpu, regs_ptr);
 362                 return H_TOO_HARD;
 363         }
 364
 365         return vcpu->arch.trap;
 366 }
 367
 368 long kvmhv_nested_init(void)
 369 {
 370         long int ptb_order;
 371         unsigned long ptcr;
 372         long rc;
 373
 374         if (!kvmhv_on_pseries())
 375                 return 0;
 376         if (!radix_enabled())
 377                 return -ENODEV;
 378
 379         /* find log base 2 of KVMPPC_NR_LPIDS, rounding up */
 380         ptb_order = __ilog2(KVMPPC_NR_LPIDS - 1) + 1;
 381         if (ptb_order < 8)
 382                 ptb_order = 8;
 383         pseries_partition_tb = kmalloc(sizeof(struct patb_entry) << ptb_order,
 384                                        GFP_KERNEL);
 385         if (!pseries_partition_tb) {
 386                 pr_err("kvm-hv: failed to allocated nested partition table\n");
 387                 return -ENOMEM;
 388         }
 389
 390         ptcr = __pa(pseries_partition_tb) | (ptb_order - 8);
 391         rc = plpar_hcall_norets(H_SET_PARTITION_TABLE, ptcr);
 392         if (rc != H_SUCCESS) {
 393                 pr_err("kvm-hv: Parent hypervisor does not support nesting (rc=%ld)\n",
 394                        rc);
 395                 kfree(pseries_partition_tb);
 396                 pseries_partition_tb = NULL;
 397                 return -ENODEV;
 398         }
 399
 400         return 0;
 401 }
 402
 403 void kvmhv_nested_exit(void)
 404 {
 405         /*
 406          * N.B. the kvmhv_on_pseries() test is there because it enables
 407          * the compiler to remove the call to plpar_hcall_norets()
 408          * when CONFIG_PPC_PSERIES=n.
 409          */
 410         if (kvmhv_on_pseries() && pseries_partition_tb) {
 411                 plpar_hcall_norets(H_SET_PARTITION_TABLE, 0);
 412                 kfree(pseries_partition_tb);
 413                 pseries_partition_tb = NULL;
 414         }
 415 }
 416
 417 static void kvmhv_flush_lpid(unsigned int lpid)
 418 {
 419         long rc;
 420
 421         if (!kvmhv_on_pseries()) {
 422                 radix__flush_all_lpid(lpid);
 423                 return;
 424         }
 425
 426         rc = plpar_hcall_norets(H_TLB_INVALIDATE, H_TLBIE_P1_ENC(2, 0, 1),
 427                                 lpid, TLBIEL_INVAL_SET_LPID);
 428         if (rc)
 429                 pr_err("KVM: TLB LPID invalidation hcall failed, rc=%ld\n", rc);
 430 }
 431
 432 void kvmhv_set_ptbl_entry(unsigned int lpid, u64 dw0, u64 dw1)
 433 {
 434         if (!kvmhv_on_pseries()) {
 435                 mmu_partition_table_set_entry(lpid, dw0, dw1, true);
 436                 return;
 437         }
 438
 439         pseries_partition_tb[lpid].patb0 = cpu_to_be64(dw0);
 440         pseries_partition_tb[lpid].patb1 = cpu_to_be64(dw1);
 441         /* L0 will do the necessary barriers */
 442         kvmhv_flush_lpid(lpid);
 443 }
 444
 445 static void kvmhv_set_nested_ptbl(struct kvm_nested_guest *gp)
 446 {
 447         unsigned long dw0;
 448
 449         dw0 = PATB_HR | radix__get_tree_size() |
 450                 __pa(gp->shadow_pgtable) | RADIX_PGD_INDEX_SIZE;
 451         kvmhv_set_ptbl_entry(gp->shadow_lpid, dw0, gp->process_table);
 452 }
 453
 454 void kvmhv_vm_nested_init(struct kvm *kvm)
 455 {
 456         kvm->arch.max_nested_lpid = -1;
 457 }
 458
 459 /*
 460  * Handle the H_SET_PARTITION_TABLE hcall.
 461  * r4 = guest real address of partition table + log_2(size) - 12
 462  * (formatted as for the PTCR).
 463  */
 464 long kvmhv_set_partition_table(struct kvm_vcpu *vcpu)
 465 {
 466         struct kvm *kvm = vcpu->kvm;
 467         unsigned long ptcr = kvmppc_get_gpr(vcpu, 4);
 468         int srcu_idx;
 469         long ret = H_SUCCESS;
 470
 471         srcu_idx = srcu_read_lock(&kvm->srcu);
 472         /*
 473          * Limit the partition table to 4096 entries (because that's what
 474          * hardware supports), and check the base address.
 475          */
 476         if ((ptcr & PRTS_MASK) > 12 - 8 ||
 477             !kvm_is_visible_gfn(vcpu->kvm, (ptcr & PRTB_MASK) >> PAGE_SHIFT))
 478                 ret = H_PARAMETER;
 479         srcu_read_unlock(&kvm->srcu, srcu_idx);
 480         if (ret == H_SUCCESS)
 481                 kvm->arch.l1_ptcr = ptcr;
 482         return ret;
 483 }
 484
 485 /*
 486  * Handle the H_COPY_TOFROM_GUEST hcall.
 487  * r4 = L1 lpid of nested guest
 488  * r5 = pid
 489  * r6 = eaddr to access
 490  * r7 = to buffer (L1 gpa)
 491  * r8 = from buffer (L1 gpa)
 492  * r9 = n bytes to copy
 493  */
 494 long kvmhv_copy_tofrom_guest_nested(struct kvm_vcpu *vcpu)
 495 {
 496         struct kvm_nested_guest *gp;
 497         int l1_lpid = kvmppc_get_gpr(vcpu, 4);
 498         int pid = kvmppc_get_gpr(vcpu, 5);
 499         gva_t eaddr = kvmppc_get_gpr(vcpu, 6);
 500         gpa_t gp_to = (gpa_t) kvmppc_get_gpr(vcpu, 7);
 501         gpa_t gp_from = (gpa_t) kvmppc_get_gpr(vcpu, 8);
 502         void *buf;
 503         unsigned long n = kvmppc_get_gpr(vcpu, 9);
 504         bool is_load = !!gp_to;
 505         long rc;
 506
 507         if (gp_to && gp_from) /* One must be NULL to determine the direction */
 508                 return H_PARAMETER;
 509
 510         if (eaddr & (0xFFFUL << 52))
 511                 return H_PARAMETER;
 512
 513         buf = kzalloc(n, GFP_KERNEL | __GFP_NOWARN);
 514         if (!buf)
 515                 return H_NO_MEM;
 516
 517         gp = kvmhv_get_nested(vcpu->kvm, l1_lpid, false);
 518         if (!gp) {
 519                 rc = H_PARAMETER;
 520                 goto out_free;
 521         }
 522
 523         mutex_lock(&gp->tlb_lock);
 524
 525         if (is_load) {
 526                 /* Load from the nested guest into our buffer */
 527                 rc = __kvmhv_copy_tofrom_guest_radix(gp->shadow_lpid, pid,
 528                                                      eaddr, buf, NULL, n);
 529                 if (rc)
 530                         goto not_found;
 531
 532                 /* Write what was loaded into our buffer back to the L1 guest */
 533                 rc = kvm_vcpu_write_guest(vcpu, gp_to, buf, n);
 534                 if (rc)
 535                         goto not_found;
 536         } else {
 537                 /* Load the data to be stored from the L1 guest into our buf */
 538                 rc = kvm_vcpu_read_guest(vcpu, gp_from, buf, n);
 539                 if (rc)
 540                         goto not_found;
 541
 542                 /* Store from our buffer into the nested guest */
 543                 rc = __kvmhv_copy_tofrom_guest_radix(gp->shadow_lpid, pid,
 544                                                      eaddr, NULL, buf, n);
 545                 if (rc)
 546                         goto not_found;
 547         }
 548
 549 out_unlock:
 550         mutex_unlock(&gp->tlb_lock);
 551         kvmhv_put_nested(gp);
 552 out_free:
 553         kfree(buf);
 554         return rc;
 555 not_found:
 556         rc = H_NOT_FOUND;
 557         goto out_unlock;
 558 }
 559
 560 /*
 561  * Reload the partition table entry for a guest.
 562  * Caller must hold gp->tlb_lock.
 563  */
 564 static void kvmhv_update_ptbl_cache(struct kvm_nested_guest *gp)
 565 {
 566         int ret;
 567         struct patb_entry ptbl_entry;
 568         unsigned long ptbl_addr;
 569         struct kvm *kvm = gp->l1_host;
 570
 571         ret = -EFAULT;
 572         ptbl_addr = (kvm->arch.l1_ptcr & PRTB_MASK) + (gp->l1_lpid << 4);
 573         if (gp->l1_lpid < (1ul << ((kvm->arch.l1_ptcr & PRTS_MASK) + 8)))
 574                 ret = kvm_read_guest(kvm, ptbl_addr,
 575                                      &ptbl_entry, sizeof(ptbl_entry));
 576         if (ret) {
 577                 gp->l1_gr_to_hr = 0;
 578                 gp->process_table = 0;
 579         } else {
 580                 gp->l1_gr_to_hr = be64_to_cpu(ptbl_entry.patb0);
 581                 gp->process_table = be64_to_cpu(ptbl_entry.patb1);
 582         }
 583         kvmhv_set_nested_ptbl(gp);
 584 }
 585
 586 struct kvm_nested_guest *kvmhv_alloc_nested(struct kvm *kvm, unsigned int lpid)
 587 {
 588         struct kvm_nested_guest *gp;
 589         long shadow_lpid;
 590
 591         gp = kzalloc(sizeof(*gp), GFP_KERNEL);
 592         if (!gp)
 593                 return NULL;
 594         gp->l1_host = kvm;
 595         gp->l1_lpid = lpid;
 596         mutex_init(&gp->tlb_lock);
 597         gp->shadow_pgtable = pgd_alloc(kvm->mm);
 598         if (!gp->shadow_pgtable)
 599                 goto out_free;
 600         shadow_lpid = kvmppc_alloc_lpid();
 601         if (shadow_lpid < 0)
 602                 goto out_free2;
 603         gp->shadow_lpid = shadow_lpid;
 604         gp->radix = 1;
 605
 606         memset(gp->prev_cpu, -1, sizeof(gp->prev_cpu));
 607
 608         return gp;
 609
 610  out_free2:
 611         pgd_free(kvm->mm, gp->shadow_pgtable);
 612  out_free:
 613         kfree(gp);
 614         return NULL;
 615 }
 616
 617 /*
 618  * Free up any resources allocated for a nested guest.
 619  */
 620 static void kvmhv_release_nested(struct kvm_nested_guest *gp)
 621 {
 622         struct kvm *kvm = gp->l1_host;
 623
 624         if (gp->shadow_pgtable) {
 625                 /*
 626                  * No vcpu is using this struct and no call to
 627                  * kvmhv_get_nested can find this struct,
 628                  * so we don't need to hold kvm->mmu_lock.
 629                  */
 630                 kvmppc_free_pgtable_radix(kvm, gp->shadow_pgtable,
 631                                           gp->shadow_lpid);
 632                 pgd_free(kvm->mm, gp->shadow_pgtable);
 633         }
 634         kvmhv_set_ptbl_entry(gp->shadow_lpid, 0, 0);
 635         kvmppc_free_lpid(gp->shadow_lpid);
 636         kfree(gp);
 637 }
 638
 639 static void kvmhv_remove_nested(struct kvm_nested_guest *gp)
 640 {
 641         struct kvm *kvm = gp->l1_host;
 642         int lpid = gp->l1_lpid;
 643         long ref;
 644
 645         spin_lock(&kvm->mmu_lock);
 646         if (gp == kvm->arch.nested_guests[lpid]) {
 647                 kvm->arch.nested_guests[lpid] = NULL;
 648                 if (lpid == kvm->arch.max_nested_lpid) {
 649                         while (--lpid >= 0 && !kvm->arch.nested_guests[lpid])
 650                                 ;
 651                         kvm->arch.max_nested_lpid = lpid;
 652                 }
 653                 --gp->refcnt;
 654         }
 655         ref = gp->refcnt;
 656         spin_unlock(&kvm->mmu_lock);
 657         if (ref == 0)
 658                 kvmhv_release_nested(gp);
 659 }
 660
 661 /*
 662  * Free up all nested resources allocated for this guest.
 663  * This is called with no vcpus of the guest running, when
 664  * switching the guest to HPT mode or when destroying the
 665  * guest.
 666  */
 667 void kvmhv_release_all_nested(struct kvm *kvm)
 668 {
 669         int i;
 670         struct kvm_nested_guest *gp;
 671         struct kvm_nested_guest *freelist = NULL;
 672         struct kvm_memory_slot *memslot;
 673         int srcu_idx;
 674
 675         spin_lock(&kvm->mmu_lock);
 676         for (i = 0; i <= kvm->arch.max_nested_lpid; i++) {
 677                 gp = kvm->arch.nested_guests[i];
 678                 if (!gp)
 679                         continue;
 680                 kvm->arch.nested_guests[i] = NULL;
 681                 if (--gp->refcnt == 0) {
 682                         gp->next = freelist;
 683                         freelist = gp;
 684                 }
 685         }
 686         kvm->arch.max_nested_lpid = -1;
 687         spin_unlock(&kvm->mmu_lock);
 688         while ((gp = freelist) != NULL) {
 689                 freelist = gp->next;
 690                 kvmhv_release_nested(gp);
 691         }
 692
 693         srcu_idx = srcu_read_lock(&kvm->srcu);
 694         kvm_for_each_memslot(memslot, kvm_memslots(kvm))
 695                 kvmhv_free_memslot_nest_rmap(memslot);
 696         srcu_read_unlock(&kvm->srcu, srcu_idx);
 697 }
 698
 699 /* caller must hold gp->tlb_lock */
 700 static void kvmhv_flush_nested(struct kvm_nested_guest *gp)
 701 {
 702         struct kvm *kvm = gp->l1_host;
 703
 704         spin_lock(&kvm->mmu_lock);
 705         kvmppc_free_pgtable_radix(kvm, gp->shadow_pgtable, gp->shadow_lpid);
 706         spin_unlock(&kvm->mmu_lock);
 707         kvmhv_flush_lpid(gp->shadow_lpid);
 708         kvmhv_update_ptbl_cache(gp);
 709         if (gp->l1_gr_to_hr == 0)
 710                 kvmhv_remove_nested(gp);
 711 }
 712
 713 struct kvm_nested_guest *kvmhv_get_nested(struct kvm *kvm, int l1_lpid,
 714                                           bool create)
 715 {
 716         struct kvm_nested_guest *gp, *newgp;
 717
 718         if (l1_lpid >= KVM_MAX_NESTED_GUESTS ||
 719             l1_lpid >= (1ul << ((kvm->arch.l1_ptcr & PRTS_MASK) + 12 - 4)))
 720                 return NULL;
 721
 722         spin_lock(&kvm->mmu_lock);
 723         gp = kvm->arch.nested_guests[l1_lpid];
 724         if (gp)
 725                 ++gp->refcnt;
 726         spin_unlock(&kvm->mmu_lock);
 727
 728         if (gp || !create)
 729                 return gp;
 730
 731         newgp = kvmhv_alloc_nested(kvm, l1_lpid);
 732         if (!newgp)
 733                 return NULL;
 734         spin_lock(&kvm->mmu_lock);
 735         if (kvm->arch.nested_guests[l1_lpid]) {
 736                 /* someone else beat us to it */
 737                 gp = kvm->arch.nested_guests[l1_lpid];
 738         } else {
 739                 kvm->arch.nested_guests[l1_lpid] = newgp;
 740                 ++newgp->refcnt;
 741                 gp = newgp;
 742                 newgp = NULL;
 743                 if (l1_lpid > kvm->arch.max_nested_lpid)
 744                         kvm->arch.max_nested_lpid = l1_lpid;
 745         }
 746         ++gp->refcnt;
 747         spin_unlock(&kvm->mmu_lock);
 748
 749         if (newgp)
 750                 kvmhv_release_nested(newgp);
 751
 752         return gp;
 753 }
 754
 755 void kvmhv_put_nested(struct kvm_nested_guest *gp)
 756 {
 757         struct kvm *kvm = gp->l1_host;
 758         long ref;
 759
 760         spin_lock(&kvm->mmu_lock);
 761         ref = --gp->refcnt;
 762         spin_unlock(&kvm->mmu_lock);
 763         if (ref == 0)
 764                 kvmhv_release_nested(gp);
 765 }
 766
 767 static struct kvm_nested_guest *kvmhv_find_nested(struct kvm *kvm, int lpid)
 768 {
 769         if (lpid > kvm->arch.max_nested_lpid)
 770                 return NULL;
 771         return kvm->arch.nested_guests[lpid];
 772 }
 773
 774 static inline bool kvmhv_n_rmap_is_equal(u64 rmap_1, u64 rmap_2)
 775 {
 776         return !((rmap_1 ^ rmap_2) & (RMAP_NESTED_LPID_MASK |
 777                                        RMAP_NESTED_GPA_MASK));
 778 }
 779
 780 void kvmhv_insert_nest_rmap(struct kvm *kvm, unsigned long *rmapp,
 781                             struct rmap_nested **n_rmap)
 782 {
 783         struct llist_node *entry = ((struct llist_head *) rmapp)->first;
 784         struct rmap_nested *cursor;
 785         u64 rmap, new_rmap = (*n_rmap)->rmap;
 786
 787         /* Are there any existing entries? */
 788         if (!(*rmapp)) {
 789                 /* No -> use the rmap as a single entry */
 790                 *rmapp = new_rmap | RMAP_NESTED_IS_SINGLE_ENTRY;
 791                 return;
 792         }
 793
 794         /* Do any entries match what we're trying to insert? */
 795         for_each_nest_rmap_safe(cursor, entry, &rmap) {
 796                 if (kvmhv_n_rmap_is_equal(rmap, new_rmap))
 797                         return;
 798         }
 799
 800         /* Do we need to create a list or just add the new entry? */
 801         rmap = *rmapp;
 802         if (rmap & RMAP_NESTED_IS_SINGLE_ENTRY) /* Not previously a list */
 803                 *rmapp = 0UL;
 804         llist_add(&((*n_rmap)->list), (struct llist_head *) rmapp);
 805         if (rmap & RMAP_NESTED_IS_SINGLE_ENTRY) /* Not previously a list */
 806                 (*n_rmap)->list.next = (struct llist_node *) rmap;
 807
 808         /* Set NULL so not freed by caller */
 809         *n_rmap = NULL;
 810 }
 811
 812 static void kvmhv_update_nest_rmap_rc(struct kvm *kvm, u64 n_rmap,
 813                                       unsigned long clr, unsigned long set,
 814                                       unsigned long hpa, unsigned long mask)
 815 {
 816         struct kvm_nested_guest *gp;
 817         unsigned long gpa;
 818         unsigned int shift, lpid;
 819         pte_t *ptep;
 820
 821         gpa = n_rmap & RMAP_NESTED_GPA_MASK;
 822         lpid = (n_rmap & RMAP_NESTED_LPID_MASK) >> RMAP_NESTED_LPID_SHIFT;
 823         gp = kvmhv_find_nested(kvm, lpid);
 824         if (!gp)
 825                 return;
 826
 827         /* Find the pte */
 828         ptep = __find_linux_pte(gp->shadow_pgtable, gpa, NULL, &shift);
 829         /*
 830          * If the pte is present and the pfn is still the same, update the pte.
 831          * If the pfn has changed then this is a stale rmap entry, the nested
 832          * gpa actually points somewhere else now, and there is nothing to do.
 833          * XXX A future optimisation would be to remove the rmap entry here.
 834          */
 835         if (ptep && pte_present(*ptep) && ((pte_val(*ptep) & mask) == hpa)) {
 836                 __radix_pte_update(ptep, clr, set);
 837                 kvmppc_radix_tlbie_page(kvm, gpa, shift, lpid);
 838         }
 839 }
 840
 841 /*
 842  * For a given list of rmap entries, update the rc bits in all ptes in shadow
 843  * page tables for nested guests which are referenced by the rmap list.
 844  */
 845 void kvmhv_update_nest_rmap_rc_list(struct kvm *kvm, unsigned long *rmapp,
 846                                     unsigned long clr, unsigned long set,
 847                                     unsigned long hpa, unsigned long nbytes)
 848 {
 849         struct llist_node *entry = ((struct llist_head *) rmapp)->first;
 850         struct rmap_nested *cursor;
 851         unsigned long rmap, mask;
 852
 853         if ((clr | set) & ~(_PAGE_DIRTY | _PAGE_ACCESSED))
 854                 return;
 855
 856         mask = PTE_RPN_MASK & ~(nbytes - 1);
 857         hpa &= mask;
 858
 859         for_each_nest_rmap_safe(cursor, entry, &rmap)
 860                 kvmhv_update_nest_rmap_rc(kvm, rmap, clr, set, hpa, mask);
 861 }
 862
 863 static void kvmhv_remove_nest_rmap(struct kvm *kvm, u64 n_rmap,
 864                                    unsigned long hpa, unsigned long mask)
 865 {
 866         struct kvm_nested_guest *gp;
 867         unsigned long gpa;
 868         unsigned int shift, lpid;
 869         pte_t *ptep;
 870
 871         gpa = n_rmap & RMAP_NESTED_GPA_MASK;
 872         lpid = (n_rmap & RMAP_NESTED_LPID_MASK) >> RMAP_NESTED_LPID_SHIFT;
 873         gp = kvmhv_find_nested(kvm, lpid);
 874         if (!gp)
 875                 return;
 876
 877         /* Find and invalidate the pte */
 878         ptep = __find_linux_pte(gp->shadow_pgtable, gpa, NULL, &shift);
 879         /* Don't spuriously invalidate ptes if the pfn has changed */
 880         if (ptep && pte_present(*ptep) && ((pte_val(*ptep) & mask) == hpa))
 881                 kvmppc_unmap_pte(kvm, ptep, gpa, shift, NULL, gp->shadow_lpid);
 882 }
 883
 884 static void kvmhv_remove_nest_rmap_list(struct kvm *kvm, unsigned long *rmapp,
 885                                         unsigned long hpa, unsigned long mask)
 886 {
 887         struct llist_node *entry = llist_del_all((struct llist_head *) rmapp);
 888         struct rmap_nested *cursor;
 889         unsigned long rmap;
 890
 891         for_each_nest_rmap_safe(cursor, entry, &rmap) {
 892                 kvmhv_remove_nest_rmap(kvm, rmap, hpa, mask);
 893                 kfree(cursor);
 894         }
 895 }
 896
 897 /* called with kvm->mmu_lock held */
 898 void kvmhv_remove_nest_rmap_range(struct kvm *kvm,
 899                                   const struct kvm_memory_slot *memslot,
 900                                   unsigned long gpa, unsigned long hpa,
 901                                   unsigned long nbytes)
 902 {
 903         unsigned long gfn, end_gfn;
 904         unsigned long addr_mask;
 905
 906         if (!memslot)
 907                 return;
 908         gfn = (gpa >> PAGE_SHIFT) - memslot->base_gfn;
 909         end_gfn = gfn + (nbytes >> PAGE_SHIFT);
 910
 911         addr_mask = PTE_RPN_MASK & ~(nbytes - 1);
 912         hpa &= addr_mask;
 913
 914         for (; gfn < end_gfn; gfn++) {
 915                 unsigned long *rmap = &memslot->arch.rmap[gfn];
 916                 kvmhv_remove_nest_rmap_list(kvm, rmap, hpa, addr_mask);
 917         }
 918 }
 919
 920 static void kvmhv_free_memslot_nest_rmap(struct kvm_memory_slot *free)
 921 {
 922         unsigned long page;
 923
 924         for (page = 0; page < free->npages; page++) {
 925                 unsigned long rmap, *rmapp = &free->arch.rmap[page];
 926                 struct rmap_nested *cursor;
 927                 struct llist_node *entry;
 928
 929                 entry = llist_del_all((struct llist_head *) rmapp);
 930                 for_each_nest_rmap_safe(cursor, entry, &rmap)
 931                         kfree(cursor);
 932         }
 933 }
 934
 935 static bool kvmhv_invalidate_shadow_pte(struct kvm_vcpu *vcpu,
 936                                         struct kvm_nested_guest *gp,
 937                                         long gpa, int *shift_ret)
 938 {
 939         struct kvm *kvm = vcpu->kvm;
 940         bool ret = false;
 941         pte_t *ptep;
 942         int shift;
 943
 944         spin_lock(&kvm->mmu_lock);
 945         ptep = __find_linux_pte(gp->shadow_pgtable, gpa, NULL, &shift);
 946         if (!shift)
 947                 shift = PAGE_SHIFT;
 948         if (ptep && pte_present(*ptep)) {
 949                 kvmppc_unmap_pte(kvm, ptep, gpa, shift, NULL, gp->shadow_lpid);
 950                 ret = true;
 951         }
 952         spin_unlock(&kvm->mmu_lock);
 953
 954         if (shift_ret)
 955                 *shift_ret = shift;
 956         return ret;
 957 }
 958
 959 static inline int get_ric(unsigned int instr)
 960 {
 961         return (instr >> 18) & 0x3;
 962 }
 963
 964 static inline int get_prs(unsigned int instr)
 965 {
 966         return (instr >> 17) & 0x1;
 967 }
 968
 969 static inline int get_r(unsigned int instr)
 970 {
 971         return (instr >> 16) & 0x1;
 972 }
 973
 974 static inline int get_lpid(unsigned long r_val)
 975 {
 976         return r_val & 0xffffffff;
 977 }
 978
 979 static inline int get_is(unsigned long r_val)
 980 {
 981         return (r_val >> 10) & 0x3;
 982 }
 983
 984 static inline int get_ap(unsigned long r_val)
 985 {
 986         return (r_val >> 5) & 0x7;
 987 }
 988
 989 static inline long get_epn(unsigned long r_val)
 990 {
 991         return r_val >> 12;
 992 }
 993
 994 static int kvmhv_emulate_tlbie_tlb_addr(struct kvm_vcpu *vcpu, int lpid,
 995                                         int ap, long epn)
 996 {
 997         struct kvm *kvm = vcpu->kvm;
 998         struct kvm_nested_guest *gp;
 999         long npages;
1000         int shift, shadow_shift;
1001         unsigned long addr;
1002
1003         shift = ap_to_shift(ap);
1004         addr = epn << 12;
1005         if (shift < 0)
1006                 /* Invalid ap encoding */
1007                 return -EINVAL;
1008
1009         addr &= ~((1UL << shift) - 1);
1010         npages = 1UL << (shift - PAGE_SHIFT);
1011
1012         gp = kvmhv_get_nested(kvm, lpid, false);
1013         if (!gp) /* No such guest -> nothing to do */
1014                 return 0;
1015         mutex_lock(&gp->tlb_lock);
1016
1017         /* There may be more than one host page backing this single guest pte */
1018         do {
1019                 kvmhv_invalidate_shadow_pte(vcpu, gp, addr, &shadow_shift);
1020
1021                 npages -= 1UL << (shadow_shift - PAGE_SHIFT);
1022                 addr += 1UL << shadow_shift;
1023         } while (npages > 0);
1024
1025         mutex_unlock(&gp->tlb_lock);
1026         kvmhv_put_nested(gp);
1027         return 0;
1028 }
1029
1030 static void kvmhv_emulate_tlbie_lpid(struct kvm_vcpu *vcpu,
1031                                      struct kvm_nested_guest *gp, int ric)
1032 {
1033         struct kvm *kvm = vcpu->kvm;
1034
1035         mutex_lock(&gp->tlb_lock);
1036         switch (ric) {
1037         case 0:
1038                 /* Invalidate TLB */
1039                 spin_lock(&kvm->mmu_lock);
1040                 kvmppc_free_pgtable_radix(kvm, gp->shadow_pgtable,
1041                                           gp->shadow_lpid);
1042                 kvmhv_flush_lpid(gp->shadow_lpid);
1043                 spin_unlock(&kvm->mmu_lock);
1044                 break;
1045         case 1:
1046                 /*
1047                  * Invalidate PWC
1048                  * We don't cache this -> nothing to do
1049                  */
1050                 break;
1051         case 2:
1052                 /* Invalidate TLB, PWC and caching of partition table entries */
1053                 kvmhv_flush_nested(gp);
1054                 break;
1055         default:
1056                 break;
1057         }
1058         mutex_unlock(&gp->tlb_lock);
1059 }
1060
1061 static void kvmhv_emulate_tlbie_all_lpid(struct kvm_vcpu *vcpu, int ric)
1062 {
1063         struct kvm *kvm = vcpu->kvm;
1064         struct kvm_nested_guest *gp;
1065         int i;
1066
1067         spin_lock(&kvm->mmu_lock);
1068         for (i = 0; i <= kvm->arch.max_nested_lpid; i++) {
1069                 gp = kvm->arch.nested_guests[i];
1070                 if (gp) {
1071                         spin_unlock(&kvm->mmu_lock);
1072                         kvmhv_emulate_tlbie_lpid(vcpu, gp, ric);
1073                         spin_lock(&kvm->mmu_lock);
1074                 }
1075         }
1076         spin_unlock(&kvm->mmu_lock);
1077 }
1078
1079 static int kvmhv_emulate_priv_tlbie(struct kvm_vcpu *vcpu, unsigned int instr,
1080                                     unsigned long rsval, unsigned long rbval)
1081 {
1082         struct kvm *kvm = vcpu->kvm;
1083         struct kvm_nested_guest *gp;
1084         int r, ric, prs, is, ap;
1085         int lpid;
1086         long epn;
1087         int ret = 0;
1088
1089         ric = get_ric(instr);
1090         prs = get_prs(instr);
1091         r = get_r(instr);
1092         lpid = get_lpid(rsval);
1093         is = get_is(rbval);
1094
1095         /*
1096          * These cases are invalid and are not handled:
1097          * r   != 1 -> Only radix supported
1098          * prs == 1 -> Not HV privileged
1099          * ric == 3 -> No cluster bombs for radix
1100          * is  == 1 -> Partition scoped translations not associated with pid
1101          * (!is) && (ric == 1 || ric == 2) -> Not supported by ISA
1102          */
1103         if ((!r) || (prs) || (ric == 3) || (is == 1) ||
1104             ((!is) && (ric == 1 || ric == 2)))
1105                 return -EINVAL;
1106
1107         switch (is) {
1108         case 0:
1109                 /*
1110                  * We know ric == 0
1111                  * Invalidate TLB for a given target address
1112                  */
1113                 epn = get_epn(rbval);
1114                 ap = get_ap(rbval);
1115                 ret = kvmhv_emulate_tlbie_tlb_addr(vcpu, lpid, ap, epn);
1116                 break;
1117         case 2:
1118                 /* Invalidate matching LPID */
1119                 gp = kvmhv_get_nested(kvm, lpid, false);
1120                 if (gp) {
1121                         kvmhv_emulate_tlbie_lpid(vcpu, gp, ric);
1122                         kvmhv_put_nested(gp);
1123                 }
1124                 break;
1125         case 3:
1126                 /* Invalidate ALL LPIDs */
1127                 kvmhv_emulate_tlbie_all_lpid(vcpu, ric);
1128                 break;
1129         default:
1130                 ret = -EINVAL;
1131                 break;
1132         }
1133
1134         return ret;
1135 }
1136
1137 /*
1138  * This handles the H_TLB_INVALIDATE hcall.
1139  * Parameters are (r4) tlbie instruction code, (r5) rS contents,
1140  * (r6) rB contents.
1141  */
1142 long kvmhv_do_nested_tlbie(struct kvm_vcpu *vcpu)
1143 {
1144         int ret;
1145
1146         ret = kvmhv_emulate_priv_tlbie(vcpu, kvmppc_get_gpr(vcpu, 4),
1147                         kvmppc_get_gpr(vcpu, 5), kvmppc_get_gpr(vcpu, 6));
1148         if (ret)
1149                 return H_PARAMETER;
1150         return H_SUCCESS;
1151 }
1152
1153 /* Used to convert a nested guest real address to a L1 guest real address */
1154 static int kvmhv_translate_addr_nested(struct kvm_vcpu *vcpu,
1155                                        struct kvm_nested_guest *gp,
1156                                        unsigned long n_gpa, unsigned long dsisr,
1157                                        struct kvmppc_pte *gpte_p)
1158 {
1159         u64 fault_addr, flags = dsisr & DSISR_ISSTORE;
1160         int ret;
1161
1162         ret = kvmppc_mmu_walk_radix_tree(vcpu, n_gpa, gpte_p, gp->l1_gr_to_hr,
1163                                          &fault_addr);
1164
1165         if (ret) {
1166                 /* We didn't find a pte */
1167                 if (ret == -EINVAL) {
1168                         /* Unsupported mmu config */
1169                         flags |= DSISR_UNSUPP_MMU;
1170                 } else if (ret == -ENOENT) {
1171                         /* No translation found */
1172                         flags |= DSISR_NOHPTE;
1173                 } else if (ret == -EFAULT) {
1174                         /* Couldn't access L1 real address */
1175                         flags |= DSISR_PRTABLE_FAULT;
1176                         vcpu->arch.fault_gpa = fault_addr;
1177                 } else {
1178                         /* Unknown error */
1179                         return ret;
1180                 }
1181                 goto forward_to_l1;
1182         } else {
1183                 /* We found a pte -> check permissions */
1184                 if (dsisr & DSISR_ISSTORE) {
1185                         /* Can we write? */
1186                         if (!gpte_p->may_write) {
1187                                 flags |= DSISR_PROTFAULT;
1188                                 goto forward_to_l1;
1189                         }
1190                 } else if (vcpu->arch.trap == BOOK3S_INTERRUPT_H_INST_STORAGE) {
1191                         /* Can we execute? */
1192                         if (!gpte_p->may_execute) {
1193                                 flags |= SRR1_ISI_N_OR_G;
1194                                 goto forward_to_l1;
1195                         }
1196                 } else {
1197                         /* Can we read? */
1198                         if (!gpte_p->may_read && !gpte_p->may_write) {
1199                                 flags |= DSISR_PROTFAULT;
1200                                 goto forward_to_l1;
1201                         }
1202                 }
1203         }
1204
1205         return 0;
1206
1207 forward_to_l1:
1208         vcpu->arch.fault_dsisr = flags;
1209         if (vcpu->arch.trap == BOOK3S_INTERRUPT_H_INST_STORAGE) {
1210                 vcpu->arch.shregs.msr &= ~0x783f0000ul;
1211                 vcpu->arch.shregs.msr |= flags;
1212         }
1213         return RESUME_HOST;
1214 }
1215
1216 static long kvmhv_handle_nested_set_rc(struct kvm_vcpu *vcpu,
1217                                        struct kvm_nested_guest *gp,
1218                                        unsigned long n_gpa,
1219                                        struct kvmppc_pte gpte,
1220                                        unsigned long dsisr)
1221 {
1222         struct kvm *kvm = vcpu->kvm;
1223         bool writing = !!(dsisr & DSISR_ISSTORE);
1224         u64 pgflags;
1225         long ret;
1226
1227         /* Are the rc bits set in the L1 partition scoped pte? */
1228         pgflags = _PAGE_ACCESSED;
1229         if (writing)
1230                 pgflags |= _PAGE_DIRTY;
1231         if (pgflags & ~gpte.rc)
1232                 return RESUME_HOST;
1233
1234         spin_lock(&kvm->mmu_lock);
1235         /* Set the rc bit in the pte of our (L0) pgtable for the L1 guest */
1236         ret = kvmppc_hv_handle_set_rc(kvm, kvm->arch.pgtable, writing,
1237                                      gpte.raddr, kvm->arch.lpid);
1238         if (!ret) {
1239                 ret = -EINVAL;
1240                 goto out_unlock;
1241         }
1242
1243         /* Set the rc bit in the pte of the shadow_pgtable for the nest guest */
1244         ret = kvmppc_hv_handle_set_rc(kvm, gp->shadow_pgtable, writing, n_gpa,
1245                                       gp->shadow_lpid);
1246         if (!ret)
1247                 ret = -EINVAL;
1248         else
1249                 ret = 0;
1250
1251 out_unlock:
1252         spin_unlock(&kvm->mmu_lock);
1253         return ret;
1254 }
1255
1256 static inline int kvmppc_radix_level_to_shift(int level)
1257 {
1258         switch (level) {
1259         case 2:
1260                 return PUD_SHIFT;
1261         case 1:
1262                 return PMD_SHIFT;
1263         default:
1264                 return PAGE_SHIFT;
1265         }
1266 }
1267
1268 static inline int kvmppc_radix_shift_to_level(int shift)
1269 {
1270         if (shift == PUD_SHIFT)
1271                 return 2;
1272         if (shift == PMD_SHIFT)
1273                 return 1;
1274         if (shift == PAGE_SHIFT)
1275                 return 0;
1276         WARN_ON_ONCE(1);
1277         return 0;
1278 }
1279
1280 /* called with gp->tlb_lock held */
1281 static long int __kvmhv_nested_page_fault(struct kvm_run *run,
1282                                           struct kvm_vcpu *vcpu,
1283                                           struct kvm_nested_guest *gp)
1284 {
1285         struct kvm *kvm = vcpu->kvm;
1286         struct kvm_memory_slot *memslot;
1287         struct rmap_nested *n_rmap;
1288         struct kvmppc_pte gpte;
1289         pte_t pte, *pte_p;
1290         unsigned long mmu_seq;
1291         unsigned long dsisr = vcpu->arch.fault_dsisr;
1292         unsigned long ea = vcpu->arch.fault_dar;
1293         unsigned long *rmapp;
1294         unsigned long n_gpa, gpa, gfn, perm = 0UL;
1295         unsigned int shift, l1_shift, level;
1296         bool writing = !!(dsisr & DSISR_ISSTORE);
1297         bool kvm_ro = false;
1298         long int ret;
1299
1300         if (!gp->l1_gr_to_hr) {
1301                 kvmhv_update_ptbl_cache(gp);
1302                 if (!gp->l1_gr_to_hr)
1303                         return RESUME_HOST;
1304         }
1305
1306         /* Convert the nested guest real address into a L1 guest real address */
1307
1308         n_gpa = vcpu->arch.fault_gpa & ~0xF000000000000FFFULL;
1309         if (!(dsisr & DSISR_PRTABLE_FAULT))
1310                 n_gpa |= ea & 0xFFF;
1311         ret = kvmhv_translate_addr_nested(vcpu, gp, n_gpa, dsisr, &gpte);
1312
1313         /*
1314          * If the hardware found a translation but we don't now have a usable
1315          * translation in the l1 partition-scoped tree, remove the shadow pte
1316          * and let the guest retry.
1317          */
1318         if (ret == RESUME_HOST &&
1319             (dsisr & (DSISR_PROTFAULT | DSISR_BADACCESS | DSISR_NOEXEC_OR_G |
1320                       DSISR_BAD_COPYPASTE)))
1321                 goto inval;
1322         if (ret)
1323                 return ret;
1324
1325         /* Failed to set the reference/change bits */
1326         if (dsisr & DSISR_SET_RC) {
1327                 ret = kvmhv_handle_nested_set_rc(vcpu, gp, n_gpa, gpte, dsisr);
1328                 if (ret == RESUME_HOST)
1329                         return ret;
1330                 if (ret)
1331                         goto inval;
1332                 dsisr &= ~DSISR_SET_RC;
1333                 if (!(dsisr & (DSISR_BAD_FAULT_64S | DSISR_NOHPTE |
1334                                DSISR_PROTFAULT)))
1335                         return RESUME_GUEST;
1336         }
1337
1338         /*
1339          * We took an HISI or HDSI while we were running a nested guest which
1340          * means we have no partition scoped translation for that. This means
1341          * we need to insert a pte for the mapping into our shadow_pgtable.
1342          */
1343
1344         l1_shift = gpte.page_shift;
1345         if (l1_shift < PAGE_SHIFT) {
1346                 /* We don't support l1 using a page size smaller than our own */
1347                 pr_err("KVM: L1 guest page shift (%d) less than our own (%d)\n",
1348                         l1_shift, PAGE_SHIFT);
1349                 return -EINVAL;
1350         }
1351         gpa = gpte.raddr;
1352         gfn = gpa >> PAGE_SHIFT;
1353
1354         /* 1. Get the corresponding host memslot */
1355
1356         memslot = gfn_to_memslot(kvm, gfn);
1357         if (!memslot || (memslot->flags & KVM_MEMSLOT_INVALID)) {
1358                 if (dsisr & (DSISR_PRTABLE_FAULT | DSISR_BADACCESS)) {
1359                         /* unusual error -> reflect to the guest as a DSI */
1360                         kvmppc_core_queue_data_storage(vcpu, ea, dsisr);
1361                         return RESUME_GUEST;
1362                 }
1363
1364                 /* passthrough of emulated MMIO case */
1365                 return kvmppc_hv_emulate_mmio(run, vcpu, gpa, ea, writing);
1366         }
1367         if (memslot->flags & KVM_MEM_READONLY) {
1368                 if (writing) {
1369                         /* Give the guest a DSI */
1370                         kvmppc_core_queue_data_storage(vcpu, ea,
1371                                         DSISR_ISSTORE | DSISR_PROTFAULT);
1372                         return RESUME_GUEST;
1373                 }
1374                 kvm_ro = true;
1375         }
1376
1377         /* 2. Find the host pte for this L1 guest real address */
1378
1379         /* Used to check for invalidations in progress */
1380         mmu_seq = kvm->mmu_notifier_seq;
1381         smp_rmb();
1382
1383         /* See if can find translation in our partition scoped tables for L1 */
1384         pte = __pte(0);
1385         spin_lock(&kvm->mmu_lock);
1386         pte_p = __find_linux_pte(kvm->arch.pgtable, gpa, NULL, &shift);
1387         if (!shift)
1388                 shift = PAGE_SHIFT;
1389         if (pte_p)
1390                 pte = *pte_p;
1391         spin_unlock(&kvm->mmu_lock);
1392
1393         if (!pte_present(pte) || (writing && !(pte_val(pte) & _PAGE_WRITE))) {
1394                 /* No suitable pte found -> try to insert a mapping */
1395                 ret = kvmppc_book3s_instantiate_page(vcpu, gpa, memslot,
1396                                         writing, kvm_ro, &pte, &level);
1397                 if (ret == -EAGAIN)
1398                         return RESUME_GUEST;
1399                 else if (ret)
1400                         return ret;
1401                 shift = kvmppc_radix_level_to_shift(level);
1402         }
1403         /* Align gfn to the start of the page */
1404         gfn = (gpa & ~((1UL << shift) - 1)) >> PAGE_SHIFT;
1405
1406         /* 3. Compute the pte we need to insert for nest_gpa -> host r_addr */
1407
1408         /* The permissions is the combination of the host and l1 guest ptes */
1409         perm |= gpte.may_read ? 0UL : _PAGE_READ;
1410         perm |= gpte.may_write ? 0UL : _PAGE_WRITE;
1411         perm |= gpte.may_execute ? 0UL : _PAGE_EXEC;
1412         /* Only set accessed/dirty (rc) bits if set in host and l1 guest ptes */
1413         perm |= (gpte.rc & _PAGE_ACCESSED) ? 0UL : _PAGE_ACCESSED;
1414         perm |= ((gpte.rc & _PAGE_DIRTY) && writing) ? 0UL : _PAGE_DIRTY;
1415         pte = __pte(pte_val(pte) & ~perm);
1416
1417         /* What size pte can we insert? */
1418         if (shift > l1_shift) {
1419                 u64 mask;
1420                 unsigned int actual_shift = PAGE_SHIFT;
1421                 if (PMD_SHIFT < l1_shift)
1422                         actual_shift = PMD_SHIFT;
1423                 mask = (1UL << shift) - (1UL << actual_shift);
1424                 pte = __pte(pte_val(pte) | (gpa & mask));
1425                 shift = actual_shift;
1426         }
1427         level = kvmppc_radix_shift_to_level(shift);
1428         n_gpa &= ~((1UL << shift) - 1);
1429
1430         /* 4. Insert the pte into our shadow_pgtable */
1431
1432         n_rmap = kzalloc(sizeof(*n_rmap), GFP_KERNEL);
1433         if (!n_rmap)
1434                 return RESUME_GUEST; /* Let the guest try again */
1435         n_rmap->rmap = (n_gpa & RMAP_NESTED_GPA_MASK) |
1436                 (((unsigned long) gp->l1_lpid) << RMAP_NESTED_LPID_SHIFT);
1437         rmapp = &memslot->arch.rmap[gfn - memslot->base_gfn];
1438         ret = kvmppc_create_pte(kvm, gp->shadow_pgtable, pte, n_gpa, level,
1439                                 mmu_seq, gp->shadow_lpid, rmapp, &n_rmap);
1440         if (n_rmap)
1441                 kfree(n_rmap);
1442         if (ret == -EAGAIN)
1443                 ret = RESUME_GUEST;     /* Let the guest try again */
1444
1445         return ret;
1446
1447  inval:
1448         kvmhv_invalidate_shadow_pte(vcpu, gp, n_gpa, NULL);
1449         return RESUME_GUEST;
1450 }
1451
1452 long int kvmhv_nested_page_fault(struct kvm_run *run, struct kvm_vcpu *vcpu)
1453 {
1454         struct kvm_nested_guest *gp = vcpu->arch.nested;
1455         long int ret;
1456
1457         mutex_lock(&gp->tlb_lock);
1458         ret = __kvmhv_nested_page_fault(run, vcpu, gp);
1459         mutex_unlock(&gp->tlb_lock);
1460         return ret;
1461 }
1462
1463 int kvmhv_nested_next_lpid(struct kvm *kvm, int lpid)
1464 {
1465         int ret = -1;
1466
1467         spin_lock(&kvm->mmu_lock);
1468         while (++lpid <= kvm->arch.max_nested_lpid) {
1469                 if (kvm->arch.nested_guests[lpid]) {
1470                         ret = lpid;
1471                         break;
1472                 }
1473         }
1474         spin_unlock(&kvm->mmu_lock);
1475         return ret;
1476 }