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 #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
  25 static struct patb_entry *pseries_partition_tb;
  26
  27 static void kvmhv_update_ptbl_cache(struct kvm_nested_guest *gp);
  28 static void kvmhv_free_memslot_nest_rmap(struct kvm_memory_slot *free);
  29
  30 void 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 = vc->dpdes;
  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://bugs.llvm.org/show_bug.cgi?id=49610 */
  59 static 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
  67 static 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
 103 static 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 = vc->dpdes;
 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
 141 static 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         vc->dpdes = 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
 168 void 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         vc->dpdes = hr->dpdes;
 174         vcpu->arch.hfscr = hr->hfscr;
 175         vcpu->arch.purr = hr->purr;
 176         vcpu->arch.spurr = hr->spurr;
 177         vcpu->arch.ic = hr->ic;
 178         vc->vtb = hr->vtb;
 179         vcpu->arch.fault_dar = hr->hdar;
 180         vcpu->arch.fault_dsisr = hr->hdsisr;
 181         vcpu->arch.fault_gpa = hr->asdr;
 182         vcpu->arch.emul_inst = hr->heir;
 183         vcpu->arch.shregs.srr0 = hr->srr0;
 184         vcpu->arch.shregs.srr1 = hr->srr1;
 185         vcpu->arch.shregs.sprg0 = hr->sprg[0];
 186         vcpu->arch.shregs.sprg1 = hr->sprg[1];
 187         vcpu->arch.shregs.sprg2 = hr->sprg[2];
 188         vcpu->arch.shregs.sprg3 = hr->sprg[3];
 189         vcpu->arch.pid = hr->pidr;
 190         vcpu->arch.cfar = hr->cfar;
 191         vcpu->arch.ppr = hr->ppr;
 192 }
 193
 194 static void kvmhv_nested_mmio_needed(struct kvm_vcpu *vcpu, u64 regs_ptr)
 195 {
 196         /* No need to reflect the page fault to L1, we've handled it */
 197         vcpu->arch.trap = 0;
 198
 199         /*
 200          * Since the L2 gprs have already been written back into L1 memory when
 201          * we complete the mmio, store the L1 memory location of the L2 gpr
 202          * being loaded into by the mmio so that the loaded value can be
 203          * written there in kvmppc_complete_mmio_load()
 204          */
 205         if (((vcpu->arch.io_gpr & KVM_MMIO_REG_EXT_MASK) == KVM_MMIO_REG_GPR)
 206             && (vcpu->mmio_is_write == 0)) {
 207                 vcpu->arch.nested_io_gpr = (gpa_t) regs_ptr +
 208                                            offsetof(struct pt_regs,
 209                                                     gpr[vcpu->arch.io_gpr]);
 210                 vcpu->arch.io_gpr = KVM_MMIO_REG_NESTED_GPR;
 211         }
 212 }
 213
 214 static int kvmhv_read_guest_state_and_regs(struct kvm_vcpu *vcpu,
 215                                            struct hv_guest_state *l2_hv,
 216                                            struct pt_regs *l2_regs,
 217                                            u64 hv_ptr, u64 regs_ptr)
 218 {
 219         int size;
 220
 221         if (kvm_vcpu_read_guest(vcpu, hv_ptr, &l2_hv->version,
 222                                 sizeof(l2_hv->version)))
 223                 return -1;
 224
 225         if (kvmppc_need_byteswap(vcpu))
 226                 l2_hv->version = swab64(l2_hv->version);
 227
 228         size = hv_guest_state_size(l2_hv->version);
 229         if (size < 0)
 230                 return -1;
 231
 232         return kvm_vcpu_read_guest(vcpu, hv_ptr, l2_hv, size) ||
 233                 kvm_vcpu_read_guest(vcpu, regs_ptr, l2_regs,
 234                                     sizeof(struct pt_regs));
 235 }
 236
 237 static int kvmhv_write_guest_state_and_regs(struct kvm_vcpu *vcpu,
 238                                             struct hv_guest_state *l2_hv,
 239                                             struct pt_regs *l2_regs,
 240                                             u64 hv_ptr, u64 regs_ptr)
 241 {
 242         int size;
 243
 244         size = hv_guest_state_size(l2_hv->version);
 245         if (size < 0)
 246                 return -1;
 247
 248         return kvm_vcpu_write_guest(vcpu, hv_ptr, l2_hv, size) ||
 249                 kvm_vcpu_write_guest(vcpu, regs_ptr, l2_regs,
 250                                      sizeof(struct pt_regs));
 251 }
 252
 253 static void load_l2_hv_regs(struct kvm_vcpu *vcpu,
 254                             const struct hv_guest_state *l2_hv,
 255                             const struct hv_guest_state *l1_hv, u64 *lpcr)
 256 {
 257         struct kvmppc_vcore *vc = vcpu->arch.vcore;
 258         u64 mask;
 259
 260         restore_hv_regs(vcpu, l2_hv);
 261
 262         /*
 263          * Don't let L1 change LPCR bits for the L2 except these:
 264          */
 265         mask = LPCR_DPFD | LPCR_ILE | LPCR_TC | LPCR_AIL | LPCR_LD | LPCR_MER;
 266
 267         /*
 268          * Additional filtering is required depending on hardware
 269          * and configuration.
 270          */
 271         *lpcr = kvmppc_filter_lpcr_hv(vcpu->kvm,
 272                                       (vc->lpcr & ~mask) | (*lpcr & mask));
 273
 274         /*
 275          * Don't let L1 enable features for L2 which we don't allow for L1,
 276          * but preserve the interrupt cause field.
 277          */
 278         vcpu->arch.hfscr = l2_hv->hfscr & (HFSCR_INTR_CAUSE | vcpu->arch.hfscr_permitted);
 279
 280         /* Don't let data address watchpoint match in hypervisor state */
 281         vcpu->arch.dawrx0 = l2_hv->dawrx0 & ~DAWRX_HYP;
 282         vcpu->arch.dawrx1 = l2_hv->dawrx1 & ~DAWRX_HYP;
 283
 284         /* Don't let completed instruction address breakpt match in HV state */
 285         if ((l2_hv->ciabr & CIABR_PRIV) == CIABR_PRIV_HYPER)
 286                 vcpu->arch.ciabr = l2_hv->ciabr & ~CIABR_PRIV;
 287 }
 288
 289 long kvmhv_enter_nested_guest(struct kvm_vcpu *vcpu)
 290 {
 291         long int err, r;
 292         struct kvm_nested_guest *l2;
 293         struct pt_regs l2_regs, saved_l1_regs;
 294         struct hv_guest_state l2_hv = {0}, saved_l1_hv;
 295         struct kvmppc_vcore *vc = vcpu->arch.vcore;
 296         u64 hv_ptr, regs_ptr;
 297         u64 hdec_exp, lpcr;
 298         s64 delta_purr, delta_spurr, delta_ic, delta_vtb;
 299
 300         if (vcpu->kvm->arch.l1_ptcr == 0)
 301                 return H_NOT_AVAILABLE;
 302
 303         if (MSR_TM_TRANSACTIONAL(vcpu->arch.shregs.msr))
 304                 return H_BAD_MODE;
 305
 306         /* copy parameters in */
 307         hv_ptr = kvmppc_get_gpr(vcpu, 4);
 308         regs_ptr = kvmppc_get_gpr(vcpu, 5);
 309         kvm_vcpu_srcu_read_lock(vcpu);
 310         err = kvmhv_read_guest_state_and_regs(vcpu, &l2_hv, &l2_regs,
 311                                               hv_ptr, regs_ptr);
 312         kvm_vcpu_srcu_read_unlock(vcpu);
 313         if (err)
 314                 return H_PARAMETER;
 315
 316         if (kvmppc_need_byteswap(vcpu))
 317                 byteswap_hv_regs(&l2_hv);
 318         if (l2_hv.version > HV_GUEST_STATE_VERSION)
 319                 return H_P2;
 320
 321         if (kvmppc_need_byteswap(vcpu))
 322                 byteswap_pt_regs(&l2_regs);
 323         if (l2_hv.vcpu_token >= NR_CPUS)
 324                 return H_PARAMETER;
 325
 326         /*
 327          * L1 must have set up a suspended state to enter the L2 in a
 328          * transactional state, and only in that case. These have to be
 329          * filtered out here to prevent causing a TM Bad Thing in the
 330          * host HRFID. We could synthesize a TM Bad Thing back to the L1
 331          * here but there doesn't seem like much point.
 332          */
 333         if (MSR_TM_SUSPENDED(vcpu->arch.shregs.msr)) {
 334                 if (!MSR_TM_ACTIVE(l2_regs.msr))
 335                         return H_BAD_MODE;
 336         } else {
 337                 if (l2_regs.msr & MSR_TS_MASK)
 338                         return H_BAD_MODE;
 339                 if (WARN_ON_ONCE(vcpu->arch.shregs.msr & MSR_TS_MASK))
 340                         return H_BAD_MODE;
 341         }
 342
 343         /* translate lpid */
 344         l2 = kvmhv_get_nested(vcpu->kvm, l2_hv.lpid, true);
 345         if (!l2)
 346                 return H_PARAMETER;
 347         if (!l2->l1_gr_to_hr) {
 348                 mutex_lock(&l2->tlb_lock);
 349                 kvmhv_update_ptbl_cache(l2);
 350                 mutex_unlock(&l2->tlb_lock);
 351         }
 352
 353         /* save l1 values of things */
 354         vcpu->arch.regs.msr = vcpu->arch.shregs.msr;
 355         saved_l1_regs = vcpu->arch.regs;
 356         kvmhv_save_hv_regs(vcpu, &saved_l1_hv);
 357
 358         /* convert TB values/offsets to host (L0) values */
 359         hdec_exp = l2_hv.hdec_expiry - vc->tb_offset;
 360         vc->tb_offset += l2_hv.tb_offset;
 361         vcpu->arch.dec_expires += l2_hv.tb_offset;
 362
 363         /* set L1 state to L2 state */
 364         vcpu->arch.nested = l2;
 365         vcpu->arch.nested_vcpu_id = l2_hv.vcpu_token;
 366         vcpu->arch.nested_hfscr = l2_hv.hfscr;
 367         vcpu->arch.regs = l2_regs;
 368
 369         /* Guest must always run with ME enabled, HV disabled. */
 370         vcpu->arch.shregs.msr = (vcpu->arch.regs.msr | MSR_ME) & ~MSR_HV;
 371
 372         lpcr = l2_hv.lpcr;
 373         load_l2_hv_regs(vcpu, &l2_hv, &saved_l1_hv, &lpcr);
 374
 375         vcpu->arch.ret = RESUME_GUEST;
 376         vcpu->arch.trap = 0;
 377         do {
 378                 r = kvmhv_run_single_vcpu(vcpu, hdec_exp, lpcr);
 379         } while (is_kvmppc_resume_guest(r));
 380
 381         /* save L2 state for return */
 382         l2_regs = vcpu->arch.regs;
 383         l2_regs.msr = vcpu->arch.shregs.msr;
 384         delta_purr = vcpu->arch.purr - l2_hv.purr;
 385         delta_spurr = vcpu->arch.spurr - l2_hv.spurr;
 386         delta_ic = vcpu->arch.ic - l2_hv.ic;
 387         delta_vtb = vc->vtb - l2_hv.vtb;
 388         save_hv_return_state(vcpu, &l2_hv);
 389
 390         /* restore L1 state */
 391         vcpu->arch.nested = NULL;
 392         vcpu->arch.regs = saved_l1_regs;
 393         vcpu->arch.shregs.msr = saved_l1_regs.msr & ~MSR_TS_MASK;
 394         /* set L1 MSR TS field according to L2 transaction state */
 395         if (l2_regs.msr & MSR_TS_MASK)
 396                 vcpu->arch.shregs.msr |= MSR_TS_S;
 397         vc->tb_offset = saved_l1_hv.tb_offset;
 398         /* XXX: is this always the same delta as saved_l1_hv.tb_offset? */
 399         vcpu->arch.dec_expires -= l2_hv.tb_offset;
 400         restore_hv_regs(vcpu, &saved_l1_hv);
 401         vcpu->arch.purr += delta_purr;
 402         vcpu->arch.spurr += delta_spurr;
 403         vcpu->arch.ic += delta_ic;
 404         vc->vtb += delta_vtb;
 405
 406         kvmhv_put_nested(l2);
 407
 408         /* copy l2_hv_state and regs back to guest */
 409         if (kvmppc_need_byteswap(vcpu)) {
 410                 byteswap_hv_regs(&l2_hv);
 411                 byteswap_pt_regs(&l2_regs);
 412         }
 413         kvm_vcpu_srcu_read_lock(vcpu);
 414         err = kvmhv_write_guest_state_and_regs(vcpu, &l2_hv, &l2_regs,
 415                                                hv_ptr, regs_ptr);
 416         kvm_vcpu_srcu_read_unlock(vcpu);
 417         if (err)
 418                 return H_AUTHORITY;
 419
 420         if (r == -EINTR)
 421                 return H_INTERRUPT;
 422
 423         if (vcpu->mmio_needed) {
 424                 kvmhv_nested_mmio_needed(vcpu, regs_ptr);
 425                 return H_TOO_HARD;
 426         }
 427
 428         return vcpu->arch.trap;
 429 }
 430
 431 long kvmhv_nested_init(void)
 432 {
 433         long int ptb_order;
 434         unsigned long ptcr;
 435         long rc;
 436
 437         if (!kvmhv_on_pseries())
 438                 return 0;
 439         if (!radix_enabled())
 440                 return -ENODEV;
 441
 442         /* Partition table entry is 1<<4 bytes in size, hence the 4. */
 443         ptb_order = KVM_MAX_NESTED_GUESTS_SHIFT + 4;
 444         /* Minimum partition table size is 1<<12 bytes */
 445         if (ptb_order < 12)
 446                 ptb_order = 12;
 447         pseries_partition_tb = kmalloc(sizeof(struct patb_entry) << ptb_order,
 448                                        GFP_KERNEL);
 449         if (!pseries_partition_tb) {
 450                 pr_err("kvm-hv: failed to allocated nested partition table\n");
 451                 return -ENOMEM;
 452         }
 453
 454         ptcr = __pa(pseries_partition_tb) | (ptb_order - 12);
 455         rc = plpar_hcall_norets(H_SET_PARTITION_TABLE, ptcr);
 456         if (rc != H_SUCCESS) {
 457                 pr_err("kvm-hv: Parent hypervisor does not support nesting (rc=%ld)\n",
 458                        rc);
 459                 kfree(pseries_partition_tb);
 460                 pseries_partition_tb = NULL;
 461                 return -ENODEV;
 462         }
 463
 464         return 0;
 465 }
 466
 467 void kvmhv_nested_exit(void)
 468 {
 469         /*
 470          * N.B. the kvmhv_on_pseries() test is there because it enables
 471          * the compiler to remove the call to plpar_hcall_norets()
 472          * when CONFIG_PPC_PSERIES=n.
 473          */
 474         if (kvmhv_on_pseries() && pseries_partition_tb) {
 475                 plpar_hcall_norets(H_SET_PARTITION_TABLE, 0);
 476                 kfree(pseries_partition_tb);
 477                 pseries_partition_tb = NULL;
 478         }
 479 }
 480
 481 static void kvmhv_flush_lpid(unsigned int lpid)
 482 {
 483         long rc;
 484
 485         if (!kvmhv_on_pseries()) {
 486                 radix__flush_all_lpid(lpid);
 487                 return;
 488         }
 489
 490         if (!firmware_has_feature(FW_FEATURE_RPT_INVALIDATE))
 491                 rc = plpar_hcall_norets(H_TLB_INVALIDATE, H_TLBIE_P1_ENC(2, 0, 1),
 492                                         lpid, TLBIEL_INVAL_SET_LPID);
 493         else
 494                 rc = pseries_rpt_invalidate(lpid, H_RPTI_TARGET_CMMU,
 495                                             H_RPTI_TYPE_NESTED |
 496                                             H_RPTI_TYPE_TLB | H_RPTI_TYPE_PWC |
 497                                             H_RPTI_TYPE_PAT,
 498                                             H_RPTI_PAGE_ALL, 0, -1UL);
 499         if (rc)
 500                 pr_err("KVM: TLB LPID invalidation hcall failed, rc=%ld\n", rc);
 501 }
 502
 503 void kvmhv_set_ptbl_entry(unsigned int lpid, u64 dw0, u64 dw1)
 504 {
 505         if (!kvmhv_on_pseries()) {
 506                 mmu_partition_table_set_entry(lpid, dw0, dw1, true);
 507                 return;
 508         }
 509
 510         pseries_partition_tb[lpid].patb0 = cpu_to_be64(dw0);
 511         pseries_partition_tb[lpid].patb1 = cpu_to_be64(dw1);
 512         /* L0 will do the necessary barriers */
 513         kvmhv_flush_lpid(lpid);
 514 }
 515
 516 static void kvmhv_set_nested_ptbl(struct kvm_nested_guest *gp)
 517 {
 518         unsigned long dw0;
 519
 520         dw0 = PATB_HR | radix__get_tree_size() |
 521                 __pa(gp->shadow_pgtable) | RADIX_PGD_INDEX_SIZE;
 522         kvmhv_set_ptbl_entry(gp->shadow_lpid, dw0, gp->process_table);
 523 }
 524
 525 /*
 526  * Handle the H_SET_PARTITION_TABLE hcall.
 527  * r4 = guest real address of partition table + log_2(size) - 12
 528  * (formatted as for the PTCR).
 529  */
 530 long kvmhv_set_partition_table(struct kvm_vcpu *vcpu)
 531 {
 532         struct kvm *kvm = vcpu->kvm;
 533         unsigned long ptcr = kvmppc_get_gpr(vcpu, 4);
 534         int srcu_idx;
 535         long ret = H_SUCCESS;
 536
 537         srcu_idx = srcu_read_lock(&kvm->srcu);
 538         /* Check partition size and base address. */
 539         if ((ptcr & PRTS_MASK) + 12 - 4 > KVM_MAX_NESTED_GUESTS_SHIFT ||
 540             !kvm_is_visible_gfn(vcpu->kvm, (ptcr & PRTB_MASK) >> PAGE_SHIFT))
 541                 ret = H_PARAMETER;
 542         srcu_read_unlock(&kvm->srcu, srcu_idx);
 543         if (ret == H_SUCCESS)
 544                 kvm->arch.l1_ptcr = ptcr;
 545
 546         return ret;
 547 }
 548
 549 /*
 550  * Handle the H_COPY_TOFROM_GUEST hcall.
 551  * r4 = L1 lpid of nested guest
 552  * r5 = pid
 553  * r6 = eaddr to access
 554  * r7 = to buffer (L1 gpa)
 555  * r8 = from buffer (L1 gpa)
 556  * r9 = n bytes to copy
 557  */
 558 long kvmhv_copy_tofrom_guest_nested(struct kvm_vcpu *vcpu)
 559 {
 560         struct kvm_nested_guest *gp;
 561         int l1_lpid = kvmppc_get_gpr(vcpu, 4);
 562         int pid = kvmppc_get_gpr(vcpu, 5);
 563         gva_t eaddr = kvmppc_get_gpr(vcpu, 6);
 564         gpa_t gp_to = (gpa_t) kvmppc_get_gpr(vcpu, 7);
 565         gpa_t gp_from = (gpa_t) kvmppc_get_gpr(vcpu, 8);
 566         void *buf;
 567         unsigned long n = kvmppc_get_gpr(vcpu, 9);
 568         bool is_load = !!gp_to;
 569         long rc;
 570
 571         if (gp_to && gp_from) /* One must be NULL to determine the direction */
 572                 return H_PARAMETER;
 573
 574         if (eaddr & (0xFFFUL << 52))
 575                 return H_PARAMETER;
 576
 577         buf = kzalloc(n, GFP_KERNEL | __GFP_NOWARN);
 578         if (!buf)
 579                 return H_NO_MEM;
 580
 581         gp = kvmhv_get_nested(vcpu->kvm, l1_lpid, false);
 582         if (!gp) {
 583                 rc = H_PARAMETER;
 584                 goto out_free;
 585         }
 586
 587         mutex_lock(&gp->tlb_lock);
 588
 589         if (is_load) {
 590                 /* Load from the nested guest into our buffer */
 591                 rc = __kvmhv_copy_tofrom_guest_radix(gp->shadow_lpid, pid,
 592                                                      eaddr, buf, NULL, n);
 593                 if (rc)
 594                         goto not_found;
 595
 596                 /* Write what was loaded into our buffer back to the L1 guest */
 597                 kvm_vcpu_srcu_read_lock(vcpu);
 598                 rc = kvm_vcpu_write_guest(vcpu, gp_to, buf, n);
 599                 kvm_vcpu_srcu_read_unlock(vcpu);
 600                 if (rc)
 601                         goto not_found;
 602         } else {
 603                 /* Load the data to be stored from the L1 guest into our buf */
 604                 kvm_vcpu_srcu_read_lock(vcpu);
 605                 rc = kvm_vcpu_read_guest(vcpu, gp_from, buf, n);
 606                 kvm_vcpu_srcu_read_unlock(vcpu);
 607                 if (rc)
 608                         goto not_found;
 609
 610                 /* Store from our buffer into the nested guest */
 611                 rc = __kvmhv_copy_tofrom_guest_radix(gp->shadow_lpid, pid,
 612                                                      eaddr, NULL, buf, n);
 613                 if (rc)
 614                         goto not_found;
 615         }
 616
 617 out_unlock:
 618         mutex_unlock(&gp->tlb_lock);
 619         kvmhv_put_nested(gp);
 620 out_free:
 621         kfree(buf);
 622         return rc;
 623 not_found:
 624         rc = H_NOT_FOUND;
 625         goto out_unlock;
 626 }
 627
 628 /*
 629  * Reload the partition table entry for a guest.
 630  * Caller must hold gp->tlb_lock.
 631  */
 632 static void kvmhv_update_ptbl_cache(struct kvm_nested_guest *gp)
 633 {
 634         int ret;
 635         struct patb_entry ptbl_entry;
 636         unsigned long ptbl_addr;
 637         struct kvm *kvm = gp->l1_host;
 638
 639         ret = -EFAULT;
 640         ptbl_addr = (kvm->arch.l1_ptcr & PRTB_MASK) + (gp->l1_lpid << 4);
 641         if (gp->l1_lpid < (1ul << ((kvm->arch.l1_ptcr & PRTS_MASK) + 12 - 4))) {
 642                 int srcu_idx = srcu_read_lock(&kvm->srcu);
 643                 ret = kvm_read_guest(kvm, ptbl_addr,
 644                                      &ptbl_entry, sizeof(ptbl_entry));
 645                 srcu_read_unlock(&kvm->srcu, srcu_idx);
 646         }
 647         if (ret) {
 648                 gp->l1_gr_to_hr = 0;
 649                 gp->process_table = 0;
 650         } else {
 651                 gp->l1_gr_to_hr = be64_to_cpu(ptbl_entry.patb0);
 652                 gp->process_table = be64_to_cpu(ptbl_entry.patb1);
 653         }
 654         kvmhv_set_nested_ptbl(gp);
 655 }
 656
 657 void kvmhv_vm_nested_init(struct kvm *kvm)
 658 {
 659         idr_init(&kvm->arch.kvm_nested_guest_idr);
 660 }
 661
 662 static struct kvm_nested_guest *__find_nested(struct kvm *kvm, int lpid)
 663 {
 664         return idr_find(&kvm->arch.kvm_nested_guest_idr, lpid);
 665 }
 666
 667 static bool __prealloc_nested(struct kvm *kvm, int lpid)
 668 {
 669         if (idr_alloc(&kvm->arch.kvm_nested_guest_idr,
 670                                 NULL, lpid, lpid + 1, GFP_KERNEL) != lpid)
 671                 return false;
 672         return true;
 673 }
 674
 675 static void __add_nested(struct kvm *kvm, int lpid, struct kvm_nested_guest *gp)
 676 {
 677         if (idr_replace(&kvm->arch.kvm_nested_guest_idr, gp, lpid))
 678                 WARN_ON(1);
 679 }
 680
 681 static void __remove_nested(struct kvm *kvm, int lpid)
 682 {
 683         idr_remove(&kvm->arch.kvm_nested_guest_idr, lpid);
 684 }
 685
 686 static struct kvm_nested_guest *kvmhv_alloc_nested(struct kvm *kvm, unsigned int lpid)
 687 {
 688         struct kvm_nested_guest *gp;
 689         long shadow_lpid;
 690
 691         gp = kzalloc(sizeof(*gp), GFP_KERNEL);
 692         if (!gp)
 693                 return NULL;
 694         gp->l1_host = kvm;
 695         gp->l1_lpid = lpid;
 696         mutex_init(&gp->tlb_lock);
 697         gp->shadow_pgtable = pgd_alloc(kvm->mm);
 698         if (!gp->shadow_pgtable)
 699                 goto out_free;
 700         shadow_lpid = kvmppc_alloc_lpid();
 701         if (shadow_lpid < 0)
 702                 goto out_free2;
 703         gp->shadow_lpid = shadow_lpid;
 704         gp->radix = 1;
 705
 706         memset(gp->prev_cpu, -1, sizeof(gp->prev_cpu));
 707
 708         return gp;
 709
 710  out_free2:
 711         pgd_free(kvm->mm, gp->shadow_pgtable);
 712  out_free:
 713         kfree(gp);
 714         return NULL;
 715 }
 716
 717 /*
 718  * Free up any resources allocated for a nested guest.
 719  */
 720 static void kvmhv_release_nested(struct kvm_nested_guest *gp)
 721 {
 722         struct kvm *kvm = gp->l1_host;
 723
 724         if (gp->shadow_pgtable) {
 725                 /*
 726                  * No vcpu is using this struct and no call to
 727                  * kvmhv_get_nested can find this struct,
 728                  * so we don't need to hold kvm->mmu_lock.
 729                  */
 730                 kvmppc_free_pgtable_radix(kvm, gp->shadow_pgtable,
 731                                           gp->shadow_lpid);
 732                 pgd_free(kvm->mm, gp->shadow_pgtable);
 733         }
 734         kvmhv_set_ptbl_entry(gp->shadow_lpid, 0, 0);
 735         kvmppc_free_lpid(gp->shadow_lpid);
 736         kfree(gp);
 737 }
 738
 739 static void kvmhv_remove_nested(struct kvm_nested_guest *gp)
 740 {
 741         struct kvm *kvm = gp->l1_host;
 742         int lpid = gp->l1_lpid;
 743         long ref;
 744
 745         spin_lock(&kvm->mmu_lock);
 746         if (gp == __find_nested(kvm, lpid)) {
 747                 __remove_nested(kvm, lpid);
 748                 --gp->refcnt;
 749         }
 750         ref = gp->refcnt;
 751         spin_unlock(&kvm->mmu_lock);
 752         if (ref == 0)
 753                 kvmhv_release_nested(gp);
 754 }
 755
 756 /*
 757  * Free up all nested resources allocated for this guest.
 758  * This is called with no vcpus of the guest running, when
 759  * switching the guest to HPT mode or when destroying the
 760  * guest.
 761  */
 762 void kvmhv_release_all_nested(struct kvm *kvm)
 763 {
 764         int lpid;
 765         struct kvm_nested_guest *gp;
 766         struct kvm_nested_guest *freelist = NULL;
 767         struct kvm_memory_slot *memslot;
 768         int srcu_idx, bkt;
 769
 770         spin_lock(&kvm->mmu_lock);
 771         idr_for_each_entry(&kvm->arch.kvm_nested_guest_idr, gp, lpid) {
 772                 __remove_nested(kvm, lpid);
 773                 if (--gp->refcnt == 0) {
 774                         gp->next = freelist;
 775                         freelist = gp;
 776                 }
 777         }
 778         idr_destroy(&kvm->arch.kvm_nested_guest_idr);
 779         /* idr is empty and may be reused at this point */
 780         spin_unlock(&kvm->mmu_lock);
 781         while ((gp = freelist) != NULL) {
 782                 freelist = gp->next;
 783                 kvmhv_release_nested(gp);
 784         }
 785
 786         srcu_idx = srcu_read_lock(&kvm->srcu);
 787         kvm_for_each_memslot(memslot, bkt, kvm_memslots(kvm))
 788                 kvmhv_free_memslot_nest_rmap(memslot);
 789         srcu_read_unlock(&kvm->srcu, srcu_idx);
 790 }
 791
 792 /* caller must hold gp->tlb_lock */
 793 static void kvmhv_flush_nested(struct kvm_nested_guest *gp)
 794 {
 795         struct kvm *kvm = gp->l1_host;
 796
 797         spin_lock(&kvm->mmu_lock);
 798         kvmppc_free_pgtable_radix(kvm, gp->shadow_pgtable, gp->shadow_lpid);
 799         spin_unlock(&kvm->mmu_lock);
 800         kvmhv_flush_lpid(gp->shadow_lpid);
 801         kvmhv_update_ptbl_cache(gp);
 802         if (gp->l1_gr_to_hr == 0)
 803                 kvmhv_remove_nested(gp);
 804 }
 805
 806 struct kvm_nested_guest *kvmhv_get_nested(struct kvm *kvm, int l1_lpid,
 807                                           bool create)
 808 {
 809         struct kvm_nested_guest *gp, *newgp;
 810
 811         if (l1_lpid >= (1ul << ((kvm->arch.l1_ptcr & PRTS_MASK) + 12 - 4)))
 812                 return NULL;
 813
 814         spin_lock(&kvm->mmu_lock);
 815         gp = __find_nested(kvm, l1_lpid);
 816         if (gp)
 817                 ++gp->refcnt;
 818         spin_unlock(&kvm->mmu_lock);
 819
 820         if (gp || !create)
 821                 return gp;
 822
 823         newgp = kvmhv_alloc_nested(kvm, l1_lpid);
 824         if (!newgp)
 825                 return NULL;
 826
 827         if (!__prealloc_nested(kvm, l1_lpid)) {
 828                 kvmhv_release_nested(newgp);
 829                 return NULL;
 830         }
 831
 832         spin_lock(&kvm->mmu_lock);
 833         gp = __find_nested(kvm, l1_lpid);
 834         if (!gp) {
 835                 __add_nested(kvm, l1_lpid, newgp);
 836                 ++newgp->refcnt;
 837                 gp = newgp;
 838                 newgp = NULL;
 839         }
 840         ++gp->refcnt;
 841         spin_unlock(&kvm->mmu_lock);
 842
 843         if (newgp)
 844                 kvmhv_release_nested(newgp);
 845
 846         return gp;
 847 }
 848
 849 void kvmhv_put_nested(struct kvm_nested_guest *gp)
 850 {
 851         struct kvm *kvm = gp->l1_host;
 852         long ref;
 853
 854         spin_lock(&kvm->mmu_lock);
 855         ref = --gp->refcnt;
 856         spin_unlock(&kvm->mmu_lock);
 857         if (ref == 0)
 858                 kvmhv_release_nested(gp);
 859 }
 860
 861 pte_t *find_kvm_nested_guest_pte(struct kvm *kvm, unsigned long lpid,
 862                                  unsigned long ea, unsigned *hshift)
 863 {
 864         struct kvm_nested_guest *gp;
 865         pte_t *pte;
 866
 867         gp = __find_nested(kvm, lpid);
 868         if (!gp)
 869                 return NULL;
 870
 871         VM_WARN(!spin_is_locked(&kvm->mmu_lock),
 872                 "%s called with kvm mmu_lock not held \n", __func__);
 873         pte = __find_linux_pte(gp->shadow_pgtable, ea, NULL, hshift);
 874
 875         return pte;
 876 }
 877
 878 static inline bool kvmhv_n_rmap_is_equal(u64 rmap_1, u64 rmap_2)
 879 {
 880         return !((rmap_1 ^ rmap_2) & (RMAP_NESTED_LPID_MASK |
 881                                        RMAP_NESTED_GPA_MASK));
 882 }
 883
 884 void kvmhv_insert_nest_rmap(struct kvm *kvm, unsigned long *rmapp,
 885                             struct rmap_nested **n_rmap)
 886 {
 887         struct llist_node *entry = ((struct llist_head *) rmapp)->first;
 888         struct rmap_nested *cursor;
 889         u64 rmap, new_rmap = (*n_rmap)->rmap;
 890
 891         /* Are there any existing entries? */
 892         if (!(*rmapp)) {
 893                 /* No -> use the rmap as a single entry */
 894                 *rmapp = new_rmap | RMAP_NESTED_IS_SINGLE_ENTRY;
 895                 return;
 896         }
 897
 898         /* Do any entries match what we're trying to insert? */
 899         for_each_nest_rmap_safe(cursor, entry, &rmap) {
 900                 if (kvmhv_n_rmap_is_equal(rmap, new_rmap))
 901                         return;
 902         }
 903
 904         /* Do we need to create a list or just add the new entry? */
 905         rmap = *rmapp;
 906         if (rmap & RMAP_NESTED_IS_SINGLE_ENTRY) /* Not previously a list */
 907                 *rmapp = 0UL;
 908         llist_add(&((*n_rmap)->list), (struct llist_head *) rmapp);
 909         if (rmap & RMAP_NESTED_IS_SINGLE_ENTRY) /* Not previously a list */
 910                 (*n_rmap)->list.next = (struct llist_node *) rmap;
 911
 912         /* Set NULL so not freed by caller */
 913         *n_rmap = NULL;
 914 }
 915
 916 static void kvmhv_update_nest_rmap_rc(struct kvm *kvm, u64 n_rmap,
 917                                       unsigned long clr, unsigned long set,
 918                                       unsigned long hpa, unsigned long mask)
 919 {
 920         unsigned long gpa;
 921         unsigned int shift, lpid;
 922         pte_t *ptep;
 923
 924         gpa = n_rmap & RMAP_NESTED_GPA_MASK;
 925         lpid = (n_rmap & RMAP_NESTED_LPID_MASK) >> RMAP_NESTED_LPID_SHIFT;
 926
 927         /* Find the pte */
 928         ptep = find_kvm_nested_guest_pte(kvm, lpid, gpa, &shift);
 929         /*
 930          * If the pte is present and the pfn is still the same, update the pte.
 931          * If the pfn has changed then this is a stale rmap entry, the nested
 932          * gpa actually points somewhere else now, and there is nothing to do.
 933          * XXX A future optimisation would be to remove the rmap entry here.
 934          */
 935         if (ptep && pte_present(*ptep) && ((pte_val(*ptep) & mask) == hpa)) {
 936                 __radix_pte_update(ptep, clr, set);
 937                 kvmppc_radix_tlbie_page(kvm, gpa, shift, lpid);
 938         }
 939 }
 940
 941 /*
 942  * For a given list of rmap entries, update the rc bits in all ptes in shadow
 943  * page tables for nested guests which are referenced by the rmap list.
 944  */
 945 void kvmhv_update_nest_rmap_rc_list(struct kvm *kvm, unsigned long *rmapp,
 946                                     unsigned long clr, unsigned long set,
 947                                     unsigned long hpa, unsigned long nbytes)
 948 {
 949         struct llist_node *entry = ((struct llist_head *) rmapp)->first;
 950         struct rmap_nested *cursor;
 951         unsigned long rmap, mask;
 952
 953         if ((clr | set) & ~(_PAGE_DIRTY | _PAGE_ACCESSED))
 954                 return;
 955
 956         mask = PTE_RPN_MASK & ~(nbytes - 1);
 957         hpa &= mask;
 958
 959         for_each_nest_rmap_safe(cursor, entry, &rmap)
 960                 kvmhv_update_nest_rmap_rc(kvm, rmap, clr, set, hpa, mask);
 961 }
 962
 963 static void kvmhv_remove_nest_rmap(struct kvm *kvm, u64 n_rmap,
 964                                    unsigned long hpa, unsigned long mask)
 965 {
 966         struct kvm_nested_guest *gp;
 967         unsigned long gpa;
 968         unsigned int shift, lpid;
 969         pte_t *ptep;
 970
 971         gpa = n_rmap & RMAP_NESTED_GPA_MASK;
 972         lpid = (n_rmap & RMAP_NESTED_LPID_MASK) >> RMAP_NESTED_LPID_SHIFT;
 973         gp = __find_nested(kvm, lpid);
 974         if (!gp)
 975                 return;
 976
 977         /* Find and invalidate the pte */
 978         ptep = find_kvm_nested_guest_pte(kvm, lpid, gpa, &shift);
 979         /* Don't spuriously invalidate ptes if the pfn has changed */
 980         if (ptep && pte_present(*ptep) && ((pte_val(*ptep) & mask) == hpa))
 981                 kvmppc_unmap_pte(kvm, ptep, gpa, shift, NULL, gp->shadow_lpid);
 982 }
 983
 984 static void kvmhv_remove_nest_rmap_list(struct kvm *kvm, unsigned long *rmapp,
 985                                         unsigned long hpa, unsigned long mask)
 986 {
 987         struct llist_node *entry = llist_del_all((struct llist_head *) rmapp);
 988         struct rmap_nested *cursor;
 989         unsigned long rmap;
 990
 991         for_each_nest_rmap_safe(cursor, entry, &rmap) {
 992                 kvmhv_remove_nest_rmap(kvm, rmap, hpa, mask);
 993                 kfree(cursor);
 994         }
 995 }
 996
 997 /* called with kvm->mmu_lock held */
 998 void kvmhv_remove_nest_rmap_range(struct kvm *kvm,
 999                                   const struct kvm_memory_slot *memslot,
1000                                   unsigned long gpa, unsigned long hpa,
1001                                   unsigned long nbytes)
1002 {
1003         unsigned long gfn, end_gfn;
1004         unsigned long addr_mask;
1005
1006         if (!memslot)
1007                 return;
1008         gfn = (gpa >> PAGE_SHIFT) - memslot->base_gfn;
1009         end_gfn = gfn + (nbytes >> PAGE_SHIFT);
1010
1011         addr_mask = PTE_RPN_MASK & ~(nbytes - 1);
1012         hpa &= addr_mask;
1013
1014         for (; gfn < end_gfn; gfn++) {
1015                 unsigned long *rmap = &memslot->arch.rmap[gfn];
1016                 kvmhv_remove_nest_rmap_list(kvm, rmap, hpa, addr_mask);
1017         }
1018 }
1019
1020 static void kvmhv_free_memslot_nest_rmap(struct kvm_memory_slot *free)
1021 {
1022         unsigned long page;
1023
1024         for (page = 0; page < free->npages; page++) {
1025                 unsigned long rmap, *rmapp = &free->arch.rmap[page];
1026                 struct rmap_nested *cursor;
1027                 struct llist_node *entry;
1028
1029                 entry = llist_del_all((struct llist_head *) rmapp);
1030                 for_each_nest_rmap_safe(cursor, entry, &rmap)
1031                         kfree(cursor);
1032         }
1033 }
1034
1035 static bool kvmhv_invalidate_shadow_pte(struct kvm_vcpu *vcpu,
1036                                         struct kvm_nested_guest *gp,
1037                                         long gpa, int *shift_ret)
1038 {
1039         struct kvm *kvm = vcpu->kvm;
1040         bool ret = false;
1041         pte_t *ptep;
1042         int shift;
1043
1044         spin_lock(&kvm->mmu_lock);
1045         ptep = find_kvm_nested_guest_pte(kvm, gp->l1_lpid, gpa, &shift);
1046         if (!shift)
1047                 shift = PAGE_SHIFT;
1048         if (ptep && pte_present(*ptep)) {
1049                 kvmppc_unmap_pte(kvm, ptep, gpa, shift, NULL, gp->shadow_lpid);
1050                 ret = true;
1051         }
1052         spin_unlock(&kvm->mmu_lock);
1053
1054         if (shift_ret)
1055                 *shift_ret = shift;
1056         return ret;
1057 }
1058
1059 static inline int get_ric(unsigned int instr)
1060 {
1061         return (instr >> 18) & 0x3;
1062 }
1063
1064 static inline int get_prs(unsigned int instr)
1065 {
1066         return (instr >> 17) & 0x1;
1067 }
1068
1069 static inline int get_r(unsigned int instr)
1070 {
1071         return (instr >> 16) & 0x1;
1072 }
1073
1074 static inline int get_lpid(unsigned long r_val)
1075 {
1076         return r_val & 0xffffffff;
1077 }
1078
1079 static inline int get_is(unsigned long r_val)
1080 {
1081         return (r_val >> 10) & 0x3;
1082 }
1083
1084 static inline int get_ap(unsigned long r_val)
1085 {
1086         return (r_val >> 5) & 0x7;
1087 }
1088
1089 static inline long get_epn(unsigned long r_val)
1090 {
1091         return r_val >> 12;
1092 }
1093
1094 static int kvmhv_emulate_tlbie_tlb_addr(struct kvm_vcpu *vcpu, int lpid,
1095                                         int ap, long epn)
1096 {
1097         struct kvm *kvm = vcpu->kvm;
1098         struct kvm_nested_guest *gp;
1099         long npages;
1100         int shift, shadow_shift;
1101         unsigned long addr;
1102
1103         shift = ap_to_shift(ap);
1104         addr = epn << 12;
1105         if (shift < 0)
1106                 /* Invalid ap encoding */
1107                 return -EINVAL;
1108
1109         addr &= ~((1UL << shift) - 1);
1110         npages = 1UL << (shift - PAGE_SHIFT);
1111
1112         gp = kvmhv_get_nested(kvm, lpid, false);
1113         if (!gp) /* No such guest -> nothing to do */
1114                 return 0;
1115         mutex_lock(&gp->tlb_lock);
1116
1117         /* There may be more than one host page backing this single guest pte */
1118         do {
1119                 kvmhv_invalidate_shadow_pte(vcpu, gp, addr, &shadow_shift);
1120
1121                 npages -= 1UL << (shadow_shift - PAGE_SHIFT);
1122                 addr += 1UL << shadow_shift;
1123         } while (npages > 0);
1124
1125         mutex_unlock(&gp->tlb_lock);
1126         kvmhv_put_nested(gp);
1127         return 0;
1128 }
1129
1130 static void kvmhv_emulate_tlbie_lpid(struct kvm_vcpu *vcpu,
1131                                      struct kvm_nested_guest *gp, int ric)
1132 {
1133         struct kvm *kvm = vcpu->kvm;
1134
1135         mutex_lock(&gp->tlb_lock);
1136         switch (ric) {
1137         case 0:
1138                 /* Invalidate TLB */
1139                 spin_lock(&kvm->mmu_lock);
1140                 kvmppc_free_pgtable_radix(kvm, gp->shadow_pgtable,
1141                                           gp->shadow_lpid);
1142                 kvmhv_flush_lpid(gp->shadow_lpid);
1143                 spin_unlock(&kvm->mmu_lock);
1144                 break;
1145         case 1:
1146                 /*
1147                  * Invalidate PWC
1148                  * We don't cache this -> nothing to do
1149                  */
1150                 break;
1151         case 2:
1152                 /* Invalidate TLB, PWC and caching of partition table entries */
1153                 kvmhv_flush_nested(gp);
1154                 break;
1155         default:
1156                 break;
1157         }
1158         mutex_unlock(&gp->tlb_lock);
1159 }
1160
1161 static void kvmhv_emulate_tlbie_all_lpid(struct kvm_vcpu *vcpu, int ric)
1162 {
1163         struct kvm *kvm = vcpu->kvm;
1164         struct kvm_nested_guest *gp;
1165         int lpid;
1166
1167         spin_lock(&kvm->mmu_lock);
1168         idr_for_each_entry(&kvm->arch.kvm_nested_guest_idr, gp, lpid) {
1169                 spin_unlock(&kvm->mmu_lock);
1170                 kvmhv_emulate_tlbie_lpid(vcpu, gp, ric);
1171                 spin_lock(&kvm->mmu_lock);
1172         }
1173         spin_unlock(&kvm->mmu_lock);
1174 }
1175
1176 static int kvmhv_emulate_priv_tlbie(struct kvm_vcpu *vcpu, unsigned int instr,
1177                                     unsigned long rsval, unsigned long rbval)
1178 {
1179         struct kvm *kvm = vcpu->kvm;
1180         struct kvm_nested_guest *gp;
1181         int r, ric, prs, is, ap;
1182         int lpid;
1183         long epn;
1184         int ret = 0;
1185
1186         ric = get_ric(instr);
1187         prs = get_prs(instr);
1188         r = get_r(instr);
1189         lpid = get_lpid(rsval);
1190         is = get_is(rbval);
1191
1192         /*
1193          * These cases are invalid and are not handled:
1194          * r   != 1 -> Only radix supported
1195          * prs == 1 -> Not HV privileged
1196          * ric == 3 -> No cluster bombs for radix
1197          * is  == 1 -> Partition scoped translations not associated with pid
1198          * (!is) && (ric == 1 || ric == 2) -> Not supported by ISA
1199          */
1200         if ((!r) || (prs) || (ric == 3) || (is == 1) ||
1201             ((!is) && (ric == 1 || ric == 2)))
1202                 return -EINVAL;
1203
1204         switch (is) {
1205         case 0:
1206                 /*
1207                  * We know ric == 0
1208                  * Invalidate TLB for a given target address
1209                  */
1210                 epn = get_epn(rbval);
1211                 ap = get_ap(rbval);
1212                 ret = kvmhv_emulate_tlbie_tlb_addr(vcpu, lpid, ap, epn);
1213                 break;
1214         case 2:
1215                 /* Invalidate matching LPID */
1216                 gp = kvmhv_get_nested(kvm, lpid, false);
1217                 if (gp) {
1218                         kvmhv_emulate_tlbie_lpid(vcpu, gp, ric);
1219                         kvmhv_put_nested(gp);
1220                 }
1221                 break;
1222         case 3:
1223                 /* Invalidate ALL LPIDs */
1224                 kvmhv_emulate_tlbie_all_lpid(vcpu, ric);
1225                 break;
1226         default:
1227                 ret = -EINVAL;
1228                 break;
1229         }
1230
1231         return ret;
1232 }
1233
1234 /*
1235  * This handles the H_TLB_INVALIDATE hcall.
1236  * Parameters are (r4) tlbie instruction code, (r5) rS contents,
1237  * (r6) rB contents.
1238  */
1239 long kvmhv_do_nested_tlbie(struct kvm_vcpu *vcpu)
1240 {
1241         int ret;
1242
1243         ret = kvmhv_emulate_priv_tlbie(vcpu, kvmppc_get_gpr(vcpu, 4),
1244                         kvmppc_get_gpr(vcpu, 5), kvmppc_get_gpr(vcpu, 6));
1245         if (ret)
1246                 return H_PARAMETER;
1247         return H_SUCCESS;
1248 }
1249
1250 static long do_tlb_invalidate_nested_all(struct kvm_vcpu *vcpu,
1251                                          unsigned long lpid, unsigned long ric)
1252 {
1253         struct kvm *kvm = vcpu->kvm;
1254         struct kvm_nested_guest *gp;
1255
1256         gp = kvmhv_get_nested(kvm, lpid, false);
1257         if (gp) {
1258                 kvmhv_emulate_tlbie_lpid(vcpu, gp, ric);
1259                 kvmhv_put_nested(gp);
1260         }
1261         return H_SUCCESS;
1262 }
1263
1264 /*
1265  * Number of pages above which we invalidate the entire LPID rather than
1266  * flush individual pages.
1267  */
1268 static unsigned long tlb_range_flush_page_ceiling __read_mostly = 33;
1269
1270 static long do_tlb_invalidate_nested_tlb(struct kvm_vcpu *vcpu,
1271                                          unsigned long lpid,
1272                                          unsigned long pg_sizes,
1273                                          unsigned long start,
1274                                          unsigned long end)
1275 {
1276         int ret = H_P4;
1277         unsigned long addr, nr_pages;
1278         struct mmu_psize_def *def;
1279         unsigned long psize, ap, page_size;
1280         bool flush_lpid;
1281
1282         for (psize = 0; psize < MMU_PAGE_COUNT; psize++) {
1283                 def = &mmu_psize_defs[psize];
1284                 if (!(pg_sizes & def->h_rpt_pgsize))
1285                         continue;
1286
1287                 nr_pages = (end - start) >> def->shift;
1288                 flush_lpid = nr_pages > tlb_range_flush_page_ceiling;
1289                 if (flush_lpid)
1290                         return do_tlb_invalidate_nested_all(vcpu, lpid,
1291                                                         RIC_FLUSH_TLB);
1292                 addr = start;
1293                 ap = mmu_get_ap(psize);
1294                 page_size = 1UL << def->shift;
1295                 do {
1296                         ret = kvmhv_emulate_tlbie_tlb_addr(vcpu, lpid, ap,
1297                                                    get_epn(addr));
1298                         if (ret)
1299                                 return H_P4;
1300                         addr += page_size;
1301                 } while (addr < end);
1302         }
1303         return ret;
1304 }
1305
1306 /*
1307  * Performs partition-scoped invalidations for nested guests
1308  * as part of H_RPT_INVALIDATE hcall.
1309  */
1310 long do_h_rpt_invalidate_pat(struct kvm_vcpu *vcpu, unsigned long lpid,
1311                              unsigned long type, unsigned long pg_sizes,
1312                              unsigned long start, unsigned long end)
1313 {
1314         /*
1315          * If L2 lpid isn't valid, we need to return H_PARAMETER.
1316          *
1317          * However, nested KVM issues a L2 lpid flush call when creating
1318          * partition table entries for L2. This happens even before the
1319          * corresponding shadow lpid is created in HV which happens in
1320          * H_ENTER_NESTED call. Since we can't differentiate this case from
1321          * the invalid case, we ignore such flush requests and return success.
1322          */
1323         if (!__find_nested(vcpu->kvm, lpid))
1324                 return H_SUCCESS;
1325
1326         /*
1327          * A flush all request can be handled by a full lpid flush only.
1328          */
1329         if ((type & H_RPTI_TYPE_NESTED_ALL) == H_RPTI_TYPE_NESTED_ALL)
1330                 return do_tlb_invalidate_nested_all(vcpu, lpid, RIC_FLUSH_ALL);
1331
1332         /*
1333          * We don't need to handle a PWC flush like process table here,
1334          * because intermediate partition scoped table in nested guest doesn't
1335          * really have PWC. Only level we have PWC is in L0 and for nested
1336          * invalidate at L0 we always do kvm_flush_lpid() which does
1337          * radix__flush_all_lpid(). For range invalidate at any level, we
1338          * are not removing the higher level page tables and hence there is
1339          * no PWC invalidate needed.
1340          *
1341          * if (type & H_RPTI_TYPE_PWC) {
1342          *      ret = do_tlb_invalidate_nested_all(vcpu, lpid, RIC_FLUSH_PWC);
1343          *      if (ret)
1344          *              return H_P4;
1345          * }
1346          */
1347
1348         if (start == 0 && end == -1)
1349                 return do_tlb_invalidate_nested_all(vcpu, lpid, RIC_FLUSH_TLB);
1350
1351         if (type & H_RPTI_TYPE_TLB)
1352                 return do_tlb_invalidate_nested_tlb(vcpu, lpid, pg_sizes,
1353                                                     start, end);
1354         return H_SUCCESS;
1355 }
1356
1357 /* Used to convert a nested guest real address to a L1 guest real address */
1358 static int kvmhv_translate_addr_nested(struct kvm_vcpu *vcpu,
1359                                        struct kvm_nested_guest *gp,
1360                                        unsigned long n_gpa, unsigned long dsisr,
1361                                        struct kvmppc_pte *gpte_p)
1362 {
1363         u64 fault_addr, flags = dsisr & DSISR_ISSTORE;
1364         int ret;
1365
1366         ret = kvmppc_mmu_walk_radix_tree(vcpu, n_gpa, gpte_p, gp->l1_gr_to_hr,
1367                                          &fault_addr);
1368
1369         if (ret) {
1370                 /* We didn't find a pte */
1371                 if (ret == -EINVAL) {
1372                         /* Unsupported mmu config */
1373                         flags |= DSISR_UNSUPP_MMU;
1374                 } else if (ret == -ENOENT) {
1375                         /* No translation found */
1376                         flags |= DSISR_NOHPTE;
1377                 } else if (ret == -EFAULT) {
1378                         /* Couldn't access L1 real address */
1379                         flags |= DSISR_PRTABLE_FAULT;
1380                         vcpu->arch.fault_gpa = fault_addr;
1381                 } else {
1382                         /* Unknown error */
1383                         return ret;
1384                 }
1385                 goto forward_to_l1;
1386         } else {
1387                 /* We found a pte -> check permissions */
1388                 if (dsisr & DSISR_ISSTORE) {
1389                         /* Can we write? */
1390                         if (!gpte_p->may_write) {
1391                                 flags |= DSISR_PROTFAULT;
1392                                 goto forward_to_l1;
1393                         }
1394                 } else if (vcpu->arch.trap == BOOK3S_INTERRUPT_H_INST_STORAGE) {
1395                         /* Can we execute? */
1396                         if (!gpte_p->may_execute) {
1397                                 flags |= SRR1_ISI_N_G_OR_CIP;
1398                                 goto forward_to_l1;
1399                         }
1400                 } else {
1401                         /* Can we read? */
1402                         if (!gpte_p->may_read && !gpte_p->may_write) {
1403                                 flags |= DSISR_PROTFAULT;
1404                                 goto forward_to_l1;
1405                         }
1406                 }
1407         }
1408
1409         return 0;
1410
1411 forward_to_l1:
1412         vcpu->arch.fault_dsisr = flags;
1413         if (vcpu->arch.trap == BOOK3S_INTERRUPT_H_INST_STORAGE) {
1414                 vcpu->arch.shregs.msr &= SRR1_MSR_BITS;
1415                 vcpu->arch.shregs.msr |= flags;
1416         }
1417         return RESUME_HOST;
1418 }
1419
1420 static long kvmhv_handle_nested_set_rc(struct kvm_vcpu *vcpu,
1421                                        struct kvm_nested_guest *gp,
1422                                        unsigned long n_gpa,
1423                                        struct kvmppc_pte gpte,
1424                                        unsigned long dsisr)
1425 {
1426         struct kvm *kvm = vcpu->kvm;
1427         bool writing = !!(dsisr & DSISR_ISSTORE);
1428         u64 pgflags;
1429         long ret;
1430
1431         /* Are the rc bits set in the L1 partition scoped pte? */
1432         pgflags = _PAGE_ACCESSED;
1433         if (writing)
1434                 pgflags |= _PAGE_DIRTY;
1435         if (pgflags & ~gpte.rc)
1436                 return RESUME_HOST;
1437
1438         spin_lock(&kvm->mmu_lock);
1439         /* Set the rc bit in the pte of our (L0) pgtable for the L1 guest */
1440         ret = kvmppc_hv_handle_set_rc(kvm, false, writing,
1441                                       gpte.raddr, kvm->arch.lpid);
1442         if (!ret) {
1443                 ret = -EINVAL;
1444                 goto out_unlock;
1445         }
1446
1447         /* Set the rc bit in the pte of the shadow_pgtable for the nest guest */
1448         ret = kvmppc_hv_handle_set_rc(kvm, true, writing,
1449                                       n_gpa, gp->l1_lpid);
1450         if (!ret)
1451                 ret = -EINVAL;
1452         else
1453                 ret = 0;
1454
1455 out_unlock:
1456         spin_unlock(&kvm->mmu_lock);
1457         return ret;
1458 }
1459
1460 static inline int kvmppc_radix_level_to_shift(int level)
1461 {
1462         switch (level) {
1463         case 2:
1464                 return PUD_SHIFT;
1465         case 1:
1466                 return PMD_SHIFT;
1467         default:
1468                 return PAGE_SHIFT;
1469         }
1470 }
1471
1472 static inline int kvmppc_radix_shift_to_level(int shift)
1473 {
1474         if (shift == PUD_SHIFT)
1475                 return 2;
1476         if (shift == PMD_SHIFT)
1477                 return 1;
1478         if (shift == PAGE_SHIFT)
1479                 return 0;
1480         WARN_ON_ONCE(1);
1481         return 0;
1482 }
1483
1484 /* called with gp->tlb_lock held */
1485 static long int __kvmhv_nested_page_fault(struct kvm_vcpu *vcpu,
1486                                           struct kvm_nested_guest *gp)
1487 {
1488         struct kvm *kvm = vcpu->kvm;
1489         struct kvm_memory_slot *memslot;
1490         struct rmap_nested *n_rmap;
1491         struct kvmppc_pte gpte;
1492         pte_t pte, *pte_p;
1493         unsigned long mmu_seq;
1494         unsigned long dsisr = vcpu->arch.fault_dsisr;
1495         unsigned long ea = vcpu->arch.fault_dar;
1496         unsigned long *rmapp;
1497         unsigned long n_gpa, gpa, gfn, perm = 0UL;
1498         unsigned int shift, l1_shift, level;
1499         bool writing = !!(dsisr & DSISR_ISSTORE);
1500         bool kvm_ro = false;
1501         long int ret;
1502
1503         if (!gp->l1_gr_to_hr) {
1504                 kvmhv_update_ptbl_cache(gp);
1505                 if (!gp->l1_gr_to_hr)
1506                         return RESUME_HOST;
1507         }
1508
1509         /* Convert the nested guest real address into a L1 guest real address */
1510
1511         n_gpa = vcpu->arch.fault_gpa & ~0xF000000000000FFFULL;
1512         if (!(dsisr & DSISR_PRTABLE_FAULT))
1513                 n_gpa |= ea & 0xFFF;
1514         ret = kvmhv_translate_addr_nested(vcpu, gp, n_gpa, dsisr, &gpte);
1515
1516         /*
1517          * If the hardware found a translation but we don't now have a usable
1518          * translation in the l1 partition-scoped tree, remove the shadow pte
1519          * and let the guest retry.
1520          */
1521         if (ret == RESUME_HOST &&
1522             (dsisr & (DSISR_PROTFAULT | DSISR_BADACCESS | DSISR_NOEXEC_OR_G |
1523                       DSISR_BAD_COPYPASTE)))
1524                 goto inval;
1525         if (ret)
1526                 return ret;
1527
1528         /* Failed to set the reference/change bits */
1529         if (dsisr & DSISR_SET_RC) {
1530                 ret = kvmhv_handle_nested_set_rc(vcpu, gp, n_gpa, gpte, dsisr);
1531                 if (ret == RESUME_HOST)
1532                         return ret;
1533                 if (ret)
1534                         goto inval;
1535                 dsisr &= ~DSISR_SET_RC;
1536                 if (!(dsisr & (DSISR_BAD_FAULT_64S | DSISR_NOHPTE |
1537                                DSISR_PROTFAULT)))
1538                         return RESUME_GUEST;
1539         }
1540
1541         /*
1542          * We took an HISI or HDSI while we were running a nested guest which
1543          * means we have no partition scoped translation for that. This means
1544          * we need to insert a pte for the mapping into our shadow_pgtable.
1545          */
1546
1547         l1_shift = gpte.page_shift;
1548         if (l1_shift < PAGE_SHIFT) {
1549                 /* We don't support l1 using a page size smaller than our own */
1550                 pr_err("KVM: L1 guest page shift (%d) less than our own (%d)\n",
1551                         l1_shift, PAGE_SHIFT);
1552                 return -EINVAL;
1553         }
1554         gpa = gpte.raddr;
1555         gfn = gpa >> PAGE_SHIFT;
1556
1557         /* 1. Get the corresponding host memslot */
1558
1559         memslot = gfn_to_memslot(kvm, gfn);
1560         if (!memslot || (memslot->flags & KVM_MEMSLOT_INVALID)) {
1561                 if (dsisr & (DSISR_PRTABLE_FAULT | DSISR_BADACCESS)) {
1562                         /* unusual error -> reflect to the guest as a DSI */
1563                         kvmppc_core_queue_data_storage(vcpu,
1564                                         kvmppc_get_msr(vcpu) & SRR1_PREFIXED,
1565                                         ea, dsisr);
1566                         return RESUME_GUEST;
1567                 }
1568
1569                 /* passthrough of emulated MMIO case */
1570                 return kvmppc_hv_emulate_mmio(vcpu, gpa, ea, writing);
1571         }
1572         if (memslot->flags & KVM_MEM_READONLY) {
1573                 if (writing) {
1574                         /* Give the guest a DSI */
1575                         kvmppc_core_queue_data_storage(vcpu,
1576                                         kvmppc_get_msr(vcpu) & SRR1_PREFIXED,
1577                                         ea, DSISR_ISSTORE | DSISR_PROTFAULT);
1578                         return RESUME_GUEST;
1579                 }
1580                 kvm_ro = true;
1581         }
1582
1583         /* 2. Find the host pte for this L1 guest real address */
1584
1585         /* Used to check for invalidations in progress */
1586         mmu_seq = kvm->mmu_invalidate_seq;
1587         smp_rmb();
1588
1589         /* See if can find translation in our partition scoped tables for L1 */
1590         pte = __pte(0);
1591         spin_lock(&kvm->mmu_lock);
1592         pte_p = find_kvm_secondary_pte(kvm, gpa, &shift);
1593         if (!shift)
1594                 shift = PAGE_SHIFT;
1595         if (pte_p)
1596                 pte = *pte_p;
1597         spin_unlock(&kvm->mmu_lock);
1598
1599         if (!pte_present(pte) || (writing && !(pte_val(pte) & _PAGE_WRITE))) {
1600                 /* No suitable pte found -> try to insert a mapping */
1601                 ret = kvmppc_book3s_instantiate_page(vcpu, gpa, memslot,
1602                                         writing, kvm_ro, &pte, &level);
1603                 if (ret == -EAGAIN)
1604                         return RESUME_GUEST;
1605                 else if (ret)
1606                         return ret;
1607                 shift = kvmppc_radix_level_to_shift(level);
1608         }
1609         /* Align gfn to the start of the page */
1610         gfn = (gpa & ~((1UL << shift) - 1)) >> PAGE_SHIFT;
1611
1612         /* 3. Compute the pte we need to insert for nest_gpa -> host r_addr */
1613
1614         /* The permissions is the combination of the host and l1 guest ptes */
1615         perm |= gpte.may_read ? 0UL : _PAGE_READ;
1616         perm |= gpte.may_write ? 0UL : _PAGE_WRITE;
1617         perm |= gpte.may_execute ? 0UL : _PAGE_EXEC;
1618         /* Only set accessed/dirty (rc) bits if set in host and l1 guest ptes */
1619         perm |= (gpte.rc & _PAGE_ACCESSED) ? 0UL : _PAGE_ACCESSED;
1620         perm |= ((gpte.rc & _PAGE_DIRTY) && writing) ? 0UL : _PAGE_DIRTY;
1621         pte = __pte(pte_val(pte) & ~perm);
1622
1623         /* What size pte can we insert? */
1624         if (shift > l1_shift) {
1625                 u64 mask;
1626                 unsigned int actual_shift = PAGE_SHIFT;
1627                 if (PMD_SHIFT < l1_shift)
1628                         actual_shift = PMD_SHIFT;
1629                 mask = (1UL << shift) - (1UL << actual_shift);
1630                 pte = __pte(pte_val(pte) | (gpa & mask));
1631                 shift = actual_shift;
1632         }
1633         level = kvmppc_radix_shift_to_level(shift);
1634         n_gpa &= ~((1UL << shift) - 1);
1635
1636         /* 4. Insert the pte into our shadow_pgtable */
1637
1638         n_rmap = kzalloc(sizeof(*n_rmap), GFP_KERNEL);
1639         if (!n_rmap)
1640                 return RESUME_GUEST; /* Let the guest try again */
1641         n_rmap->rmap = (n_gpa & RMAP_NESTED_GPA_MASK) |
1642                 (((unsigned long) gp->l1_lpid) << RMAP_NESTED_LPID_SHIFT);
1643         rmapp = &memslot->arch.rmap[gfn - memslot->base_gfn];
1644         ret = kvmppc_create_pte(kvm, gp->shadow_pgtable, pte, n_gpa, level,
1645                                 mmu_seq, gp->shadow_lpid, rmapp, &n_rmap);
1646         kfree(n_rmap);
1647         if (ret == -EAGAIN)
1648                 ret = RESUME_GUEST;     /* Let the guest try again */
1649
1650         return ret;
1651
1652  inval:
1653         kvmhv_invalidate_shadow_pte(vcpu, gp, n_gpa, NULL);
1654         return RESUME_GUEST;
1655 }
1656
1657 long int kvmhv_nested_page_fault(struct kvm_vcpu *vcpu)
1658 {
1659         struct kvm_nested_guest *gp = vcpu->arch.nested;
1660         long int ret;
1661
1662         mutex_lock(&gp->tlb_lock);
1663         ret = __kvmhv_nested_page_fault(vcpu, gp);
1664         mutex_unlock(&gp->tlb_lock);
1665         return ret;
1666 }
1667
1668 int kvmhv_nested_next_lpid(struct kvm *kvm, int lpid)
1669 {
1670         int ret = lpid + 1;
1671
1672         spin_lock(&kvm->mmu_lock);
1673         if (!idr_get_next(&kvm->arch.kvm_nested_guest_idr, &ret))
1674                 ret = -1;
1675         spin_unlock(&kvm->mmu_lock);
1676
1677         return ret;
1678 }