arch/powerpc/kvm/e500_mmu_host.c

   1 /*
   2  * Copyright (C) 2008-2013 Freescale Semiconductor, Inc. All rights reserved.
   3  *
   4  * Author: Yu Liu, yu.liu@freescale.com
   5  *         Scott Wood, scottwood@freescale.com
   6  *         Ashish Kalra, ashish.kalra@freescale.com
   7  *         Varun Sethi, varun.sethi@freescale.com
   8  *         Alexander Graf, agraf@suse.de
   9  *
  10  * Description:
  11  * This file is based on arch/powerpc/kvm/44x_tlb.c,
  12  * by Hollis Blanchard <hollisb@us.ibm.com>.
  13  *
  14  * This program is free software; you can redistribute it and/or modify
  15  * it under the terms of the GNU General Public License, version 2, as
  16  * published by the Free Software Foundation.
  17  */
  18
  19 #include <linux/kernel.h>
  20 #include <linux/types.h>
  21 #include <linux/slab.h>
  22 #include <linux/string.h>
  23 #include <linux/kvm.h>
  24 #include <linux/kvm_host.h>
  25 #include <linux/highmem.h>
  26 #include <linux/log2.h>
  27 #include <linux/uaccess.h>
  28 #include <linux/sched.h>
  29 #include <linux/rwsem.h>
  30 #include <linux/vmalloc.h>
  31 #include <linux/hugetlb.h>
  32 #include <asm/kvm_ppc.h>
  33
  34 #include "e500.h"
  35 #include "timing.h"
  36 #include "e500_mmu_host.h"
  37
  38 #include "trace_booke.h"
  39
  40 #define to_htlb1_esel(esel) (host_tlb_params[1].entries - (esel) - 1)
  41
  42 static struct kvmppc_e500_tlb_params host_tlb_params[E500_TLB_NUM];
  43
  44 static inline unsigned int tlb1_max_shadow_size(void)
  45 {
  46         /* reserve one entry for magic page */
  47         return host_tlb_params[1].entries - tlbcam_index - 1;
  48 }
  49
  50 static inline u32 e500_shadow_mas3_attrib(u32 mas3, int usermode)
  51 {
  52         /* Mask off reserved bits. */
  53         mas3 &= MAS3_ATTRIB_MASK;
  54
  55 #ifndef CONFIG_KVM_BOOKE_HV
  56         if (!usermode) {
  57                 /* Guest is in supervisor mode,
  58                  * so we need to translate guest
  59                  * supervisor permissions into user permissions. */
  60                 mas3 &= ~E500_TLB_USER_PERM_MASK;
  61                 mas3 |= (mas3 & E500_TLB_SUPER_PERM_MASK) << 1;
  62         }
  63         mas3 |= E500_TLB_SUPER_PERM_MASK;
  64 #endif
  65         return mas3;
  66 }
  67
  68 /*
  69  * writing shadow tlb entry to host TLB
  70  */
  71 static inline void __write_host_tlbe(struct kvm_book3e_206_tlb_entry *stlbe,
  72                                      uint32_t mas0)
  73 {
  74         unsigned long flags;
  75
  76         local_irq_save(flags);
  77         mtspr(SPRN_MAS0, mas0);
  78         mtspr(SPRN_MAS1, stlbe->mas1);
  79         mtspr(SPRN_MAS2, (unsigned long)stlbe->mas2);
  80         mtspr(SPRN_MAS3, (u32)stlbe->mas7_3);
  81         mtspr(SPRN_MAS7, (u32)(stlbe->mas7_3 >> 32));
  82 #ifdef CONFIG_KVM_BOOKE_HV
  83         mtspr(SPRN_MAS8, stlbe->mas8);
  84 #endif
  85         asm volatile("isync; tlbwe" : : : "memory");
  86
  87 #ifdef CONFIG_KVM_BOOKE_HV
  88         /* Must clear mas8 for other host tlbwe's */
  89         mtspr(SPRN_MAS8, 0);
  90         isync();
  91 #endif
  92         local_irq_restore(flags);
  93
  94         trace_kvm_booke206_stlb_write(mas0, stlbe->mas8, stlbe->mas1,
  95                                       stlbe->mas2, stlbe->mas7_3);
  96 }
  97
  98 /*
  99  * Acquire a mas0 with victim hint, as if we just took a TLB miss.
 100  *
 101  * We don't care about the address we're searching for, other than that it's
 102  * in the right set and is not present in the TLB.  Using a zero PID and a
 103  * userspace address means we don't have to set and then restore MAS5, or
 104  * calculate a proper MAS6 value.
 105  */
 106 static u32 get_host_mas0(unsigned long eaddr)
 107 {
 108         unsigned long flags;
 109         u32 mas0;
 110
 111         local_irq_save(flags);
 112         mtspr(SPRN_MAS6, 0);
 113         asm volatile("tlbsx 0, %0" : : "b" (eaddr & ~CONFIG_PAGE_OFFSET));
 114         mas0 = mfspr(SPRN_MAS0);
 115         local_irq_restore(flags);
 116
 117         return mas0;
 118 }
 119
 120 /* sesel is for tlb1 only */
 121 static inline void write_host_tlbe(struct kvmppc_vcpu_e500 *vcpu_e500,
 122                 int tlbsel, int sesel, struct kvm_book3e_206_tlb_entry *stlbe)
 123 {
 124         u32 mas0;
 125
 126         if (tlbsel == 0) {
 127                 mas0 = get_host_mas0(stlbe->mas2);
 128                 __write_host_tlbe(stlbe, mas0);
 129         } else {
 130                 __write_host_tlbe(stlbe,
 131                                   MAS0_TLBSEL(1) |
 132                                   MAS0_ESEL(to_htlb1_esel(sesel)));
 133         }
 134 }
 135
 136 /* sesel is for tlb1 only */
 137 static void write_stlbe(struct kvmppc_vcpu_e500 *vcpu_e500,
 138                         struct kvm_book3e_206_tlb_entry *gtlbe,
 139                         struct kvm_book3e_206_tlb_entry *stlbe,
 140                         int stlbsel, int sesel)
 141 {
 142         int stid;
 143
 144         preempt_disable();
 145         stid = kvmppc_e500_get_tlb_stid(&vcpu_e500->vcpu, gtlbe);
 146
 147         stlbe->mas1 |= MAS1_TID(stid);
 148         write_host_tlbe(vcpu_e500, stlbsel, sesel, stlbe);
 149         preempt_enable();
 150 }
 151
 152 #ifdef CONFIG_KVM_E500V2
 153 /* XXX should be a hook in the gva2hpa translation */
 154 void kvmppc_map_magic(struct kvm_vcpu *vcpu)
 155 {
 156         struct kvmppc_vcpu_e500 *vcpu_e500 = to_e500(vcpu);
 157         struct kvm_book3e_206_tlb_entry magic;
 158         ulong shared_page = ((ulong)vcpu->arch.shared) & PAGE_MASK;
 159         unsigned int stid;
 160         pfn_t pfn;
 161
 162         pfn = (pfn_t)virt_to_phys((void *)shared_page) >> PAGE_SHIFT;
 163         get_page(pfn_to_page(pfn));
 164
 165         preempt_disable();
 166         stid = kvmppc_e500_get_sid(vcpu_e500, 0, 0, 0, 0);
 167
 168         magic.mas1 = MAS1_VALID | MAS1_TS | MAS1_TID(stid) |
 169                      MAS1_TSIZE(BOOK3E_PAGESZ_4K);
 170         magic.mas2 = vcpu->arch.magic_page_ea | MAS2_M;
 171         magic.mas7_3 = ((u64)pfn << PAGE_SHIFT) |
 172                        MAS3_SW | MAS3_SR | MAS3_UW | MAS3_UR;
 173         magic.mas8 = 0;
 174
 175         __write_host_tlbe(&magic, MAS0_TLBSEL(1) | MAS0_ESEL(tlbcam_index));
 176         preempt_enable();
 177 }
 178 #endif
 179
 180 void inval_gtlbe_on_host(struct kvmppc_vcpu_e500 *vcpu_e500, int tlbsel,
 181                          int esel)
 182 {
 183         struct kvm_book3e_206_tlb_entry *gtlbe =
 184                 get_entry(vcpu_e500, tlbsel, esel);
 185         struct tlbe_ref *ref = &vcpu_e500->gtlb_priv[tlbsel][esel].ref;
 186
 187         /* Don't bother with unmapped entries */
 188         if (!(ref->flags & E500_TLB_VALID)) {
 189                 WARN(ref->flags & (E500_TLB_BITMAP | E500_TLB_TLB0),
 190                      "%s: flags %x\n", __func__, ref->flags);
 191                 WARN_ON(tlbsel == 1 && vcpu_e500->g2h_tlb1_map[esel]);
 192         }
 193
 194         if (tlbsel == 1 && ref->flags & E500_TLB_BITMAP) {
 195                 u64 tmp = vcpu_e500->g2h_tlb1_map[esel];
 196                 int hw_tlb_indx;
 197                 unsigned long flags;
 198
 199                 local_irq_save(flags);
 200                 while (tmp) {
 201                         hw_tlb_indx = __ilog2_u64(tmp & -tmp);
 202                         mtspr(SPRN_MAS0,
 203                               MAS0_TLBSEL(1) |
 204                               MAS0_ESEL(to_htlb1_esel(hw_tlb_indx)));
 205                         mtspr(SPRN_MAS1, 0);
 206                         asm volatile("tlbwe");
 207                         vcpu_e500->h2g_tlb1_rmap[hw_tlb_indx] = 0;
 208                         tmp &= tmp - 1;
 209                 }
 210                 mb();
 211                 vcpu_e500->g2h_tlb1_map[esel] = 0;
 212                 ref->flags &= ~(E500_TLB_BITMAP | E500_TLB_VALID);
 213                 local_irq_restore(flags);
 214         }
 215
 216         if (tlbsel == 1 && ref->flags & E500_TLB_TLB0) {
 217                 /*
 218                  * TLB1 entry is backed by 4k pages. This should happen
 219                  * rarely and is not worth optimizing. Invalidate everything.
 220                  */
 221                 kvmppc_e500_tlbil_all(vcpu_e500);
 222                 ref->flags &= ~(E500_TLB_TLB0 | E500_TLB_VALID);
 223         }
 224
 225         /*
 226          * If TLB entry is still valid then it's a TLB0 entry, and thus
 227          * backed by at most one host tlbe per shadow pid
 228          */
 229         if (ref->flags & E500_TLB_VALID)
 230                 kvmppc_e500_tlbil_one(vcpu_e500, gtlbe);
 231
 232         /* Mark the TLB as not backed by the host anymore */
 233         ref->flags = 0;
 234 }
 235
 236 static inline int tlbe_is_writable(struct kvm_book3e_206_tlb_entry *tlbe)
 237 {
 238         return tlbe->mas7_3 & (MAS3_SW|MAS3_UW);
 239 }
 240
 241 static inline void kvmppc_e500_ref_setup(struct tlbe_ref *ref,
 242                                          struct kvm_book3e_206_tlb_entry *gtlbe,
 243                                          pfn_t pfn, unsigned int wimg)
 244 {
 245         ref->pfn = pfn;
 246         ref->flags = E500_TLB_VALID;
 247
 248         /* Use guest supplied MAS2_G and MAS2_E */
 249         ref->flags |= (gtlbe->mas2 & MAS2_ATTRIB_MASK) | wimg;
 250
 251         /* Mark the page accessed */
 252         kvm_set_pfn_accessed(pfn);
 253
 254         if (tlbe_is_writable(gtlbe))
 255                 kvm_set_pfn_dirty(pfn);
 256 }
 257
 258 static inline void kvmppc_e500_ref_release(struct tlbe_ref *ref)
 259 {
 260         if (ref->flags & E500_TLB_VALID) {
 261                 /* FIXME: don't log bogus pfn for TLB1 */
 262                 trace_kvm_booke206_ref_release(ref->pfn, ref->flags);
 263                 ref->flags = 0;
 264         }
 265 }
 266
 267 static void clear_tlb1_bitmap(struct kvmppc_vcpu_e500 *vcpu_e500)
 268 {
 269         if (vcpu_e500->g2h_tlb1_map)
 270                 memset(vcpu_e500->g2h_tlb1_map, 0,
 271                        sizeof(u64) * vcpu_e500->gtlb_params[1].entries);
 272         if (vcpu_e500->h2g_tlb1_rmap)
 273                 memset(vcpu_e500->h2g_tlb1_rmap, 0,
 274                        sizeof(unsigned int) * host_tlb_params[1].entries);
 275 }
 276
 277 static void clear_tlb_privs(struct kvmppc_vcpu_e500 *vcpu_e500)
 278 {
 279         int tlbsel;
 280         int i;
 281
 282         for (tlbsel = 0; tlbsel <= 1; tlbsel++) {
 283                 for (i = 0; i < vcpu_e500->gtlb_params[tlbsel].entries; i++) {
 284                         struct tlbe_ref *ref =
 285                                 &vcpu_e500->gtlb_priv[tlbsel][i].ref;
 286                         kvmppc_e500_ref_release(ref);
 287                 }
 288         }
 289 }
 290
 291 void kvmppc_core_flush_tlb(struct kvm_vcpu *vcpu)
 292 {
 293         struct kvmppc_vcpu_e500 *vcpu_e500 = to_e500(vcpu);
 294         kvmppc_e500_tlbil_all(vcpu_e500);
 295         clear_tlb_privs(vcpu_e500);
 296         clear_tlb1_bitmap(vcpu_e500);
 297 }
 298
 299 /* TID must be supplied by the caller */
 300 static void kvmppc_e500_setup_stlbe(
 301         struct kvm_vcpu *vcpu,
 302         struct kvm_book3e_206_tlb_entry *gtlbe,
 303         int tsize, struct tlbe_ref *ref, u64 gvaddr,
 304         struct kvm_book3e_206_tlb_entry *stlbe)
 305 {
 306         pfn_t pfn = ref->pfn;
 307         u32 pr = vcpu->arch.shared->msr & MSR_PR;
 308
 309         BUG_ON(!(ref->flags & E500_TLB_VALID));
 310
 311         /* Force IPROT=0 for all guest mappings. */
 312         stlbe->mas1 = MAS1_TSIZE(tsize) | get_tlb_sts(gtlbe) | MAS1_VALID;
 313         stlbe->mas2 = (gvaddr & MAS2_EPN) | (ref->flags & E500_TLB_MAS2_ATTR);
 314         stlbe->mas7_3 = ((u64)pfn << PAGE_SHIFT) |
 315                         e500_shadow_mas3_attrib(gtlbe->mas7_3, pr);
 316
 317 #ifdef CONFIG_KVM_BOOKE_HV
 318         stlbe->mas8 = MAS8_TGS | vcpu->kvm->arch.lpid;
 319 #endif
 320 }
 321
 322 static inline int kvmppc_e500_shadow_map(struct kvmppc_vcpu_e500 *vcpu_e500,
 323         u64 gvaddr, gfn_t gfn, struct kvm_book3e_206_tlb_entry *gtlbe,
 324         int tlbsel, struct kvm_book3e_206_tlb_entry *stlbe,
 325         struct tlbe_ref *ref)
 326 {
 327         struct kvm_memory_slot *slot;
 328         unsigned long pfn = 0; /* silence GCC warning */
 329         unsigned long hva;
 330         int pfnmap = 0;
 331         int tsize = BOOK3E_PAGESZ_4K;
 332         int ret = 0;
 333         unsigned long mmu_seq;
 334         struct kvm *kvm = vcpu_e500->vcpu.kvm;
 335         unsigned long tsize_pages = 0;
 336         pte_t *ptep;
 337         unsigned int wimg = 0;
 338         pgd_t *pgdir;
 339
 340         /* used to check for invalidations in progress */
 341         mmu_seq = kvm->mmu_notifier_seq;
 342         smp_rmb();
 343
 344         /*
 345          * Translate guest physical to true physical, acquiring
 346          * a page reference if it is normal, non-reserved memory.
 347          *
 348          * gfn_to_memslot() must succeed because otherwise we wouldn't
 349          * have gotten this far.  Eventually we should just pass the slot
 350          * pointer through from the first lookup.
 351          */
 352         slot = gfn_to_memslot(vcpu_e500->vcpu.kvm, gfn);
 353         hva = gfn_to_hva_memslot(slot, gfn);
 354
 355         if (tlbsel == 1) {
 356                 struct vm_area_struct *vma;
 357                 down_read(&current->mm->mmap_sem);
 358
 359                 vma = find_vma(current->mm, hva);
 360                 if (vma && hva >= vma->vm_start &&
 361                     (vma->vm_flags & VM_PFNMAP)) {
 362                         /*
 363                          * This VMA is a physically contiguous region (e.g.
 364                          * /dev/mem) that bypasses normal Linux page
 365                          * management.  Find the overlap between the
 366                          * vma and the memslot.
 367                          */
 368
 369                         unsigned long start, end;
 370                         unsigned long slot_start, slot_end;
 371
 372                         pfnmap = 1;
 373
 374                         start = vma->vm_pgoff;
 375                         end = start +
 376                               ((vma->vm_end - vma->vm_start) >> PAGE_SHIFT);
 377
 378                         pfn = start + ((hva - vma->vm_start) >> PAGE_SHIFT);
 379
 380                         slot_start = pfn - (gfn - slot->base_gfn);
 381                         slot_end = slot_start + slot->npages;
 382
 383                         if (start < slot_start)
 384                                 start = slot_start;
 385                         if (end > slot_end)
 386                                 end = slot_end;
 387
 388                         tsize = (gtlbe->mas1 & MAS1_TSIZE_MASK) >>
 389                                 MAS1_TSIZE_SHIFT;
 390
 391                         /*
 392                          * e500 doesn't implement the lowest tsize bit,
 393                          * or 1K pages.
 394                          */
 395                         tsize = max(BOOK3E_PAGESZ_4K, tsize & ~1);
 396
 397                         /*
 398                          * Now find the largest tsize (up to what the guest
 399                          * requested) that will cover gfn, stay within the
 400                          * range, and for which gfn and pfn are mutually
 401                          * aligned.
 402                          */
 403
 404                         for (; tsize > BOOK3E_PAGESZ_4K; tsize -= 2) {
 405                                 unsigned long gfn_start, gfn_end;
 406                                 tsize_pages = 1 << (tsize - 2);
 407
 408                                 gfn_start = gfn & ~(tsize_pages - 1);
 409                                 gfn_end = gfn_start + tsize_pages;
 410
 411                                 if (gfn_start + pfn - gfn < start)
 412                                         continue;
 413                                 if (gfn_end + pfn - gfn > end)
 414                                         continue;
 415                                 if ((gfn & (tsize_pages - 1)) !=
 416                                     (pfn & (tsize_pages - 1)))
 417                                         continue;
 418
 419                                 gvaddr &= ~((tsize_pages << PAGE_SHIFT) - 1);
 420                                 pfn &= ~(tsize_pages - 1);
 421                                 break;
 422                         }
 423                 } else if (vma && hva >= vma->vm_start &&
 424                            (vma->vm_flags & VM_HUGETLB)) {
 425                         unsigned long psize = vma_kernel_pagesize(vma);
 426
 427                         tsize = (gtlbe->mas1 & MAS1_TSIZE_MASK) >>
 428                                 MAS1_TSIZE_SHIFT;
 429
 430                         /*
 431                          * Take the largest page size that satisfies both host
 432                          * and guest mapping
 433                          */
 434                         tsize = min(__ilog2(psize) - 10, tsize);
 435
 436                         /*
 437                          * e500 doesn't implement the lowest tsize bit,
 438                          * or 1K pages.
 439                          */
 440                         tsize = max(BOOK3E_PAGESZ_4K, tsize & ~1);
 441                 }
 442
 443                 up_read(&current->mm->mmap_sem);
 444         }
 445
 446         if (likely(!pfnmap)) {
 447                 tsize_pages = 1 << (tsize + 10 - PAGE_SHIFT);
 448                 pfn = gfn_to_pfn_memslot(slot, gfn);
 449                 if (is_error_noslot_pfn(pfn)) {
 450                         if (printk_ratelimit())
 451                                 pr_err("%s: real page not found for gfn %lx\n",
 452                                        __func__, (long)gfn);
 453                         return -EINVAL;
 454                 }
 455
 456                 /* Align guest and physical address to page map boundaries */
 457                 pfn &= ~(tsize_pages - 1);
 458                 gvaddr &= ~((tsize_pages << PAGE_SHIFT) - 1);
 459         }
 460
 461         spin_lock(&kvm->mmu_lock);
 462         if (mmu_notifier_retry(kvm, mmu_seq)) {
 463                 ret = -EAGAIN;
 464                 goto out;
 465         }
 466
 467
 468         pgdir = vcpu_e500->vcpu.arch.pgdir;
 469         ptep = lookup_linux_ptep(pgdir, hva, &tsize_pages);
 470         if (pte_present(*ptep))
 471                 wimg = (*ptep >> PTE_WIMGE_SHIFT) & MAS2_WIMGE_MASK;
 472         else {
 473                 if (printk_ratelimit())
 474                         pr_err("%s: pte not present: gfn %lx, pfn %lx\n",
 475                                 __func__, (long)gfn, pfn);
 476                 return -EINVAL;
 477         }
 478         kvmppc_e500_ref_setup(ref, gtlbe, pfn, wimg);
 479
 480         kvmppc_e500_setup_stlbe(&vcpu_e500->vcpu, gtlbe, tsize,
 481                                 ref, gvaddr, stlbe);
 482
 483         /* Clear i-cache for new pages */
 484         kvmppc_mmu_flush_icache(pfn);
 485
 486 out:
 487         spin_unlock(&kvm->mmu_lock);
 488
 489         /* Drop refcount on page, so that mmu notifiers can clear it */
 490         kvm_release_pfn_clean(pfn);
 491
 492         return ret;
 493 }
 494
 495 /* XXX only map the one-one case, for now use TLB0 */
 496 static int kvmppc_e500_tlb0_map(struct kvmppc_vcpu_e500 *vcpu_e500, int esel,
 497                                 struct kvm_book3e_206_tlb_entry *stlbe)
 498 {
 499         struct kvm_book3e_206_tlb_entry *gtlbe;
 500         struct tlbe_ref *ref;
 501         int stlbsel = 0;
 502         int sesel = 0;
 503         int r;
 504
 505         gtlbe = get_entry(vcpu_e500, 0, esel);
 506         ref = &vcpu_e500->gtlb_priv[0][esel].ref;
 507
 508         r = kvmppc_e500_shadow_map(vcpu_e500, get_tlb_eaddr(gtlbe),
 509                         get_tlb_raddr(gtlbe) >> PAGE_SHIFT,
 510                         gtlbe, 0, stlbe, ref);
 511         if (r)
 512                 return r;
 513
 514         write_stlbe(vcpu_e500, gtlbe, stlbe, stlbsel, sesel);
 515
 516         return 0;
 517 }
 518
 519 static int kvmppc_e500_tlb1_map_tlb1(struct kvmppc_vcpu_e500 *vcpu_e500,
 520                                      struct tlbe_ref *ref,
 521                                      int esel)
 522 {
 523         unsigned int sesel = vcpu_e500->host_tlb1_nv++;
 524
 525         if (unlikely(vcpu_e500->host_tlb1_nv >= tlb1_max_shadow_size()))
 526                 vcpu_e500->host_tlb1_nv = 0;
 527
 528         if (vcpu_e500->h2g_tlb1_rmap[sesel]) {
 529                 unsigned int idx = vcpu_e500->h2g_tlb1_rmap[sesel] - 1;
 530                 vcpu_e500->g2h_tlb1_map[idx] &= ~(1ULL << sesel);
 531         }
 532
 533         vcpu_e500->gtlb_priv[1][esel].ref.flags |= E500_TLB_BITMAP;
 534         vcpu_e500->g2h_tlb1_map[esel] |= (u64)1 << sesel;
 535         vcpu_e500->h2g_tlb1_rmap[sesel] = esel + 1;
 536         WARN_ON(!(ref->flags & E500_TLB_VALID));
 537
 538         return sesel;
 539 }
 540
 541 /* Caller must ensure that the specified guest TLB entry is safe to insert into
 542  * the shadow TLB. */
 543 /* For both one-one and one-to-many */
 544 static int kvmppc_e500_tlb1_map(struct kvmppc_vcpu_e500 *vcpu_e500,
 545                 u64 gvaddr, gfn_t gfn, struct kvm_book3e_206_tlb_entry *gtlbe,
 546                 struct kvm_book3e_206_tlb_entry *stlbe, int esel)
 547 {
 548         struct tlbe_ref *ref = &vcpu_e500->gtlb_priv[1][esel].ref;
 549         int sesel;
 550         int r;
 551
 552         r = kvmppc_e500_shadow_map(vcpu_e500, gvaddr, gfn, gtlbe, 1, stlbe,
 553                                    ref);
 554         if (r)
 555                 return r;
 556
 557         /* Use TLB0 when we can only map a page with 4k */
 558         if (get_tlb_tsize(stlbe) == BOOK3E_PAGESZ_4K) {
 559                 vcpu_e500->gtlb_priv[1][esel].ref.flags |= E500_TLB_TLB0;
 560                 write_stlbe(vcpu_e500, gtlbe, stlbe, 0, 0);
 561                 return 0;
 562         }
 563
 564         /* Otherwise map into TLB1 */
 565         sesel = kvmppc_e500_tlb1_map_tlb1(vcpu_e500, ref, esel);
 566         write_stlbe(vcpu_e500, gtlbe, stlbe, 1, sesel);
 567
 568         return 0;
 569 }
 570
 571 void kvmppc_mmu_map(struct kvm_vcpu *vcpu, u64 eaddr, gpa_t gpaddr,
 572                     unsigned int index)
 573 {
 574         struct kvmppc_vcpu_e500 *vcpu_e500 = to_e500(vcpu);
 575         struct tlbe_priv *priv;
 576         struct kvm_book3e_206_tlb_entry *gtlbe, stlbe;
 577         int tlbsel = tlbsel_of(index);
 578         int esel = esel_of(index);
 579
 580         gtlbe = get_entry(vcpu_e500, tlbsel, esel);
 581
 582         switch (tlbsel) {
 583         case 0:
 584                 priv = &vcpu_e500->gtlb_priv[tlbsel][esel];
 585
 586                 /* Triggers after clear_tlb_privs or on initial mapping */
 587                 if (!(priv->ref.flags & E500_TLB_VALID)) {
 588                         kvmppc_e500_tlb0_map(vcpu_e500, esel, &stlbe);
 589                 } else {
 590                         kvmppc_e500_setup_stlbe(vcpu, gtlbe, BOOK3E_PAGESZ_4K,
 591                                                 &priv->ref, eaddr, &stlbe);
 592                         write_stlbe(vcpu_e500, gtlbe, &stlbe, 0, 0);
 593                 }
 594                 break;
 595
 596         case 1: {
 597                 gfn_t gfn = gpaddr >> PAGE_SHIFT;
 598                 kvmppc_e500_tlb1_map(vcpu_e500, eaddr, gfn, gtlbe, &stlbe,
 599                                      esel);
 600                 break;
 601         }
 602
 603         default:
 604                 BUG();
 605                 break;
 606         }
 607 }
 608
 609 /************* MMU Notifiers *************/
 610
 611 int kvm_unmap_hva(struct kvm *kvm, unsigned long hva)
 612 {
 613         trace_kvm_unmap_hva(hva);
 614
 615         /*
 616          * Flush all shadow tlb entries everywhere. This is slow, but
 617          * we are 100% sure that we catch the to be unmapped page
 618          */
 619         kvm_flush_remote_tlbs(kvm);
 620
 621         return 0;
 622 }
 623
 624 int kvm_unmap_hva_range(struct kvm *kvm, unsigned long start, unsigned long end)
 625 {
 626         /* kvm_unmap_hva flushes everything anyways */
 627         kvm_unmap_hva(kvm, start);
 628
 629         return 0;
 630 }
 631
 632 int kvm_age_hva(struct kvm *kvm, unsigned long hva)
 633 {
 634         /* XXX could be more clever ;) */
 635         return 0;
 636 }
 637
 638 int kvm_test_age_hva(struct kvm *kvm, unsigned long hva)
 639 {
 640         /* XXX could be more clever ;) */
 641         return 0;
 642 }
 643
 644 void kvm_set_spte_hva(struct kvm *kvm, unsigned long hva, pte_t pte)
 645 {
 646         /* The page will get remapped properly on its next fault */
 647         kvm_unmap_hva(kvm, hva);
 648 }
 649
 650 /*****************************************/
 651
 652 int e500_mmu_host_init(struct kvmppc_vcpu_e500 *vcpu_e500)
 653 {
 654         host_tlb_params[0].entries = mfspr(SPRN_TLB0CFG) & TLBnCFG_N_ENTRY;
 655         host_tlb_params[1].entries = mfspr(SPRN_TLB1CFG) & TLBnCFG_N_ENTRY;
 656
 657         /*
 658          * This should never happen on real e500 hardware, but is
 659          * architecturally possible -- e.g. in some weird nested
 660          * virtualization case.
 661          */
 662         if (host_tlb_params[0].entries == 0 ||
 663             host_tlb_params[1].entries == 0) {
 664                 pr_err("%s: need to know host tlb size\n", __func__);
 665                 return -ENODEV;
 666         }
 667
 668         host_tlb_params[0].ways = (mfspr(SPRN_TLB0CFG) & TLBnCFG_ASSOC) >>
 669                                   TLBnCFG_ASSOC_SHIFT;
 670         host_tlb_params[1].ways = host_tlb_params[1].entries;
 671
 672         if (!is_power_of_2(host_tlb_params[0].entries) ||
 673             !is_power_of_2(host_tlb_params[0].ways) ||
 674             host_tlb_params[0].entries < host_tlb_params[0].ways ||
 675             host_tlb_params[0].ways == 0) {
 676                 pr_err("%s: bad tlb0 host config: %u entries %u ways\n",
 677                        __func__, host_tlb_params[0].entries,
 678                        host_tlb_params[0].ways);
 679                 return -ENODEV;
 680         }
 681
 682         host_tlb_params[0].sets =
 683                 host_tlb_params[0].entries / host_tlb_params[0].ways;
 684         host_tlb_params[1].sets = 1;
 685
 686         vcpu_e500->h2g_tlb1_rmap = kzalloc(sizeof(unsigned int) *
 687                                            host_tlb_params[1].entries,
 688                                            GFP_KERNEL);
 689         if (!vcpu_e500->h2g_tlb1_rmap)
 690                 return -EINVAL;
 691
 692         return 0;
 693 }
 694
 695 void e500_mmu_host_uninit(struct kvmppc_vcpu_e500 *vcpu_e500)
 696 {
 697         kfree(vcpu_e500->h2g_tlb1_rmap);
 698 }