arch/powerpc/kvm/book3s_64_mmu_host.c

   1 /*
   2  * Copyright (C) 2009 SUSE Linux Products GmbH. All rights reserved.
   3  *
   4  * Authors:
   5  *     Alexander Graf <agraf@suse.de>
   6  *     Kevin Wolf <mail@kevin-wolf.de>
   7  *
   8  * This program is free software; you can redistribute it and/or modify
   9  * it under the terms of the GNU General Public License, version 2, as
  10  * published by the Free Software Foundation.
  11  *
  12  * This program is distributed in the hope that it will be useful,
  13  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  14  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  15  * GNU General Public License for more details.
  16  *
  17  * You should have received a copy of the GNU General Public License
  18  * along with this program; if not, write to the Free Software
  19  * Foundation, 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301, USA.
  20  */
  21
  22 #include <linux/kvm_host.h>
  23
  24 #include <asm/kvm_ppc.h>
  25 #include <asm/kvm_book3s.h>
  26 #include <asm/mmu-hash64.h>
  27 #include <asm/machdep.h>
  28 #include <asm/mmu_context.h>
  29 #include <asm/hw_irq.h>
  30
  31 #define PTE_SIZE 12
  32 #define VSID_ALL 0
  33
  34 /* #define DEBUG_MMU */
  35 /* #define DEBUG_SLB */
  36
  37 #ifdef DEBUG_MMU
  38 #define dprintk_mmu(a, ...) printk(KERN_INFO a, __VA_ARGS__)
  39 #else
  40 #define dprintk_mmu(a, ...) do { } while(0)
  41 #endif
  42
  43 #ifdef DEBUG_SLB
  44 #define dprintk_slb(a, ...) printk(KERN_INFO a, __VA_ARGS__)
  45 #else
  46 #define dprintk_slb(a, ...) do { } while(0)
  47 #endif
  48
  49 static void invalidate_pte(struct hpte_cache *pte)
  50 {
  51         dprintk_mmu("KVM: Flushing SPT %d: 0x%llx (0x%llx) -> 0x%llx\n",
  52                     i, pte->pte.eaddr, pte->pte.vpage, pte->host_va);
  53
  54         ppc_md.hpte_invalidate(pte->slot, pte->host_va,
  55                                MMU_PAGE_4K, MMU_SEGSIZE_256M,
  56                                false);
  57         pte->host_va = 0;
  58         kvm_release_pfn_dirty(pte->pfn);
  59 }
  60
  61 void kvmppc_mmu_pte_flush(struct kvm_vcpu *vcpu, u64 guest_ea, u64 ea_mask)
  62 {
  63         int i;
  64
  65         dprintk_mmu("KVM: Flushing %d Shadow PTEs: 0x%llx & 0x%llx\n",
  66                     vcpu->arch.hpte_cache_offset, guest_ea, ea_mask);
  67         BUG_ON(vcpu->arch.hpte_cache_offset > HPTEG_CACHE_NUM);
  68
  69         guest_ea &= ea_mask;
  70         for (i = 0; i < vcpu->arch.hpte_cache_offset; i++) {
  71                 struct hpte_cache *pte;
  72
  73                 pte = &vcpu->arch.hpte_cache[i];
  74                 if (!pte->host_va)
  75                         continue;
  76
  77                 if ((pte->pte.eaddr & ea_mask) == guest_ea) {
  78                         invalidate_pte(pte);
  79                 }
  80         }
  81
  82         /* Doing a complete flush -> start from scratch */
  83         if (!ea_mask)
  84                 vcpu->arch.hpte_cache_offset = 0;
  85 }
  86
  87 void kvmppc_mmu_pte_vflush(struct kvm_vcpu *vcpu, u64 guest_vp, u64 vp_mask)
  88 {
  89         int i;
  90
  91         dprintk_mmu("KVM: Flushing %d Shadow vPTEs: 0x%llx & 0x%llx\n",
  92                     vcpu->arch.hpte_cache_offset, guest_vp, vp_mask);
  93         BUG_ON(vcpu->arch.hpte_cache_offset > HPTEG_CACHE_NUM);
  94
  95         guest_vp &= vp_mask;
  96         for (i = 0; i < vcpu->arch.hpte_cache_offset; i++) {
  97                 struct hpte_cache *pte;
  98
  99                 pte = &vcpu->arch.hpte_cache[i];
 100                 if (!pte->host_va)
 101                         continue;
 102
 103                 if ((pte->pte.vpage & vp_mask) == guest_vp) {
 104                         invalidate_pte(pte);
 105                 }
 106         }
 107 }
 108
 109 void kvmppc_mmu_pte_pflush(struct kvm_vcpu *vcpu, u64 pa_start, u64 pa_end)
 110 {
 111         int i;
 112
 113         dprintk_mmu("KVM: Flushing %d Shadow pPTEs: 0x%llx & 0x%llx\n",
 114                     vcpu->arch.hpte_cache_offset, guest_pa, pa_mask);
 115         BUG_ON(vcpu->arch.hpte_cache_offset > HPTEG_CACHE_NUM);
 116
 117         for (i = 0; i < vcpu->arch.hpte_cache_offset; i++) {
 118                 struct hpte_cache *pte;
 119
 120                 pte = &vcpu->arch.hpte_cache[i];
 121                 if (!pte->host_va)
 122                         continue;
 123
 124                 if ((pte->pte.raddr >= pa_start) &&
 125                     (pte->pte.raddr < pa_end)) {
 126                         invalidate_pte(pte);
 127                 }
 128         }
 129 }
 130
 131 struct kvmppc_pte *kvmppc_mmu_find_pte(struct kvm_vcpu *vcpu, u64 ea, bool data)
 132 {
 133         int i;
 134         u64 guest_vp;
 135
 136         guest_vp = vcpu->arch.mmu.ea_to_vp(vcpu, ea, false);
 137         for (i=0; i<vcpu->arch.hpte_cache_offset; i++) {
 138                 struct hpte_cache *pte;
 139
 140                 pte = &vcpu->arch.hpte_cache[i];
 141                 if (!pte->host_va)
 142                         continue;
 143
 144                 if (pte->pte.vpage == guest_vp)
 145                         return &pte->pte;
 146         }
 147
 148         return NULL;
 149 }
 150
 151 static int kvmppc_mmu_hpte_cache_next(struct kvm_vcpu *vcpu)
 152 {
 153         if (vcpu->arch.hpte_cache_offset == HPTEG_CACHE_NUM)
 154                 kvmppc_mmu_pte_flush(vcpu, 0, 0);
 155
 156         return vcpu->arch.hpte_cache_offset++;
 157 }
 158
 159 /* We keep 512 gvsid->hvsid entries, mapping the guest ones to the array using
 160  * a hash, so we don't waste cycles on looping */
 161 static u16 kvmppc_sid_hash(struct kvm_vcpu *vcpu, u64 gvsid)
 162 {
 163         return (u16)(((gvsid >> (SID_MAP_BITS * 7)) & SID_MAP_MASK) ^
 164                      ((gvsid >> (SID_MAP_BITS * 6)) & SID_MAP_MASK) ^
 165                      ((gvsid >> (SID_MAP_BITS * 5)) & SID_MAP_MASK) ^
 166                      ((gvsid >> (SID_MAP_BITS * 4)) & SID_MAP_MASK) ^
 167                      ((gvsid >> (SID_MAP_BITS * 3)) & SID_MAP_MASK) ^
 168                      ((gvsid >> (SID_MAP_BITS * 2)) & SID_MAP_MASK) ^
 169                      ((gvsid >> (SID_MAP_BITS * 1)) & SID_MAP_MASK) ^
 170                      ((gvsid >> (SID_MAP_BITS * 0)) & SID_MAP_MASK));
 171 }
 172
 173
 174 static struct kvmppc_sid_map *find_sid_vsid(struct kvm_vcpu *vcpu, u64 gvsid)
 175 {
 176         struct kvmppc_sid_map *map;
 177         u16 sid_map_mask;
 178
 179         if (vcpu->arch.msr & MSR_PR)
 180                 gvsid |= VSID_PR;
 181
 182         sid_map_mask = kvmppc_sid_hash(vcpu, gvsid);
 183         map = &to_book3s(vcpu)->sid_map[sid_map_mask];
 184         if (map->guest_vsid == gvsid) {
 185                 dprintk_slb("SLB: Searching 0x%llx -> 0x%llx\n",
 186                             gvsid, map->host_vsid);
 187                 return map;
 188         }
 189
 190         map = &to_book3s(vcpu)->sid_map[SID_MAP_MASK - sid_map_mask];
 191         if (map->guest_vsid == gvsid) {
 192                 dprintk_slb("SLB: Searching 0x%llx -> 0x%llx\n",
 193                             gvsid, map->host_vsid);
 194                 return map;
 195         }
 196
 197         dprintk_slb("SLB: Searching 0x%llx -> not found\n", gvsid);
 198         return NULL;
 199 }
 200
 201 int kvmppc_mmu_map_page(struct kvm_vcpu *vcpu, struct kvmppc_pte *orig_pte)
 202 {
 203         pfn_t hpaddr;
 204         ulong hash, hpteg, va;
 205         u64 vsid;
 206         int ret;
 207         int rflags = 0x192;
 208         int vflags = 0;
 209         int attempt = 0;
 210         struct kvmppc_sid_map *map;
 211
 212         /* Get host physical address for gpa */
 213         hpaddr = gfn_to_pfn(vcpu->kvm, orig_pte->raddr >> PAGE_SHIFT);
 214         if (kvm_is_error_hva(hpaddr)) {
 215                 printk(KERN_INFO "Couldn't get guest page for gfn %llx!\n", orig_pte->eaddr);
 216                 return -EINVAL;
 217         }
 218         hpaddr <<= PAGE_SHIFT;
 219 #if PAGE_SHIFT == 12
 220 #elif PAGE_SHIFT == 16
 221         hpaddr |= orig_pte->raddr & 0xf000;
 222 #else
 223 #error Unknown page size
 224 #endif
 225
 226         /* and write the mapping ea -> hpa into the pt */
 227         vcpu->arch.mmu.esid_to_vsid(vcpu, orig_pte->eaddr >> SID_SHIFT, &vsid);
 228         map = find_sid_vsid(vcpu, vsid);
 229         if (!map) {
 230                 kvmppc_mmu_map_segment(vcpu, orig_pte->eaddr);
 231                 map = find_sid_vsid(vcpu, vsid);
 232         }
 233         BUG_ON(!map);
 234
 235         vsid = map->host_vsid;
 236         va = hpt_va(orig_pte->eaddr, vsid, MMU_SEGSIZE_256M);
 237
 238         if (!orig_pte->may_write)
 239                 rflags |= HPTE_R_PP;
 240         else
 241                 mark_page_dirty(vcpu->kvm, orig_pte->raddr >> PAGE_SHIFT);
 242
 243         if (!orig_pte->may_execute)
 244                 rflags |= HPTE_R_N;
 245
 246         hash = hpt_hash(va, PTE_SIZE, MMU_SEGSIZE_256M);
 247
 248 map_again:
 249         hpteg = ((hash & htab_hash_mask) * HPTES_PER_GROUP);
 250
 251         /* In case we tried normal mapping already, let's nuke old entries */
 252         if (attempt > 1)
 253                 if (ppc_md.hpte_remove(hpteg) < 0)
 254                         return -1;
 255
 256         ret = ppc_md.hpte_insert(hpteg, va, hpaddr, rflags, vflags, MMU_PAGE_4K, MMU_SEGSIZE_256M);
 257
 258         if (ret < 0) {
 259                 /* If we couldn't map a primary PTE, try a secondary */
 260 #ifdef USE_SECONDARY
 261                 hash = ~hash;
 262                 attempt++;
 263                 if (attempt % 2)
 264                         vflags = HPTE_V_SECONDARY;
 265                 else
 266                         vflags = 0;
 267 #else
 268                 attempt = 2;
 269 #endif
 270                 goto map_again;
 271         } else {
 272                 int hpte_id = kvmppc_mmu_hpte_cache_next(vcpu);
 273                 struct hpte_cache *pte = &vcpu->arch.hpte_cache[hpte_id];
 274
 275                 dprintk_mmu("KVM: %c%c Map 0x%llx: [%lx] 0x%lx (0x%llx) -> %lx\n",
 276                             ((rflags & HPTE_R_PP) == 3) ? '-' : 'w',
 277                             (rflags & HPTE_R_N) ? '-' : 'x',
 278                             orig_pte->eaddr, hpteg, va, orig_pte->vpage, hpaddr);
 279
 280                 pte->slot = hpteg + (ret & 7);
 281                 pte->host_va = va;
 282                 pte->pte = *orig_pte;
 283                 pte->pfn = hpaddr >> PAGE_SHIFT;
 284         }
 285
 286         return 0;
 287 }
 288
 289 static struct kvmppc_sid_map *create_sid_map(struct kvm_vcpu *vcpu, u64 gvsid)
 290 {
 291         struct kvmppc_sid_map *map;
 292         struct kvmppc_vcpu_book3s *vcpu_book3s = to_book3s(vcpu);
 293         u16 sid_map_mask;
 294         static int backwards_map = 0;
 295
 296         if (vcpu->arch.msr & MSR_PR)
 297                 gvsid |= VSID_PR;
 298
 299         /* We might get collisions that trap in preceding order, so let's
 300            map them differently */
 301
 302         sid_map_mask = kvmppc_sid_hash(vcpu, gvsid);
 303         if (backwards_map)
 304                 sid_map_mask = SID_MAP_MASK - sid_map_mask;
 305
 306         map = &to_book3s(vcpu)->sid_map[sid_map_mask];
 307
 308         /* Make sure we're taking the other map next time */
 309         backwards_map = !backwards_map;
 310
 311         /* Uh-oh ... out of mappings. Let's flush! */
 312         if (vcpu_book3s->vsid_next == vcpu_book3s->vsid_max) {
 313                 vcpu_book3s->vsid_next = vcpu_book3s->vsid_first;
 314                 memset(vcpu_book3s->sid_map, 0,
 315                        sizeof(struct kvmppc_sid_map) * SID_MAP_NUM);
 316                 kvmppc_mmu_pte_flush(vcpu, 0, 0);
 317                 kvmppc_mmu_flush_segments(vcpu);
 318         }
 319         map->host_vsid = vcpu_book3s->vsid_next++;
 320
 321         map->guest_vsid = gvsid;
 322         map->valid = true;
 323
 324         return map;
 325 }
 326
 327 static int kvmppc_mmu_next_segment(struct kvm_vcpu *vcpu, ulong esid)
 328 {
 329         int i;
 330         int max_slb_size = 64;
 331         int found_inval = -1;
 332         int r;
 333
 334         if (!get_paca()->kvm_slb_max)
 335                 get_paca()->kvm_slb_max = 1;
 336
 337         /* Are we overwriting? */
 338         for (i = 1; i < get_paca()->kvm_slb_max; i++) {
 339                 if (!(get_paca()->kvm_slb[i].esid & SLB_ESID_V))
 340                         found_inval = i;
 341                 else if ((get_paca()->kvm_slb[i].esid & ESID_MASK) == esid)
 342                         return i;
 343         }
 344
 345         /* Found a spare entry that was invalidated before */
 346         if (found_inval > 0)
 347                 return found_inval;
 348
 349         /* No spare invalid entry, so create one */
 350
 351         if (mmu_slb_size < 64)
 352                 max_slb_size = mmu_slb_size;
 353
 354         /* Overflowing -> purge */
 355         if ((get_paca()->kvm_slb_max) == max_slb_size)
 356                 kvmppc_mmu_flush_segments(vcpu);
 357
 358         r = get_paca()->kvm_slb_max;
 359         get_paca()->kvm_slb_max++;
 360
 361         return r;
 362 }
 363
 364 int kvmppc_mmu_map_segment(struct kvm_vcpu *vcpu, ulong eaddr)
 365 {
 366         u64 esid = eaddr >> SID_SHIFT;
 367         u64 slb_esid = (eaddr & ESID_MASK) | SLB_ESID_V;
 368         u64 slb_vsid = SLB_VSID_USER;
 369         u64 gvsid;
 370         int slb_index;
 371         struct kvmppc_sid_map *map;
 372
 373         slb_index = kvmppc_mmu_next_segment(vcpu, eaddr & ESID_MASK);
 374
 375         if (vcpu->arch.mmu.esid_to_vsid(vcpu, esid, &gvsid)) {
 376                 /* Invalidate an entry */
 377                 get_paca()->kvm_slb[slb_index].esid = 0;
 378                 return -ENOENT;
 379         }
 380
 381         map = find_sid_vsid(vcpu, gvsid);
 382         if (!map)
 383                 map = create_sid_map(vcpu, gvsid);
 384
 385         map->guest_esid = esid;
 386
 387         slb_vsid |= (map->host_vsid << 12);
 388         slb_vsid &= ~SLB_VSID_KP;
 389         slb_esid |= slb_index;
 390
 391         get_paca()->kvm_slb[slb_index].esid = slb_esid;
 392         get_paca()->kvm_slb[slb_index].vsid = slb_vsid;
 393
 394         dprintk_slb("slbmte %#llx, %#llx\n", slb_vsid, slb_esid);
 395
 396         return 0;
 397 }
 398
 399 void kvmppc_mmu_flush_segments(struct kvm_vcpu *vcpu)
 400 {
 401         get_paca()->kvm_slb_max = 1;
 402         get_paca()->kvm_slb[0].esid = 0;
 403 }
 404
 405 void kvmppc_mmu_destroy(struct kvm_vcpu *vcpu)
 406 {
 407         kvmppc_mmu_pte_flush(vcpu, 0, 0);
 408 }