Linux Kernel  3.7.1
 All Data Structures Namespaces Files Functions Variables Typedefs Enumerations Enumerator Macros Groups Pages
book3s_hv_rm_mmu.c
Go to the documentation of this file.
1 /*
2  * This program is free software; you can redistribute it and/or modify
3  * it under the terms of the GNU General Public License, version 2, as
4  * published by the Free Software Foundation.
5  *
6  * Copyright 2010-2011 Paul Mackerras, IBM Corp. <[email protected]>
7  */
8 
9 #include <linux/types.h>
10 #include <linux/string.h>
11 #include <linux/kvm.h>
12 #include <linux/kvm_host.h>
13 #include <linux/hugetlb.h>
14 #include <linux/module.h>
15 
16 #include <asm/tlbflush.h>
17 #include <asm/kvm_ppc.h>
18 #include <asm/kvm_book3s.h>
19 #include <asm/mmu-hash64.h>
20 #include <asm/hvcall.h>
21 #include <asm/synch.h>
22 #include <asm/ppc-opcode.h>
23 
24 /* Translate address of a vmalloc'd thing to a linear map address */
25 static void *real_vmalloc_addr(void *x)
26 {
27  unsigned long addr = (unsigned long) x;
28  pte_t *p;
29 
30  p = find_linux_pte(swapper_pg_dir, addr);
31  if (!p || !pte_present(*p))
32  return NULL;
33  /* assume we don't have huge pages in vmalloc space... */
34  addr = (pte_pfn(*p) << PAGE_SHIFT) | (addr & ~PAGE_MASK);
35  return __va(addr);
36 }
37 
38 /*
39  * Add this HPTE into the chain for the real page.
40  * Must be called with the chain locked; it unlocks the chain.
41  */
43  unsigned long *rmap, long pte_index, int realmode)
44 {
45  struct revmap_entry *head, *tail;
46  unsigned long i;
47 
48  if (*rmap & KVMPPC_RMAP_PRESENT) {
49  i = *rmap & KVMPPC_RMAP_INDEX;
50  head = &kvm->arch.revmap[i];
51  if (realmode)
52  head = real_vmalloc_addr(head);
53  tail = &kvm->arch.revmap[head->back];
54  if (realmode)
55  tail = real_vmalloc_addr(tail);
56  rev->forw = i;
57  rev->back = head->back;
58  tail->forw = pte_index;
59  head->back = pte_index;
60  } else {
61  rev->forw = rev->back = pte_index;
62  i = pte_index;
63  }
64  smp_wmb();
65  *rmap = i | KVMPPC_RMAP_REFERENCED | KVMPPC_RMAP_PRESENT; /* unlock */
66 }
68 
69 /* Remove this HPTE from the chain for a real page */
70 static void remove_revmap_chain(struct kvm *kvm, long pte_index,
71  struct revmap_entry *rev,
72  unsigned long hpte_v, unsigned long hpte_r)
73 {
74  struct revmap_entry *next, *prev;
75  unsigned long gfn, ptel, head;
76  struct kvm_memory_slot *memslot;
77  unsigned long *rmap;
78  unsigned long rcbits;
79 
80  rcbits = hpte_r & (HPTE_R_R | HPTE_R_C);
81  ptel = rev->guest_rpte |= rcbits;
82  gfn = hpte_rpn(ptel, hpte_page_size(hpte_v, ptel));
83  memslot = __gfn_to_memslot(kvm_memslots(kvm), gfn);
84  if (!memslot || (memslot->flags & KVM_MEMSLOT_INVALID))
85  return;
86 
87  rmap = real_vmalloc_addr(&memslot->arch.rmap[gfn - memslot->base_gfn]);
88  lock_rmap(rmap);
89 
90  head = *rmap & KVMPPC_RMAP_INDEX;
91  next = real_vmalloc_addr(&kvm->arch.revmap[rev->forw]);
92  prev = real_vmalloc_addr(&kvm->arch.revmap[rev->back]);
93  next->back = rev->back;
94  prev->forw = rev->forw;
95  if (head == pte_index) {
96  head = rev->forw;
97  if (head == pte_index)
99  else
100  *rmap = (*rmap & ~KVMPPC_RMAP_INDEX) | head;
101  }
102  *rmap |= rcbits << KVMPPC_RMAP_RC_SHIFT;
103  unlock_rmap(rmap);
104 }
105 
106 static pte_t lookup_linux_pte(struct kvm_vcpu *vcpu, unsigned long hva,
107  int writing, unsigned long *pte_sizep)
108 {
109  pte_t *ptep;
110  unsigned long ps = *pte_sizep;
111  unsigned int shift;
112 
113  ptep = find_linux_pte_or_hugepte(vcpu->arch.pgdir, hva, &shift);
114  if (!ptep)
115  return __pte(0);
116  if (shift)
117  *pte_sizep = 1ul << shift;
118  else
119  *pte_sizep = PAGE_SIZE;
120  if (ps > *pte_sizep)
121  return __pte(0);
122  if (!pte_present(*ptep))
123  return __pte(0);
124  return kvmppc_read_update_linux_pte(ptep, writing);
125 }
126 
127 static inline void unlock_hpte(unsigned long *hpte, unsigned long hpte_v)
128 {
129  asm volatile(PPC_RELEASE_BARRIER "" : : : "memory");
130  hpte[0] = hpte_v;
131 }
132 
133 long kvmppc_h_enter(struct kvm_vcpu *vcpu, unsigned long flags,
134  long pte_index, unsigned long pteh, unsigned long ptel)
135 {
136  struct kvm *kvm = vcpu->kvm;
137  unsigned long i, pa, gpa, gfn, psize;
138  unsigned long slot_fn, hva;
139  unsigned long *hpte;
140  struct revmap_entry *rev;
141  unsigned long g_ptel = ptel;
142  struct kvm_memory_slot *memslot;
143  unsigned long *physp, pte_size;
144  unsigned long is_io;
145  unsigned long *rmap;
146  pte_t pte;
147  unsigned int writing;
148  unsigned long mmu_seq;
149  unsigned long rcbits;
150  bool realmode = vcpu->arch.vcore->vcore_state == VCORE_RUNNING;
151 
152  psize = hpte_page_size(pteh, ptel);
153  if (!psize)
154  return H_PARAMETER;
155  writing = hpte_is_writable(ptel);
157 
158  /* used later to detect if we might have been invalidated */
159  mmu_seq = kvm->mmu_notifier_seq;
160  smp_rmb();
161 
162  /* Find the memslot (if any) for this address */
163  gpa = (ptel & HPTE_R_RPN) & ~(psize - 1);
164  gfn = gpa >> PAGE_SHIFT;
165  memslot = __gfn_to_memslot(kvm_memslots(kvm), gfn);
166  pa = 0;
167  is_io = ~0ul;
168  rmap = NULL;
169  if (!(memslot && !(memslot->flags & KVM_MEMSLOT_INVALID))) {
170  /* PPC970 can't do emulated MMIO */
172  return H_PARAMETER;
173  /* Emulated MMIO - mark this with key=31 */
174  pteh |= HPTE_V_ABSENT;
175  ptel |= HPTE_R_KEY_HI | HPTE_R_KEY_LO;
176  goto do_insert;
177  }
178 
179  /* Check if the requested page fits entirely in the memslot. */
180  if (!slot_is_aligned(memslot, psize))
181  return H_PARAMETER;
182  slot_fn = gfn - memslot->base_gfn;
183  rmap = &memslot->arch.rmap[slot_fn];
184 
185  if (!kvm->arch.using_mmu_notifiers) {
186  physp = kvm->arch.slot_phys[memslot->id];
187  if (!physp)
188  return H_PARAMETER;
189  physp += slot_fn;
190  if (realmode)
191  physp = real_vmalloc_addr(physp);
192  pa = *physp;
193  if (!pa)
194  return H_TOO_HARD;
195  is_io = pa & (HPTE_R_I | HPTE_R_W);
196  pte_size = PAGE_SIZE << (pa & KVMPPC_PAGE_ORDER_MASK);
197  pa &= PAGE_MASK;
198  } else {
199  /* Translate to host virtual address */
200  hva = __gfn_to_hva_memslot(memslot, gfn);
201 
202  /* Look up the Linux PTE for the backing page */
203  pte_size = psize;
204  pte = lookup_linux_pte(vcpu, hva, writing, &pte_size);
205  if (pte_present(pte)) {
206  if (writing && !pte_write(pte))
207  /* make the actual HPTE be read-only */
208  ptel = hpte_make_readonly(ptel);
209  is_io = hpte_cache_bits(pte_val(pte));
210  pa = pte_pfn(pte) << PAGE_SHIFT;
211  }
212  }
213  if (pte_size < psize)
214  return H_PARAMETER;
215  if (pa && pte_size > psize)
216  pa |= gpa & (pte_size - 1);
217 
218  ptel &= ~(HPTE_R_PP0 - psize);
219  ptel |= pa;
220 
221  if (pa)
222  pteh |= HPTE_V_VALID;
223  else
224  pteh |= HPTE_V_ABSENT;
225 
226  /* Check WIMG */
227  if (is_io != ~0ul && !hpte_cache_flags_ok(ptel, is_io)) {
228  if (is_io)
229  return H_PARAMETER;
230  /*
231  * Allow guest to map emulated device memory as
232  * uncacheable, but actually make it cacheable.
233  */
234  ptel &= ~(HPTE_R_W|HPTE_R_I|HPTE_R_G);
235  ptel |= HPTE_R_M;
236  }
237 
238  /* Find and lock the HPTEG slot to use */
239  do_insert:
240  if (pte_index >= kvm->arch.hpt_npte)
241  return H_PARAMETER;
242  if (likely((flags & H_EXACT) == 0)) {
243  pte_index &= ~7UL;
244  hpte = (unsigned long *)(kvm->arch.hpt_virt + (pte_index << 4));
245  for (i = 0; i < 8; ++i) {
246  if ((*hpte & HPTE_V_VALID) == 0 &&
247  try_lock_hpte(hpte, HPTE_V_HVLOCK | HPTE_V_VALID |
248  HPTE_V_ABSENT))
249  break;
250  hpte += 2;
251  }
252  if (i == 8) {
253  /*
254  * Since try_lock_hpte doesn't retry (not even stdcx.
255  * failures), it could be that there is a free slot
256  * but we transiently failed to lock it. Try again,
257  * actually locking each slot and checking it.
258  */
259  hpte -= 16;
260  for (i = 0; i < 8; ++i) {
261  while (!try_lock_hpte(hpte, HPTE_V_HVLOCK))
262  cpu_relax();
263  if (!(*hpte & (HPTE_V_VALID | HPTE_V_ABSENT)))
264  break;
265  *hpte &= ~HPTE_V_HVLOCK;
266  hpte += 2;
267  }
268  if (i == 8)
269  return H_PTEG_FULL;
270  }
271  pte_index += i;
272  } else {
273  hpte = (unsigned long *)(kvm->arch.hpt_virt + (pte_index << 4));
274  if (!try_lock_hpte(hpte, HPTE_V_HVLOCK | HPTE_V_VALID |
275  HPTE_V_ABSENT)) {
276  /* Lock the slot and check again */
277  while (!try_lock_hpte(hpte, HPTE_V_HVLOCK))
278  cpu_relax();
279  if (*hpte & (HPTE_V_VALID | HPTE_V_ABSENT)) {
280  *hpte &= ~HPTE_V_HVLOCK;
281  return H_PTEG_FULL;
282  }
283  }
284  }
285 
286  /* Save away the guest's idea of the second HPTE dword */
287  rev = &kvm->arch.revmap[pte_index];
288  if (realmode)
289  rev = real_vmalloc_addr(rev);
290  if (rev)
291  rev->guest_rpte = g_ptel;
292 
293  /* Link HPTE into reverse-map chain */
294  if (pteh & HPTE_V_VALID) {
295  if (realmode)
296  rmap = real_vmalloc_addr(rmap);
297  lock_rmap(rmap);
298  /* Check for pending invalidations under the rmap chain lock */
299  if (kvm->arch.using_mmu_notifiers &&
300  mmu_notifier_retry(vcpu, mmu_seq)) {
301  /* inval in progress, write a non-present HPTE */
302  pteh |= HPTE_V_ABSENT;
303  pteh &= ~HPTE_V_VALID;
304  unlock_rmap(rmap);
305  } else {
306  kvmppc_add_revmap_chain(kvm, rev, rmap, pte_index,
307  realmode);
308  /* Only set R/C in real HPTE if already set in *rmap */
309  rcbits = *rmap >> KVMPPC_RMAP_RC_SHIFT;
310  ptel &= rcbits | ~(HPTE_R_R | HPTE_R_C);
311  }
312  }
313 
314  hpte[1] = ptel;
315 
316  /* Write the first HPTE dword, unlocking the HPTE and making it valid */
317  eieio();
318  hpte[0] = pteh;
319  asm volatile("ptesync" : : : "memory");
320 
321  vcpu->arch.gpr[4] = pte_index;
322  return H_SUCCESS;
323 }
325 
326 #define LOCK_TOKEN (*(u32 *)(&get_paca()->lock_token))
327 
328 static inline int try_lock_tlbie(unsigned int *lock)
329 {
330  unsigned int tmp, old;
331  unsigned int token = LOCK_TOKEN;
332 
333  asm volatile("1:lwarx %1,0,%2\n"
334  " cmpwi cr0,%1,0\n"
335  " bne 2f\n"
336  " stwcx. %3,0,%2\n"
337  " bne- 1b\n"
338  " isync\n"
339  "2:"
340  : "=&r" (tmp), "=&r" (old)
341  : "r" (lock), "r" (token)
342  : "cc", "memory");
343  return old == 0;
344 }
345 
346 long kvmppc_h_remove(struct kvm_vcpu *vcpu, unsigned long flags,
347  unsigned long pte_index, unsigned long avpn,
348  unsigned long va)
349 {
350  struct kvm *kvm = vcpu->kvm;
351  unsigned long *hpte;
352  unsigned long v, r, rb;
353  struct revmap_entry *rev;
354 
355  if (pte_index >= kvm->arch.hpt_npte)
356  return H_PARAMETER;
357  hpte = (unsigned long *)(kvm->arch.hpt_virt + (pte_index << 4));
358  while (!try_lock_hpte(hpte, HPTE_V_HVLOCK))
359  cpu_relax();
360  if ((hpte[0] & (HPTE_V_ABSENT | HPTE_V_VALID)) == 0 ||
361  ((flags & H_AVPN) && (hpte[0] & ~0x7fUL) != avpn) ||
362  ((flags & H_ANDCOND) && (hpte[0] & avpn) != 0)) {
363  hpte[0] &= ~HPTE_V_HVLOCK;
364  return H_NOT_FOUND;
365  }
366 
367  rev = real_vmalloc_addr(&kvm->arch.revmap[pte_index]);
368  v = hpte[0] & ~HPTE_V_HVLOCK;
369  if (v & HPTE_V_VALID) {
370  hpte[0] &= ~HPTE_V_VALID;
371  rb = compute_tlbie_rb(v, hpte[1], pte_index);
372  if (!(flags & H_LOCAL) && atomic_read(&kvm->online_vcpus) > 1) {
373  while (!try_lock_tlbie(&kvm->arch.tlbie_lock))
374  cpu_relax();
375  asm volatile("ptesync" : : : "memory");
376  asm volatile(PPC_TLBIE(%1,%0)"; eieio; tlbsync"
377  : : "r" (rb), "r" (kvm->arch.lpid));
378  asm volatile("ptesync" : : : "memory");
379  kvm->arch.tlbie_lock = 0;
380  } else {
381  asm volatile("ptesync" : : : "memory");
382  asm volatile("tlbiel %0" : : "r" (rb));
383  asm volatile("ptesync" : : : "memory");
384  }
385  /* Read PTE low word after tlbie to get final R/C values */
386  remove_revmap_chain(kvm, pte_index, rev, v, hpte[1]);
387  }
388  r = rev->guest_rpte;
389  unlock_hpte(hpte, 0);
390 
391  vcpu->arch.gpr[4] = v;
392  vcpu->arch.gpr[5] = r;
393  return H_SUCCESS;
394 }
395 
396 long kvmppc_h_bulk_remove(struct kvm_vcpu *vcpu)
397 {
398  struct kvm *kvm = vcpu->kvm;
399  unsigned long *args = &vcpu->arch.gpr[4];
400  unsigned long *hp, *hptes[4], tlbrb[4];
401  long int i, j, k, n, found, indexes[4];
402  unsigned long flags, req, pte_index, rcbits;
403  long int local = 0;
404  long int ret = H_SUCCESS;
405  struct revmap_entry *rev, *revs[4];
406 
407  if (atomic_read(&kvm->online_vcpus) == 1)
408  local = 1;
409  for (i = 0; i < 4 && ret == H_SUCCESS; ) {
410  n = 0;
411  for (; i < 4; ++i) {
412  j = i * 2;
413  pte_index = args[j];
414  flags = pte_index >> 56;
415  pte_index &= ((1ul << 56) - 1);
416  req = flags >> 6;
417  flags &= 3;
418  if (req == 3) { /* no more requests */
419  i = 4;
420  break;
421  }
422  if (req != 1 || flags == 3 ||
423  pte_index >= kvm->arch.hpt_npte) {
424  /* parameter error */
425  args[j] = ((0xa0 | flags) << 56) + pte_index;
426  ret = H_PARAMETER;
427  break;
428  }
429  hp = (unsigned long *)
430  (kvm->arch.hpt_virt + (pte_index << 4));
431  /* to avoid deadlock, don't spin except for first */
432  if (!try_lock_hpte(hp, HPTE_V_HVLOCK)) {
433  if (n)
434  break;
435  while (!try_lock_hpte(hp, HPTE_V_HVLOCK))
436  cpu_relax();
437  }
438  found = 0;
439  if (hp[0] & (HPTE_V_ABSENT | HPTE_V_VALID)) {
440  switch (flags & 3) {
441  case 0: /* absolute */
442  found = 1;
443  break;
444  case 1: /* andcond */
445  if (!(hp[0] & args[j + 1]))
446  found = 1;
447  break;
448  case 2: /* AVPN */
449  if ((hp[0] & ~0x7fUL) == args[j + 1])
450  found = 1;
451  break;
452  }
453  }
454  if (!found) {
455  hp[0] &= ~HPTE_V_HVLOCK;
456  args[j] = ((0x90 | flags) << 56) + pte_index;
457  continue;
458  }
459 
460  args[j] = ((0x80 | flags) << 56) + pte_index;
461  rev = real_vmalloc_addr(&kvm->arch.revmap[pte_index]);
462 
463  if (!(hp[0] & HPTE_V_VALID)) {
464  /* insert R and C bits from PTE */
465  rcbits = rev->guest_rpte & (HPTE_R_R|HPTE_R_C);
466  args[j] |= rcbits << (56 - 5);
467  hp[0] = 0;
468  continue;
469  }
470 
471  hp[0] &= ~HPTE_V_VALID; /* leave it locked */
472  tlbrb[n] = compute_tlbie_rb(hp[0], hp[1], pte_index);
473  indexes[n] = j;
474  hptes[n] = hp;
475  revs[n] = rev;
476  ++n;
477  }
478 
479  if (!n)
480  break;
481 
482  /* Now that we've collected a batch, do the tlbies */
483  if (!local) {
484  while(!try_lock_tlbie(&kvm->arch.tlbie_lock))
485  cpu_relax();
486  asm volatile("ptesync" : : : "memory");
487  for (k = 0; k < n; ++k)
488  asm volatile(PPC_TLBIE(%1,%0) : :
489  "r" (tlbrb[k]),
490  "r" (kvm->arch.lpid));
491  asm volatile("eieio; tlbsync; ptesync" : : : "memory");
492  kvm->arch.tlbie_lock = 0;
493  } else {
494  asm volatile("ptesync" : : : "memory");
495  for (k = 0; k < n; ++k)
496  asm volatile("tlbiel %0" : : "r" (tlbrb[k]));
497  asm volatile("ptesync" : : : "memory");
498  }
499 
500  /* Read PTE low words after tlbie to get final R/C values */
501  for (k = 0; k < n; ++k) {
502  j = indexes[k];
503  pte_index = args[j] & ((1ul << 56) - 1);
504  hp = hptes[k];
505  rev = revs[k];
506  remove_revmap_chain(kvm, pte_index, rev, hp[0], hp[1]);
507  rcbits = rev->guest_rpte & (HPTE_R_R|HPTE_R_C);
508  args[j] |= rcbits << (56 - 5);
509  hp[0] = 0;
510  }
511  }
512 
513  return ret;
514 }
515 
516 long kvmppc_h_protect(struct kvm_vcpu *vcpu, unsigned long flags,
517  unsigned long pte_index, unsigned long avpn,
518  unsigned long va)
519 {
520  struct kvm *kvm = vcpu->kvm;
521  unsigned long *hpte;
522  struct revmap_entry *rev;
523  unsigned long v, r, rb, mask, bits;
524 
525  if (pte_index >= kvm->arch.hpt_npte)
526  return H_PARAMETER;
527 
528  hpte = (unsigned long *)(kvm->arch.hpt_virt + (pte_index << 4));
529  while (!try_lock_hpte(hpte, HPTE_V_HVLOCK))
530  cpu_relax();
531  if ((hpte[0] & (HPTE_V_ABSENT | HPTE_V_VALID)) == 0 ||
532  ((flags & H_AVPN) && (hpte[0] & ~0x7fUL) != avpn)) {
533  hpte[0] &= ~HPTE_V_HVLOCK;
534  return H_NOT_FOUND;
535  }
536 
537  if (atomic_read(&kvm->online_vcpus) == 1)
538  flags |= H_LOCAL;
539  v = hpte[0];
540  bits = (flags << 55) & HPTE_R_PP0;
541  bits |= (flags << 48) & HPTE_R_KEY_HI;
542  bits |= flags & (HPTE_R_PP | HPTE_R_N | HPTE_R_KEY_LO);
543 
544  /* Update guest view of 2nd HPTE dword */
545  mask = HPTE_R_PP0 | HPTE_R_PP | HPTE_R_N |
547  rev = real_vmalloc_addr(&kvm->arch.revmap[pte_index]);
548  if (rev) {
549  r = (rev->guest_rpte & ~mask) | bits;
550  rev->guest_rpte = r;
551  }
552  r = (hpte[1] & ~mask) | bits;
553 
554  /* Update HPTE */
555  if (v & HPTE_V_VALID) {
556  rb = compute_tlbie_rb(v, r, pte_index);
557  hpte[0] = v & ~HPTE_V_VALID;
558  if (!(flags & H_LOCAL)) {
559  while(!try_lock_tlbie(&kvm->arch.tlbie_lock))
560  cpu_relax();
561  asm volatile("ptesync" : : : "memory");
562  asm volatile(PPC_TLBIE(%1,%0)"; eieio; tlbsync"
563  : : "r" (rb), "r" (kvm->arch.lpid));
564  asm volatile("ptesync" : : : "memory");
565  kvm->arch.tlbie_lock = 0;
566  } else {
567  asm volatile("ptesync" : : : "memory");
568  asm volatile("tlbiel %0" : : "r" (rb));
569  asm volatile("ptesync" : : : "memory");
570  }
571  }
572  hpte[1] = r;
573  eieio();
574  hpte[0] = v & ~HPTE_V_HVLOCK;
575  asm volatile("ptesync" : : : "memory");
576  return H_SUCCESS;
577 }
578 
579 long kvmppc_h_read(struct kvm_vcpu *vcpu, unsigned long flags,
580  unsigned long pte_index)
581 {
582  struct kvm *kvm = vcpu->kvm;
583  unsigned long *hpte, v, r;
584  int i, n = 1;
585  struct revmap_entry *rev = NULL;
586 
587  if (pte_index >= kvm->arch.hpt_npte)
588  return H_PARAMETER;
589  if (flags & H_READ_4) {
590  pte_index &= ~3;
591  n = 4;
592  }
593  rev = real_vmalloc_addr(&kvm->arch.revmap[pte_index]);
594  for (i = 0; i < n; ++i, ++pte_index) {
595  hpte = (unsigned long *)(kvm->arch.hpt_virt + (pte_index << 4));
596  v = hpte[0] & ~HPTE_V_HVLOCK;
597  r = hpte[1];
598  if (v & HPTE_V_ABSENT) {
599  v &= ~HPTE_V_ABSENT;
600  v |= HPTE_V_VALID;
601  }
602  if (v & HPTE_V_VALID)
603  r = rev[i].guest_rpte | (r & (HPTE_R_R | HPTE_R_C));
604  vcpu->arch.gpr[4 + i * 2] = v;
605  vcpu->arch.gpr[5 + i * 2] = r;
606  }
607  return H_SUCCESS;
608 }
609 
610 void kvmppc_invalidate_hpte(struct kvm *kvm, unsigned long *hptep,
611  unsigned long pte_index)
612 {
613  unsigned long rb;
614 
615  hptep[0] &= ~HPTE_V_VALID;
616  rb = compute_tlbie_rb(hptep[0], hptep[1], pte_index);
617  while (!try_lock_tlbie(&kvm->arch.tlbie_lock))
618  cpu_relax();
619  asm volatile("ptesync" : : : "memory");
620  asm volatile(PPC_TLBIE(%1,%0)"; eieio; tlbsync"
621  : : "r" (rb), "r" (kvm->arch.lpid));
622  asm volatile("ptesync" : : : "memory");
623  kvm->arch.tlbie_lock = 0;
624 }
626 
627 void kvmppc_clear_ref_hpte(struct kvm *kvm, unsigned long *hptep,
628  unsigned long pte_index)
629 {
630  unsigned long rb;
631  unsigned char rbyte;
632 
633  rb = compute_tlbie_rb(hptep[0], hptep[1], pte_index);
634  rbyte = (hptep[1] & ~HPTE_R_R) >> 8;
635  /* modify only the second-last byte, which contains the ref bit */
636  *((char *)hptep + 14) = rbyte;
637  while (!try_lock_tlbie(&kvm->arch.tlbie_lock))
638  cpu_relax();
639  asm volatile(PPC_TLBIE(%1,%0)"; eieio; tlbsync"
640  : : "r" (rb), "r" (kvm->arch.lpid));
641  asm volatile("ptesync" : : : "memory");
642  kvm->arch.tlbie_lock = 0;
643 }
645 
646 static int slb_base_page_shift[4] = {
647  24, /* 16M */
648  16, /* 64k */
649  34, /* 16G */
650  20, /* 1M, unsupported */
651 };
652 
653 long kvmppc_hv_find_lock_hpte(struct kvm *kvm, gva_t eaddr, unsigned long slb_v,
654  unsigned long valid)
655 {
656  unsigned int i;
657  unsigned int pshift;
658  unsigned long somask;
659  unsigned long vsid, hash;
660  unsigned long avpn;
661  unsigned long *hpte;
662  unsigned long mask, val;
663  unsigned long v, r;
664 
665  /* Get page shift, work out hash and AVPN etc. */
667  val = 0;
668  pshift = 12;
669  if (slb_v & SLB_VSID_L) {
670  mask |= HPTE_V_LARGE;
671  val |= HPTE_V_LARGE;
672  pshift = slb_base_page_shift[(slb_v & SLB_VSID_LP) >> 4];
673  }
674  if (slb_v & SLB_VSID_B_1T) {
675  somask = (1UL << 40) - 1;
676  vsid = (slb_v & ~SLB_VSID_B) >> SLB_VSID_SHIFT_1T;
677  vsid ^= vsid << 25;
678  } else {
679  somask = (1UL << 28) - 1;
680  vsid = (slb_v & ~SLB_VSID_B) >> SLB_VSID_SHIFT;
681  }
682  hash = (vsid ^ ((eaddr & somask) >> pshift)) & kvm->arch.hpt_mask;
683  avpn = slb_v & ~(somask >> 16); /* also includes B */
684  avpn |= (eaddr & somask) >> 16;
685 
686  if (pshift >= 24)
687  avpn &= ~((1UL << (pshift - 16)) - 1);
688  else
689  avpn &= ~0x7fUL;
690  val |= avpn;
691 
692  for (;;) {
693  hpte = (unsigned long *)(kvm->arch.hpt_virt + (hash << 7));
694 
695  for (i = 0; i < 16; i += 2) {
696  /* Read the PTE racily */
697  v = hpte[i] & ~HPTE_V_HVLOCK;
698 
699  /* Check valid/absent, hash, segment size and AVPN */
700  if (!(v & valid) || (v & mask) != val)
701  continue;
702 
703  /* Lock the PTE and read it under the lock */
704  while (!try_lock_hpte(&hpte[i], HPTE_V_HVLOCK))
705  cpu_relax();
706  v = hpte[i] & ~HPTE_V_HVLOCK;
707  r = hpte[i+1];
708 
709  /*
710  * Check the HPTE again, including large page size
711  * Since we don't currently allow any MPSS (mixed
712  * page-size segment) page sizes, it is sufficient
713  * to check against the actual page size.
714  */
715  if ((v & valid) && (v & mask) == val &&
716  hpte_page_size(v, r) == (1ul << pshift))
717  /* Return with the HPTE still locked */
718  return (hash << 3) + (i >> 1);
719 
720  /* Unlock and move on */
721  hpte[i] = v;
722  }
723 
724  if (val & HPTE_V_SECONDARY)
725  break;
726  val |= HPTE_V_SECONDARY;
727  hash = hash ^ kvm->arch.hpt_mask;
728  }
729  return -1;
730 }
732 
733 /*
734  * Called in real mode to check whether an HPTE not found fault
735  * is due to accessing a paged-out page or an emulated MMIO page,
736  * or if a protection fault is due to accessing a page that the
737  * guest wanted read/write access to but which we made read-only.
738  * Returns a possibly modified status (DSISR) value if not
739  * (i.e. pass the interrupt to the guest),
740  * -1 to pass the fault up to host kernel mode code, -2 to do that
741  * and also load the instruction word (for MMIO emulation),
742  * or 0 if we should make the guest retry the access.
743  */
744 long kvmppc_hpte_hv_fault(struct kvm_vcpu *vcpu, unsigned long addr,
745  unsigned long slb_v, unsigned int status, bool data)
746 {
747  struct kvm *kvm = vcpu->kvm;
748  long int index;
749  unsigned long v, r, gr;
750  unsigned long *hpte;
751  unsigned long valid;
752  struct revmap_entry *rev;
753  unsigned long pp, key;
754 
755  /* For protection fault, expect to find a valid HPTE */
756  valid = HPTE_V_VALID;
757  if (status & DSISR_NOHPTE)
758  valid |= HPTE_V_ABSENT;
759 
760  index = kvmppc_hv_find_lock_hpte(kvm, addr, slb_v, valid);
761  if (index < 0) {
762  if (status & DSISR_NOHPTE)
763  return status; /* there really was no HPTE */
764  return 0; /* for prot fault, HPTE disappeared */
765  }
766  hpte = (unsigned long *)(kvm->arch.hpt_virt + (index << 4));
767  v = hpte[0] & ~HPTE_V_HVLOCK;
768  r = hpte[1];
769  rev = real_vmalloc_addr(&kvm->arch.revmap[index]);
770  gr = rev->guest_rpte;
771 
772  unlock_hpte(hpte, v);
773 
774  /* For not found, if the HPTE is valid by now, retry the instruction */
775  if ((status & DSISR_NOHPTE) && (v & HPTE_V_VALID))
776  return 0;
777 
778  /* Check access permissions to the page */
779  pp = gr & (HPTE_R_PP0 | HPTE_R_PP);
780  key = (vcpu->arch.shregs.msr & MSR_PR) ? SLB_VSID_KP : SLB_VSID_KS;
781  status &= ~DSISR_NOHPTE; /* DSISR_NOHPTE == SRR1_ISI_NOPT */
782  if (!data) {
783  if (gr & (HPTE_R_N | HPTE_R_G))
784  return status | SRR1_ISI_N_OR_G;
785  if (!hpte_read_permission(pp, slb_v & key))
786  return status | SRR1_ISI_PROT;
787  } else if (status & DSISR_ISSTORE) {
788  /* check write permission */
789  if (!hpte_write_permission(pp, slb_v & key))
790  return status | DSISR_PROTFAULT;
791  } else {
792  if (!hpte_read_permission(pp, slb_v & key))
793  return status | DSISR_PROTFAULT;
794  }
795 
796  /* Check storage key, if applicable */
797  if (data && (vcpu->arch.shregs.msr & MSR_DR)) {
798  unsigned int perm = hpte_get_skey_perm(gr, vcpu->arch.amr);
799  if (status & DSISR_ISSTORE)
800  perm >>= 1;
801  if (perm & 1)
802  return status | DSISR_KEYFAULT;
803  }
804 
805  /* Save HPTE info for virtual-mode handler */
806  vcpu->arch.pgfault_addr = addr;
807  vcpu->arch.pgfault_index = index;
808  vcpu->arch.pgfault_hpte[0] = v;
809  vcpu->arch.pgfault_hpte[1] = r;
810 
811  /* Check the storage key to see if it is possibly emulated MMIO */
812  if (data && (vcpu->arch.shregs.msr & MSR_IR) &&
813  (r & (HPTE_R_KEY_HI | HPTE_R_KEY_LO)) ==
815  return -2; /* MMIO emulation - load instr word */
816 
817  return -1; /* send fault up to host kernel mode */
818 }