xref: /linux/arch/powerpc/kvm/book3s_hv_rm_mmu.c (revision 4949009eb8d40a441dcddcd96e101e77d31cf1b2)
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. <paulus@au1.ibm.com>
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_or_hugepte(swapper_pg_dir, addr, NULL);
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 /* Return 1 if we need to do a global tlbie, 0 if we can use tlbiel */
39 static int global_invalidates(struct kvm *kvm, unsigned long flags)
40 {
41 	int global;
42 
43 	/*
44 	 * If there is only one vcore, and it's currently running,
45 	 * as indicated by local_paca->kvm_hstate.kvm_vcpu being set,
46 	 * we can use tlbiel as long as we mark all other physical
47 	 * cores as potentially having stale TLB entries for this lpid.
48 	 * Otherwise, don't use tlbiel.
49 	 */
50 	if (kvm->arch.online_vcores == 1 && local_paca->kvm_hstate.kvm_vcpu)
51 		global = 0;
52 	else
53 		global = 1;
54 
55 	if (!global) {
56 		/* any other core might now have stale TLB entries... */
57 		smp_wmb();
58 		cpumask_setall(&kvm->arch.need_tlb_flush);
59 		cpumask_clear_cpu(local_paca->kvm_hstate.kvm_vcore->pcpu,
60 				  &kvm->arch.need_tlb_flush);
61 	}
62 
63 	return global;
64 }
65 
66 /*
67  * Add this HPTE into the chain for the real page.
68  * Must be called with the chain locked; it unlocks the chain.
69  */
70 void kvmppc_add_revmap_chain(struct kvm *kvm, struct revmap_entry *rev,
71 			     unsigned long *rmap, long pte_index, int realmode)
72 {
73 	struct revmap_entry *head, *tail;
74 	unsigned long i;
75 
76 	if (*rmap & KVMPPC_RMAP_PRESENT) {
77 		i = *rmap & KVMPPC_RMAP_INDEX;
78 		head = &kvm->arch.revmap[i];
79 		if (realmode)
80 			head = real_vmalloc_addr(head);
81 		tail = &kvm->arch.revmap[head->back];
82 		if (realmode)
83 			tail = real_vmalloc_addr(tail);
84 		rev->forw = i;
85 		rev->back = head->back;
86 		tail->forw = pte_index;
87 		head->back = pte_index;
88 	} else {
89 		rev->forw = rev->back = pte_index;
90 		*rmap = (*rmap & ~KVMPPC_RMAP_INDEX) |
91 			pte_index | KVMPPC_RMAP_PRESENT;
92 	}
93 	unlock_rmap(rmap);
94 }
95 EXPORT_SYMBOL_GPL(kvmppc_add_revmap_chain);
96 
97 /* Remove this HPTE from the chain for a real page */
98 static void remove_revmap_chain(struct kvm *kvm, long pte_index,
99 				struct revmap_entry *rev,
100 				unsigned long hpte_v, unsigned long hpte_r)
101 {
102 	struct revmap_entry *next, *prev;
103 	unsigned long gfn, ptel, head;
104 	struct kvm_memory_slot *memslot;
105 	unsigned long *rmap;
106 	unsigned long rcbits;
107 
108 	rcbits = hpte_r & (HPTE_R_R | HPTE_R_C);
109 	ptel = rev->guest_rpte |= rcbits;
110 	gfn = hpte_rpn(ptel, hpte_page_size(hpte_v, ptel));
111 	memslot = __gfn_to_memslot(kvm_memslots_raw(kvm), gfn);
112 	if (!memslot)
113 		return;
114 
115 	rmap = real_vmalloc_addr(&memslot->arch.rmap[gfn - memslot->base_gfn]);
116 	lock_rmap(rmap);
117 
118 	head = *rmap & KVMPPC_RMAP_INDEX;
119 	next = real_vmalloc_addr(&kvm->arch.revmap[rev->forw]);
120 	prev = real_vmalloc_addr(&kvm->arch.revmap[rev->back]);
121 	next->back = rev->back;
122 	prev->forw = rev->forw;
123 	if (head == pte_index) {
124 		head = rev->forw;
125 		if (head == pte_index)
126 			*rmap &= ~(KVMPPC_RMAP_PRESENT | KVMPPC_RMAP_INDEX);
127 		else
128 			*rmap = (*rmap & ~KVMPPC_RMAP_INDEX) | head;
129 	}
130 	*rmap |= rcbits << KVMPPC_RMAP_RC_SHIFT;
131 	unlock_rmap(rmap);
132 }
133 
134 static pte_t lookup_linux_pte_and_update(pgd_t *pgdir, unsigned long hva,
135 			      int writing, unsigned long *pte_sizep)
136 {
137 	pte_t *ptep;
138 	unsigned long ps = *pte_sizep;
139 	unsigned int hugepage_shift;
140 
141 	ptep = find_linux_pte_or_hugepte(pgdir, hva, &hugepage_shift);
142 	if (!ptep)
143 		return __pte(0);
144 	if (hugepage_shift)
145 		*pte_sizep = 1ul << hugepage_shift;
146 	else
147 		*pte_sizep = PAGE_SIZE;
148 	if (ps > *pte_sizep)
149 		return __pte(0);
150 	return kvmppc_read_update_linux_pte(ptep, writing, hugepage_shift);
151 }
152 
153 static inline void unlock_hpte(__be64 *hpte, unsigned long hpte_v)
154 {
155 	asm volatile(PPC_RELEASE_BARRIER "" : : : "memory");
156 	hpte[0] = cpu_to_be64(hpte_v);
157 }
158 
159 long kvmppc_do_h_enter(struct kvm *kvm, unsigned long flags,
160 		       long pte_index, unsigned long pteh, unsigned long ptel,
161 		       pgd_t *pgdir, bool realmode, unsigned long *pte_idx_ret)
162 {
163 	unsigned long i, pa, gpa, gfn, psize;
164 	unsigned long slot_fn, hva;
165 	__be64 *hpte;
166 	struct revmap_entry *rev;
167 	unsigned long g_ptel;
168 	struct kvm_memory_slot *memslot;
169 	unsigned long pte_size;
170 	unsigned long is_io;
171 	unsigned long *rmap;
172 	pte_t pte;
173 	unsigned int writing;
174 	unsigned long mmu_seq;
175 	unsigned long rcbits;
176 
177 	psize = hpte_page_size(pteh, ptel);
178 	if (!psize)
179 		return H_PARAMETER;
180 	writing = hpte_is_writable(ptel);
181 	pteh &= ~(HPTE_V_HVLOCK | HPTE_V_ABSENT | HPTE_V_VALID);
182 	ptel &= ~HPTE_GR_RESERVED;
183 	g_ptel = ptel;
184 
185 	/* used later to detect if we might have been invalidated */
186 	mmu_seq = kvm->mmu_notifier_seq;
187 	smp_rmb();
188 
189 	/* Find the memslot (if any) for this address */
190 	gpa = (ptel & HPTE_R_RPN) & ~(psize - 1);
191 	gfn = gpa >> PAGE_SHIFT;
192 	memslot = __gfn_to_memslot(kvm_memslots_raw(kvm), gfn);
193 	pa = 0;
194 	is_io = ~0ul;
195 	rmap = NULL;
196 	if (!(memslot && !(memslot->flags & KVM_MEMSLOT_INVALID))) {
197 		/* Emulated MMIO - mark this with key=31 */
198 		pteh |= HPTE_V_ABSENT;
199 		ptel |= HPTE_R_KEY_HI | HPTE_R_KEY_LO;
200 		goto do_insert;
201 	}
202 
203 	/* Check if the requested page fits entirely in the memslot. */
204 	if (!slot_is_aligned(memslot, psize))
205 		return H_PARAMETER;
206 	slot_fn = gfn - memslot->base_gfn;
207 	rmap = &memslot->arch.rmap[slot_fn];
208 
209 	/* Translate to host virtual address */
210 	hva = __gfn_to_hva_memslot(memslot, gfn);
211 
212 	/* Look up the Linux PTE for the backing page */
213 	pte_size = psize;
214 	pte = lookup_linux_pte_and_update(pgdir, hva, writing, &pte_size);
215 	if (pte_present(pte) && !pte_numa(pte)) {
216 		if (writing && !pte_write(pte))
217 			/* make the actual HPTE be read-only */
218 			ptel = hpte_make_readonly(ptel);
219 		is_io = hpte_cache_bits(pte_val(pte));
220 		pa = pte_pfn(pte) << PAGE_SHIFT;
221 		pa |= hva & (pte_size - 1);
222 		pa |= gpa & ~PAGE_MASK;
223 	}
224 
225 	if (pte_size < psize)
226 		return H_PARAMETER;
227 
228 	ptel &= ~(HPTE_R_PP0 - psize);
229 	ptel |= pa;
230 
231 	if (pa)
232 		pteh |= HPTE_V_VALID;
233 	else
234 		pteh |= HPTE_V_ABSENT;
235 
236 	/* Check WIMG */
237 	if (is_io != ~0ul && !hpte_cache_flags_ok(ptel, is_io)) {
238 		if (is_io)
239 			return H_PARAMETER;
240 		/*
241 		 * Allow guest to map emulated device memory as
242 		 * uncacheable, but actually make it cacheable.
243 		 */
244 		ptel &= ~(HPTE_R_W|HPTE_R_I|HPTE_R_G);
245 		ptel |= HPTE_R_M;
246 	}
247 
248 	/* Find and lock the HPTEG slot to use */
249  do_insert:
250 	if (pte_index >= kvm->arch.hpt_npte)
251 		return H_PARAMETER;
252 	if (likely((flags & H_EXACT) == 0)) {
253 		pte_index &= ~7UL;
254 		hpte = (__be64 *)(kvm->arch.hpt_virt + (pte_index << 4));
255 		for (i = 0; i < 8; ++i) {
256 			if ((be64_to_cpu(*hpte) & HPTE_V_VALID) == 0 &&
257 			    try_lock_hpte(hpte, HPTE_V_HVLOCK | HPTE_V_VALID |
258 					  HPTE_V_ABSENT))
259 				break;
260 			hpte += 2;
261 		}
262 		if (i == 8) {
263 			/*
264 			 * Since try_lock_hpte doesn't retry (not even stdcx.
265 			 * failures), it could be that there is a free slot
266 			 * but we transiently failed to lock it.  Try again,
267 			 * actually locking each slot and checking it.
268 			 */
269 			hpte -= 16;
270 			for (i = 0; i < 8; ++i) {
271 				u64 pte;
272 				while (!try_lock_hpte(hpte, HPTE_V_HVLOCK))
273 					cpu_relax();
274 				pte = be64_to_cpu(*hpte);
275 				if (!(pte & (HPTE_V_VALID | HPTE_V_ABSENT)))
276 					break;
277 				*hpte &= ~cpu_to_be64(HPTE_V_HVLOCK);
278 				hpte += 2;
279 			}
280 			if (i == 8)
281 				return H_PTEG_FULL;
282 		}
283 		pte_index += i;
284 	} else {
285 		hpte = (__be64 *)(kvm->arch.hpt_virt + (pte_index << 4));
286 		if (!try_lock_hpte(hpte, HPTE_V_HVLOCK | HPTE_V_VALID |
287 				   HPTE_V_ABSENT)) {
288 			/* Lock the slot and check again */
289 			u64 pte;
290 
291 			while (!try_lock_hpte(hpte, HPTE_V_HVLOCK))
292 				cpu_relax();
293 			pte = be64_to_cpu(*hpte);
294 			if (pte & (HPTE_V_VALID | HPTE_V_ABSENT)) {
295 				*hpte &= ~cpu_to_be64(HPTE_V_HVLOCK);
296 				return H_PTEG_FULL;
297 			}
298 		}
299 	}
300 
301 	/* Save away the guest's idea of the second HPTE dword */
302 	rev = &kvm->arch.revmap[pte_index];
303 	if (realmode)
304 		rev = real_vmalloc_addr(rev);
305 	if (rev) {
306 		rev->guest_rpte = g_ptel;
307 		note_hpte_modification(kvm, rev);
308 	}
309 
310 	/* Link HPTE into reverse-map chain */
311 	if (pteh & HPTE_V_VALID) {
312 		if (realmode)
313 			rmap = real_vmalloc_addr(rmap);
314 		lock_rmap(rmap);
315 		/* Check for pending invalidations under the rmap chain lock */
316 		if (mmu_notifier_retry(kvm, mmu_seq)) {
317 			/* inval in progress, write a non-present HPTE */
318 			pteh |= HPTE_V_ABSENT;
319 			pteh &= ~HPTE_V_VALID;
320 			unlock_rmap(rmap);
321 		} else {
322 			kvmppc_add_revmap_chain(kvm, rev, rmap, pte_index,
323 						realmode);
324 			/* Only set R/C in real HPTE if already set in *rmap */
325 			rcbits = *rmap >> KVMPPC_RMAP_RC_SHIFT;
326 			ptel &= rcbits | ~(HPTE_R_R | HPTE_R_C);
327 		}
328 	}
329 
330 	hpte[1] = cpu_to_be64(ptel);
331 
332 	/* Write the first HPTE dword, unlocking the HPTE and making it valid */
333 	eieio();
334 	hpte[0] = cpu_to_be64(pteh);
335 	asm volatile("ptesync" : : : "memory");
336 
337 	*pte_idx_ret = pte_index;
338 	return H_SUCCESS;
339 }
340 EXPORT_SYMBOL_GPL(kvmppc_do_h_enter);
341 
342 long kvmppc_h_enter(struct kvm_vcpu *vcpu, unsigned long flags,
343 		    long pte_index, unsigned long pteh, unsigned long ptel)
344 {
345 	return kvmppc_do_h_enter(vcpu->kvm, flags, pte_index, pteh, ptel,
346 				 vcpu->arch.pgdir, true, &vcpu->arch.gpr[4]);
347 }
348 
349 #ifdef __BIG_ENDIAN__
350 #define LOCK_TOKEN	(*(u32 *)(&get_paca()->lock_token))
351 #else
352 #define LOCK_TOKEN	(*(u32 *)(&get_paca()->paca_index))
353 #endif
354 
355 static inline int try_lock_tlbie(unsigned int *lock)
356 {
357 	unsigned int tmp, old;
358 	unsigned int token = LOCK_TOKEN;
359 
360 	asm volatile("1:lwarx	%1,0,%2\n"
361 		     "	cmpwi	cr0,%1,0\n"
362 		     "	bne	2f\n"
363 		     "  stwcx.	%3,0,%2\n"
364 		     "	bne-	1b\n"
365 		     "  isync\n"
366 		     "2:"
367 		     : "=&r" (tmp), "=&r" (old)
368 		     : "r" (lock), "r" (token)
369 		     : "cc", "memory");
370 	return old == 0;
371 }
372 
373 static void do_tlbies(struct kvm *kvm, unsigned long *rbvalues,
374 		      long npages, int global, bool need_sync)
375 {
376 	long i;
377 
378 	if (global) {
379 		while (!try_lock_tlbie(&kvm->arch.tlbie_lock))
380 			cpu_relax();
381 		if (need_sync)
382 			asm volatile("ptesync" : : : "memory");
383 		for (i = 0; i < npages; ++i)
384 			asm volatile(PPC_TLBIE(%1,%0) : :
385 				     "r" (rbvalues[i]), "r" (kvm->arch.lpid));
386 		asm volatile("eieio; tlbsync; ptesync" : : : "memory");
387 		kvm->arch.tlbie_lock = 0;
388 	} else {
389 		if (need_sync)
390 			asm volatile("ptesync" : : : "memory");
391 		for (i = 0; i < npages; ++i)
392 			asm volatile("tlbiel %0" : : "r" (rbvalues[i]));
393 		asm volatile("ptesync" : : : "memory");
394 	}
395 }
396 
397 long kvmppc_do_h_remove(struct kvm *kvm, unsigned long flags,
398 			unsigned long pte_index, unsigned long avpn,
399 			unsigned long *hpret)
400 {
401 	__be64 *hpte;
402 	unsigned long v, r, rb;
403 	struct revmap_entry *rev;
404 	u64 pte;
405 
406 	if (pte_index >= kvm->arch.hpt_npte)
407 		return H_PARAMETER;
408 	hpte = (__be64 *)(kvm->arch.hpt_virt + (pte_index << 4));
409 	while (!try_lock_hpte(hpte, HPTE_V_HVLOCK))
410 		cpu_relax();
411 	pte = be64_to_cpu(hpte[0]);
412 	if ((pte & (HPTE_V_ABSENT | HPTE_V_VALID)) == 0 ||
413 	    ((flags & H_AVPN) && (pte & ~0x7fUL) != avpn) ||
414 	    ((flags & H_ANDCOND) && (pte & avpn) != 0)) {
415 		hpte[0] &= ~cpu_to_be64(HPTE_V_HVLOCK);
416 		return H_NOT_FOUND;
417 	}
418 
419 	rev = real_vmalloc_addr(&kvm->arch.revmap[pte_index]);
420 	v = pte & ~HPTE_V_HVLOCK;
421 	if (v & HPTE_V_VALID) {
422 		u64 pte1;
423 
424 		pte1 = be64_to_cpu(hpte[1]);
425 		hpte[0] &= ~cpu_to_be64(HPTE_V_VALID);
426 		rb = compute_tlbie_rb(v, pte1, pte_index);
427 		do_tlbies(kvm, &rb, 1, global_invalidates(kvm, flags), true);
428 		/* Read PTE low word after tlbie to get final R/C values */
429 		remove_revmap_chain(kvm, pte_index, rev, v, pte1);
430 	}
431 	r = rev->guest_rpte & ~HPTE_GR_RESERVED;
432 	note_hpte_modification(kvm, rev);
433 	unlock_hpte(hpte, 0);
434 
435 	hpret[0] = v;
436 	hpret[1] = r;
437 	return H_SUCCESS;
438 }
439 EXPORT_SYMBOL_GPL(kvmppc_do_h_remove);
440 
441 long kvmppc_h_remove(struct kvm_vcpu *vcpu, unsigned long flags,
442 		     unsigned long pte_index, unsigned long avpn)
443 {
444 	return kvmppc_do_h_remove(vcpu->kvm, flags, pte_index, avpn,
445 				  &vcpu->arch.gpr[4]);
446 }
447 
448 long kvmppc_h_bulk_remove(struct kvm_vcpu *vcpu)
449 {
450 	struct kvm *kvm = vcpu->kvm;
451 	unsigned long *args = &vcpu->arch.gpr[4];
452 	__be64 *hp, *hptes[4];
453 	unsigned long tlbrb[4];
454 	long int i, j, k, n, found, indexes[4];
455 	unsigned long flags, req, pte_index, rcbits;
456 	int global;
457 	long int ret = H_SUCCESS;
458 	struct revmap_entry *rev, *revs[4];
459 	u64 hp0;
460 
461 	global = global_invalidates(kvm, 0);
462 	for (i = 0; i < 4 && ret == H_SUCCESS; ) {
463 		n = 0;
464 		for (; i < 4; ++i) {
465 			j = i * 2;
466 			pte_index = args[j];
467 			flags = pte_index >> 56;
468 			pte_index &= ((1ul << 56) - 1);
469 			req = flags >> 6;
470 			flags &= 3;
471 			if (req == 3) {		/* no more requests */
472 				i = 4;
473 				break;
474 			}
475 			if (req != 1 || flags == 3 ||
476 			    pte_index >= kvm->arch.hpt_npte) {
477 				/* parameter error */
478 				args[j] = ((0xa0 | flags) << 56) + pte_index;
479 				ret = H_PARAMETER;
480 				break;
481 			}
482 			hp = (__be64 *) (kvm->arch.hpt_virt + (pte_index << 4));
483 			/* to avoid deadlock, don't spin except for first */
484 			if (!try_lock_hpte(hp, HPTE_V_HVLOCK)) {
485 				if (n)
486 					break;
487 				while (!try_lock_hpte(hp, HPTE_V_HVLOCK))
488 					cpu_relax();
489 			}
490 			found = 0;
491 			hp0 = be64_to_cpu(hp[0]);
492 			if (hp0 & (HPTE_V_ABSENT | HPTE_V_VALID)) {
493 				switch (flags & 3) {
494 				case 0:		/* absolute */
495 					found = 1;
496 					break;
497 				case 1:		/* andcond */
498 					if (!(hp0 & args[j + 1]))
499 						found = 1;
500 					break;
501 				case 2:		/* AVPN */
502 					if ((hp0 & ~0x7fUL) == args[j + 1])
503 						found = 1;
504 					break;
505 				}
506 			}
507 			if (!found) {
508 				hp[0] &= ~cpu_to_be64(HPTE_V_HVLOCK);
509 				args[j] = ((0x90 | flags) << 56) + pte_index;
510 				continue;
511 			}
512 
513 			args[j] = ((0x80 | flags) << 56) + pte_index;
514 			rev = real_vmalloc_addr(&kvm->arch.revmap[pte_index]);
515 			note_hpte_modification(kvm, rev);
516 
517 			if (!(hp0 & HPTE_V_VALID)) {
518 				/* insert R and C bits from PTE */
519 				rcbits = rev->guest_rpte & (HPTE_R_R|HPTE_R_C);
520 				args[j] |= rcbits << (56 - 5);
521 				hp[0] = 0;
522 				continue;
523 			}
524 
525 			/* leave it locked */
526 			hp[0] &= ~cpu_to_be64(HPTE_V_VALID);
527 			tlbrb[n] = compute_tlbie_rb(be64_to_cpu(hp[0]),
528 				be64_to_cpu(hp[1]), pte_index);
529 			indexes[n] = j;
530 			hptes[n] = hp;
531 			revs[n] = rev;
532 			++n;
533 		}
534 
535 		if (!n)
536 			break;
537 
538 		/* Now that we've collected a batch, do the tlbies */
539 		do_tlbies(kvm, tlbrb, n, global, true);
540 
541 		/* Read PTE low words after tlbie to get final R/C values */
542 		for (k = 0; k < n; ++k) {
543 			j = indexes[k];
544 			pte_index = args[j] & ((1ul << 56) - 1);
545 			hp = hptes[k];
546 			rev = revs[k];
547 			remove_revmap_chain(kvm, pte_index, rev,
548 				be64_to_cpu(hp[0]), be64_to_cpu(hp[1]));
549 			rcbits = rev->guest_rpte & (HPTE_R_R|HPTE_R_C);
550 			args[j] |= rcbits << (56 - 5);
551 			hp[0] = 0;
552 		}
553 	}
554 
555 	return ret;
556 }
557 
558 long kvmppc_h_protect(struct kvm_vcpu *vcpu, unsigned long flags,
559 		      unsigned long pte_index, unsigned long avpn,
560 		      unsigned long va)
561 {
562 	struct kvm *kvm = vcpu->kvm;
563 	__be64 *hpte;
564 	struct revmap_entry *rev;
565 	unsigned long v, r, rb, mask, bits;
566 	u64 pte;
567 
568 	if (pte_index >= kvm->arch.hpt_npte)
569 		return H_PARAMETER;
570 
571 	hpte = (__be64 *)(kvm->arch.hpt_virt + (pte_index << 4));
572 	while (!try_lock_hpte(hpte, HPTE_V_HVLOCK))
573 		cpu_relax();
574 	pte = be64_to_cpu(hpte[0]);
575 	if ((pte & (HPTE_V_ABSENT | HPTE_V_VALID)) == 0 ||
576 	    ((flags & H_AVPN) && (pte & ~0x7fUL) != avpn)) {
577 		hpte[0] &= ~cpu_to_be64(HPTE_V_HVLOCK);
578 		return H_NOT_FOUND;
579 	}
580 
581 	v = pte;
582 	bits = (flags << 55) & HPTE_R_PP0;
583 	bits |= (flags << 48) & HPTE_R_KEY_HI;
584 	bits |= flags & (HPTE_R_PP | HPTE_R_N | HPTE_R_KEY_LO);
585 
586 	/* Update guest view of 2nd HPTE dword */
587 	mask = HPTE_R_PP0 | HPTE_R_PP | HPTE_R_N |
588 		HPTE_R_KEY_HI | HPTE_R_KEY_LO;
589 	rev = real_vmalloc_addr(&kvm->arch.revmap[pte_index]);
590 	if (rev) {
591 		r = (rev->guest_rpte & ~mask) | bits;
592 		rev->guest_rpte = r;
593 		note_hpte_modification(kvm, rev);
594 	}
595 
596 	/* Update HPTE */
597 	if (v & HPTE_V_VALID) {
598 		/*
599 		 * If the page is valid, don't let it transition from
600 		 * readonly to writable.  If it should be writable, we'll
601 		 * take a trap and let the page fault code sort it out.
602 		 */
603 		pte = be64_to_cpu(hpte[1]);
604 		r = (pte & ~mask) | bits;
605 		if (hpte_is_writable(r) && !hpte_is_writable(pte))
606 			r = hpte_make_readonly(r);
607 		/* If the PTE is changing, invalidate it first */
608 		if (r != pte) {
609 			rb = compute_tlbie_rb(v, r, pte_index);
610 			hpte[0] = cpu_to_be64((v & ~HPTE_V_VALID) |
611 					      HPTE_V_ABSENT);
612 			do_tlbies(kvm, &rb, 1, global_invalidates(kvm, flags),
613 				  true);
614 			hpte[1] = cpu_to_be64(r);
615 		}
616 	}
617 	unlock_hpte(hpte, v & ~HPTE_V_HVLOCK);
618 	asm volatile("ptesync" : : : "memory");
619 	return H_SUCCESS;
620 }
621 
622 long kvmppc_h_read(struct kvm_vcpu *vcpu, unsigned long flags,
623 		   unsigned long pte_index)
624 {
625 	struct kvm *kvm = vcpu->kvm;
626 	__be64 *hpte;
627 	unsigned long v, r;
628 	int i, n = 1;
629 	struct revmap_entry *rev = NULL;
630 
631 	if (pte_index >= kvm->arch.hpt_npte)
632 		return H_PARAMETER;
633 	if (flags & H_READ_4) {
634 		pte_index &= ~3;
635 		n = 4;
636 	}
637 	rev = real_vmalloc_addr(&kvm->arch.revmap[pte_index]);
638 	for (i = 0; i < n; ++i, ++pte_index) {
639 		hpte = (__be64 *)(kvm->arch.hpt_virt + (pte_index << 4));
640 		v = be64_to_cpu(hpte[0]) & ~HPTE_V_HVLOCK;
641 		r = be64_to_cpu(hpte[1]);
642 		if (v & HPTE_V_ABSENT) {
643 			v &= ~HPTE_V_ABSENT;
644 			v |= HPTE_V_VALID;
645 		}
646 		if (v & HPTE_V_VALID) {
647 			r = rev[i].guest_rpte | (r & (HPTE_R_R | HPTE_R_C));
648 			r &= ~HPTE_GR_RESERVED;
649 		}
650 		vcpu->arch.gpr[4 + i * 2] = v;
651 		vcpu->arch.gpr[5 + i * 2] = r;
652 	}
653 	return H_SUCCESS;
654 }
655 
656 void kvmppc_invalidate_hpte(struct kvm *kvm, __be64 *hptep,
657 			unsigned long pte_index)
658 {
659 	unsigned long rb;
660 
661 	hptep[0] &= ~cpu_to_be64(HPTE_V_VALID);
662 	rb = compute_tlbie_rb(be64_to_cpu(hptep[0]), be64_to_cpu(hptep[1]),
663 			      pte_index);
664 	do_tlbies(kvm, &rb, 1, 1, true);
665 }
666 EXPORT_SYMBOL_GPL(kvmppc_invalidate_hpte);
667 
668 void kvmppc_clear_ref_hpte(struct kvm *kvm, __be64 *hptep,
669 			   unsigned long pte_index)
670 {
671 	unsigned long rb;
672 	unsigned char rbyte;
673 
674 	rb = compute_tlbie_rb(be64_to_cpu(hptep[0]), be64_to_cpu(hptep[1]),
675 			      pte_index);
676 	rbyte = (be64_to_cpu(hptep[1]) & ~HPTE_R_R) >> 8;
677 	/* modify only the second-last byte, which contains the ref bit */
678 	*((char *)hptep + 14) = rbyte;
679 	do_tlbies(kvm, &rb, 1, 1, false);
680 }
681 EXPORT_SYMBOL_GPL(kvmppc_clear_ref_hpte);
682 
683 static int slb_base_page_shift[4] = {
684 	24,	/* 16M */
685 	16,	/* 64k */
686 	34,	/* 16G */
687 	20,	/* 1M, unsupported */
688 };
689 
690 /* When called from virtmode, this func should be protected by
691  * preempt_disable(), otherwise, the holding of HPTE_V_HVLOCK
692  * can trigger deadlock issue.
693  */
694 long kvmppc_hv_find_lock_hpte(struct kvm *kvm, gva_t eaddr, unsigned long slb_v,
695 			      unsigned long valid)
696 {
697 	unsigned int i;
698 	unsigned int pshift;
699 	unsigned long somask;
700 	unsigned long vsid, hash;
701 	unsigned long avpn;
702 	__be64 *hpte;
703 	unsigned long mask, val;
704 	unsigned long v, r;
705 
706 	/* Get page shift, work out hash and AVPN etc. */
707 	mask = SLB_VSID_B | HPTE_V_AVPN | HPTE_V_SECONDARY;
708 	val = 0;
709 	pshift = 12;
710 	if (slb_v & SLB_VSID_L) {
711 		mask |= HPTE_V_LARGE;
712 		val |= HPTE_V_LARGE;
713 		pshift = slb_base_page_shift[(slb_v & SLB_VSID_LP) >> 4];
714 	}
715 	if (slb_v & SLB_VSID_B_1T) {
716 		somask = (1UL << 40) - 1;
717 		vsid = (slb_v & ~SLB_VSID_B) >> SLB_VSID_SHIFT_1T;
718 		vsid ^= vsid << 25;
719 	} else {
720 		somask = (1UL << 28) - 1;
721 		vsid = (slb_v & ~SLB_VSID_B) >> SLB_VSID_SHIFT;
722 	}
723 	hash = (vsid ^ ((eaddr & somask) >> pshift)) & kvm->arch.hpt_mask;
724 	avpn = slb_v & ~(somask >> 16);	/* also includes B */
725 	avpn |= (eaddr & somask) >> 16;
726 
727 	if (pshift >= 24)
728 		avpn &= ~((1UL << (pshift - 16)) - 1);
729 	else
730 		avpn &= ~0x7fUL;
731 	val |= avpn;
732 
733 	for (;;) {
734 		hpte = (__be64 *)(kvm->arch.hpt_virt + (hash << 7));
735 
736 		for (i = 0; i < 16; i += 2) {
737 			/* Read the PTE racily */
738 			v = be64_to_cpu(hpte[i]) & ~HPTE_V_HVLOCK;
739 
740 			/* Check valid/absent, hash, segment size and AVPN */
741 			if (!(v & valid) || (v & mask) != val)
742 				continue;
743 
744 			/* Lock the PTE and read it under the lock */
745 			while (!try_lock_hpte(&hpte[i], HPTE_V_HVLOCK))
746 				cpu_relax();
747 			v = be64_to_cpu(hpte[i]) & ~HPTE_V_HVLOCK;
748 			r = be64_to_cpu(hpte[i+1]);
749 
750 			/*
751 			 * Check the HPTE again, including base page size
752 			 */
753 			if ((v & valid) && (v & mask) == val &&
754 			    hpte_base_page_size(v, r) == (1ul << pshift))
755 				/* Return with the HPTE still locked */
756 				return (hash << 3) + (i >> 1);
757 
758 			/* Unlock and move on */
759 			hpte[i] = cpu_to_be64(v);
760 		}
761 
762 		if (val & HPTE_V_SECONDARY)
763 			break;
764 		val |= HPTE_V_SECONDARY;
765 		hash = hash ^ kvm->arch.hpt_mask;
766 	}
767 	return -1;
768 }
769 EXPORT_SYMBOL(kvmppc_hv_find_lock_hpte);
770 
771 /*
772  * Called in real mode to check whether an HPTE not found fault
773  * is due to accessing a paged-out page or an emulated MMIO page,
774  * or if a protection fault is due to accessing a page that the
775  * guest wanted read/write access to but which we made read-only.
776  * Returns a possibly modified status (DSISR) value if not
777  * (i.e. pass the interrupt to the guest),
778  * -1 to pass the fault up to host kernel mode code, -2 to do that
779  * and also load the instruction word (for MMIO emulation),
780  * or 0 if we should make the guest retry the access.
781  */
782 long kvmppc_hpte_hv_fault(struct kvm_vcpu *vcpu, unsigned long addr,
783 			  unsigned long slb_v, unsigned int status, bool data)
784 {
785 	struct kvm *kvm = vcpu->kvm;
786 	long int index;
787 	unsigned long v, r, gr;
788 	__be64 *hpte;
789 	unsigned long valid;
790 	struct revmap_entry *rev;
791 	unsigned long pp, key;
792 
793 	/* For protection fault, expect to find a valid HPTE */
794 	valid = HPTE_V_VALID;
795 	if (status & DSISR_NOHPTE)
796 		valid |= HPTE_V_ABSENT;
797 
798 	index = kvmppc_hv_find_lock_hpte(kvm, addr, slb_v, valid);
799 	if (index < 0) {
800 		if (status & DSISR_NOHPTE)
801 			return status;	/* there really was no HPTE */
802 		return 0;		/* for prot fault, HPTE disappeared */
803 	}
804 	hpte = (__be64 *)(kvm->arch.hpt_virt + (index << 4));
805 	v = be64_to_cpu(hpte[0]) & ~HPTE_V_HVLOCK;
806 	r = be64_to_cpu(hpte[1]);
807 	rev = real_vmalloc_addr(&kvm->arch.revmap[index]);
808 	gr = rev->guest_rpte;
809 
810 	unlock_hpte(hpte, v);
811 
812 	/* For not found, if the HPTE is valid by now, retry the instruction */
813 	if ((status & DSISR_NOHPTE) && (v & HPTE_V_VALID))
814 		return 0;
815 
816 	/* Check access permissions to the page */
817 	pp = gr & (HPTE_R_PP0 | HPTE_R_PP);
818 	key = (vcpu->arch.shregs.msr & MSR_PR) ? SLB_VSID_KP : SLB_VSID_KS;
819 	status &= ~DSISR_NOHPTE;	/* DSISR_NOHPTE == SRR1_ISI_NOPT */
820 	if (!data) {
821 		if (gr & (HPTE_R_N | HPTE_R_G))
822 			return status | SRR1_ISI_N_OR_G;
823 		if (!hpte_read_permission(pp, slb_v & key))
824 			return status | SRR1_ISI_PROT;
825 	} else if (status & DSISR_ISSTORE) {
826 		/* check write permission */
827 		if (!hpte_write_permission(pp, slb_v & key))
828 			return status | DSISR_PROTFAULT;
829 	} else {
830 		if (!hpte_read_permission(pp, slb_v & key))
831 			return status | DSISR_PROTFAULT;
832 	}
833 
834 	/* Check storage key, if applicable */
835 	if (data && (vcpu->arch.shregs.msr & MSR_DR)) {
836 		unsigned int perm = hpte_get_skey_perm(gr, vcpu->arch.amr);
837 		if (status & DSISR_ISSTORE)
838 			perm >>= 1;
839 		if (perm & 1)
840 			return status | DSISR_KEYFAULT;
841 	}
842 
843 	/* Save HPTE info for virtual-mode handler */
844 	vcpu->arch.pgfault_addr = addr;
845 	vcpu->arch.pgfault_index = index;
846 	vcpu->arch.pgfault_hpte[0] = v;
847 	vcpu->arch.pgfault_hpte[1] = r;
848 
849 	/* Check the storage key to see if it is possibly emulated MMIO */
850 	if (data && (vcpu->arch.shregs.msr & MSR_IR) &&
851 	    (r & (HPTE_R_KEY_HI | HPTE_R_KEY_LO)) ==
852 	    (HPTE_R_KEY_HI | HPTE_R_KEY_LO))
853 		return -2;	/* MMIO emulation - load instr word */
854 
855 	return -1;		/* send fault up to host kernel mode */
856 }
857