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 	 * If we're not using MMU notifiers, we never take pages away
49 	 * from the guest, so we can use tlbiel if requested.
50 	 * Otherwise, don't use tlbiel.
51 	 */
52 	if (kvm->arch.online_vcores == 1 && local_paca->kvm_hstate.kvm_vcpu)
53 		global = 0;
54 	else if (kvm->arch.using_mmu_notifiers)
55 		global = 1;
56 	else
57 		global = !(flags & H_LOCAL);
58 
59 	if (!global) {
60 		/* any other core might now have stale TLB entries... */
61 		smp_wmb();
62 		cpumask_setall(&kvm->arch.need_tlb_flush);
63 		cpumask_clear_cpu(local_paca->kvm_hstate.kvm_vcore->pcpu,
64 				  &kvm->arch.need_tlb_flush);
65 	}
66 
67 	return global;
68 }
69 
70 /*
71  * Add this HPTE into the chain for the real page.
72  * Must be called with the chain locked; it unlocks the chain.
73  */
74 void kvmppc_add_revmap_chain(struct kvm *kvm, struct revmap_entry *rev,
75 			     unsigned long *rmap, long pte_index, int realmode)
76 {
77 	struct revmap_entry *head, *tail;
78 	unsigned long i;
79 
80 	if (*rmap & KVMPPC_RMAP_PRESENT) {
81 		i = *rmap & KVMPPC_RMAP_INDEX;
82 		head = &kvm->arch.revmap[i];
83 		if (realmode)
84 			head = real_vmalloc_addr(head);
85 		tail = &kvm->arch.revmap[head->back];
86 		if (realmode)
87 			tail = real_vmalloc_addr(tail);
88 		rev->forw = i;
89 		rev->back = head->back;
90 		tail->forw = pte_index;
91 		head->back = pte_index;
92 	} else {
93 		rev->forw = rev->back = pte_index;
94 		*rmap = (*rmap & ~KVMPPC_RMAP_INDEX) |
95 			pte_index | KVMPPC_RMAP_PRESENT;
96 	}
97 	unlock_rmap(rmap);
98 }
99 EXPORT_SYMBOL_GPL(kvmppc_add_revmap_chain);
100 
101 /* Remove this HPTE from the chain for a real page */
102 static void remove_revmap_chain(struct kvm *kvm, long pte_index,
103 				struct revmap_entry *rev,
104 				unsigned long hpte_v, unsigned long hpte_r)
105 {
106 	struct revmap_entry *next, *prev;
107 	unsigned long gfn, ptel, head;
108 	struct kvm_memory_slot *memslot;
109 	unsigned long *rmap;
110 	unsigned long rcbits;
111 
112 	rcbits = hpte_r & (HPTE_R_R | HPTE_R_C);
113 	ptel = rev->guest_rpte |= rcbits;
114 	gfn = hpte_rpn(ptel, hpte_page_size(hpte_v, ptel));
115 	memslot = __gfn_to_memslot(kvm_memslots_raw(kvm), gfn);
116 	if (!memslot)
117 		return;
118 
119 	rmap = real_vmalloc_addr(&memslot->arch.rmap[gfn - memslot->base_gfn]);
120 	lock_rmap(rmap);
121 
122 	head = *rmap & KVMPPC_RMAP_INDEX;
123 	next = real_vmalloc_addr(&kvm->arch.revmap[rev->forw]);
124 	prev = real_vmalloc_addr(&kvm->arch.revmap[rev->back]);
125 	next->back = rev->back;
126 	prev->forw = rev->forw;
127 	if (head == pte_index) {
128 		head = rev->forw;
129 		if (head == pte_index)
130 			*rmap &= ~(KVMPPC_RMAP_PRESENT | KVMPPC_RMAP_INDEX);
131 		else
132 			*rmap = (*rmap & ~KVMPPC_RMAP_INDEX) | head;
133 	}
134 	*rmap |= rcbits << KVMPPC_RMAP_RC_SHIFT;
135 	unlock_rmap(rmap);
136 }
137 
138 static pte_t lookup_linux_pte_and_update(pgd_t *pgdir, unsigned long hva,
139 			      int writing, unsigned long *pte_sizep)
140 {
141 	pte_t *ptep;
142 	unsigned long ps = *pte_sizep;
143 	unsigned int hugepage_shift;
144 
145 	ptep = find_linux_pte_or_hugepte(pgdir, hva, &hugepage_shift);
146 	if (!ptep)
147 		return __pte(0);
148 	if (hugepage_shift)
149 		*pte_sizep = 1ul << hugepage_shift;
150 	else
151 		*pte_sizep = PAGE_SIZE;
152 	if (ps > *pte_sizep)
153 		return __pte(0);
154 	return kvmppc_read_update_linux_pte(ptep, writing, hugepage_shift);
155 }
156 
157 static inline void unlock_hpte(__be64 *hpte, unsigned long hpte_v)
158 {
159 	asm volatile(PPC_RELEASE_BARRIER "" : : : "memory");
160 	hpte[0] = cpu_to_be64(hpte_v);
161 }
162 
163 long kvmppc_do_h_enter(struct kvm *kvm, unsigned long flags,
164 		       long pte_index, unsigned long pteh, unsigned long ptel,
165 		       pgd_t *pgdir, bool realmode, unsigned long *pte_idx_ret)
166 {
167 	unsigned long i, pa, gpa, gfn, psize;
168 	unsigned long slot_fn, hva;
169 	__be64 *hpte;
170 	struct revmap_entry *rev;
171 	unsigned long g_ptel;
172 	struct kvm_memory_slot *memslot;
173 	unsigned long *physp, pte_size;
174 	unsigned long is_io;
175 	unsigned long *rmap;
176 	pte_t pte;
177 	unsigned int writing;
178 	unsigned long mmu_seq;
179 	unsigned long rcbits;
180 
181 	psize = hpte_page_size(pteh, ptel);
182 	if (!psize)
183 		return H_PARAMETER;
184 	writing = hpte_is_writable(ptel);
185 	pteh &= ~(HPTE_V_HVLOCK | HPTE_V_ABSENT | HPTE_V_VALID);
186 	ptel &= ~HPTE_GR_RESERVED;
187 	g_ptel = ptel;
188 
189 	/* used later to detect if we might have been invalidated */
190 	mmu_seq = kvm->mmu_notifier_seq;
191 	smp_rmb();
192 
193 	/* Find the memslot (if any) for this address */
194 	gpa = (ptel & HPTE_R_RPN) & ~(psize - 1);
195 	gfn = gpa >> PAGE_SHIFT;
196 	memslot = __gfn_to_memslot(kvm_memslots_raw(kvm), gfn);
197 	pa = 0;
198 	is_io = ~0ul;
199 	rmap = NULL;
200 	if (!(memslot && !(memslot->flags & KVM_MEMSLOT_INVALID))) {
201 		/* PPC970 can't do emulated MMIO */
202 		if (!cpu_has_feature(CPU_FTR_ARCH_206))
203 			return H_PARAMETER;
204 		/* Emulated MMIO - mark this with key=31 */
205 		pteh |= HPTE_V_ABSENT;
206 		ptel |= HPTE_R_KEY_HI | HPTE_R_KEY_LO;
207 		goto do_insert;
208 	}
209 
210 	/* Check if the requested page fits entirely in the memslot. */
211 	if (!slot_is_aligned(memslot, psize))
212 		return H_PARAMETER;
213 	slot_fn = gfn - memslot->base_gfn;
214 	rmap = &memslot->arch.rmap[slot_fn];
215 
216 	if (!kvm->arch.using_mmu_notifiers) {
217 		physp = memslot->arch.slot_phys;
218 		if (!physp)
219 			return H_PARAMETER;
220 		physp += slot_fn;
221 		if (realmode)
222 			physp = real_vmalloc_addr(physp);
223 		pa = *physp;
224 		if (!pa)
225 			return H_TOO_HARD;
226 		is_io = pa & (HPTE_R_I | HPTE_R_W);
227 		pte_size = PAGE_SIZE << (pa & KVMPPC_PAGE_ORDER_MASK);
228 		pa &= PAGE_MASK;
229 		pa |= gpa & ~PAGE_MASK;
230 	} else {
231 		/* Translate to host virtual address */
232 		hva = __gfn_to_hva_memslot(memslot, gfn);
233 
234 		/* Look up the Linux PTE for the backing page */
235 		pte_size = psize;
236 		pte = lookup_linux_pte_and_update(pgdir, hva, writing,
237 						  &pte_size);
238 		if (pte_present(pte) && !pte_numa(pte)) {
239 			if (writing && !pte_write(pte))
240 				/* make the actual HPTE be read-only */
241 				ptel = hpte_make_readonly(ptel);
242 			is_io = hpte_cache_bits(pte_val(pte));
243 			pa = pte_pfn(pte) << PAGE_SHIFT;
244 			pa |= hva & (pte_size - 1);
245 			pa |= gpa & ~PAGE_MASK;
246 		}
247 	}
248 
249 	if (pte_size < psize)
250 		return H_PARAMETER;
251 
252 	ptel &= ~(HPTE_R_PP0 - psize);
253 	ptel |= pa;
254 
255 	if (pa)
256 		pteh |= HPTE_V_VALID;
257 	else
258 		pteh |= HPTE_V_ABSENT;
259 
260 	/* Check WIMG */
261 	if (is_io != ~0ul && !hpte_cache_flags_ok(ptel, is_io)) {
262 		if (is_io)
263 			return H_PARAMETER;
264 		/*
265 		 * Allow guest to map emulated device memory as
266 		 * uncacheable, but actually make it cacheable.
267 		 */
268 		ptel &= ~(HPTE_R_W|HPTE_R_I|HPTE_R_G);
269 		ptel |= HPTE_R_M;
270 	}
271 
272 	/* Find and lock the HPTEG slot to use */
273  do_insert:
274 	if (pte_index >= kvm->arch.hpt_npte)
275 		return H_PARAMETER;
276 	if (likely((flags & H_EXACT) == 0)) {
277 		pte_index &= ~7UL;
278 		hpte = (__be64 *)(kvm->arch.hpt_virt + (pte_index << 4));
279 		for (i = 0; i < 8; ++i) {
280 			if ((be64_to_cpu(*hpte) & HPTE_V_VALID) == 0 &&
281 			    try_lock_hpte(hpte, HPTE_V_HVLOCK | HPTE_V_VALID |
282 					  HPTE_V_ABSENT))
283 				break;
284 			hpte += 2;
285 		}
286 		if (i == 8) {
287 			/*
288 			 * Since try_lock_hpte doesn't retry (not even stdcx.
289 			 * failures), it could be that there is a free slot
290 			 * but we transiently failed to lock it.  Try again,
291 			 * actually locking each slot and checking it.
292 			 */
293 			hpte -= 16;
294 			for (i = 0; i < 8; ++i) {
295 				u64 pte;
296 				while (!try_lock_hpte(hpte, HPTE_V_HVLOCK))
297 					cpu_relax();
298 				pte = be64_to_cpu(*hpte);
299 				if (!(pte & (HPTE_V_VALID | HPTE_V_ABSENT)))
300 					break;
301 				*hpte &= ~cpu_to_be64(HPTE_V_HVLOCK);
302 				hpte += 2;
303 			}
304 			if (i == 8)
305 				return H_PTEG_FULL;
306 		}
307 		pte_index += i;
308 	} else {
309 		hpte = (__be64 *)(kvm->arch.hpt_virt + (pte_index << 4));
310 		if (!try_lock_hpte(hpte, HPTE_V_HVLOCK | HPTE_V_VALID |
311 				   HPTE_V_ABSENT)) {
312 			/* Lock the slot and check again */
313 			u64 pte;
314 
315 			while (!try_lock_hpte(hpte, HPTE_V_HVLOCK))
316 				cpu_relax();
317 			pte = be64_to_cpu(*hpte);
318 			if (pte & (HPTE_V_VALID | HPTE_V_ABSENT)) {
319 				*hpte &= ~cpu_to_be64(HPTE_V_HVLOCK);
320 				return H_PTEG_FULL;
321 			}
322 		}
323 	}
324 
325 	/* Save away the guest's idea of the second HPTE dword */
326 	rev = &kvm->arch.revmap[pte_index];
327 	if (realmode)
328 		rev = real_vmalloc_addr(rev);
329 	if (rev) {
330 		rev->guest_rpte = g_ptel;
331 		note_hpte_modification(kvm, rev);
332 	}
333 
334 	/* Link HPTE into reverse-map chain */
335 	if (pteh & HPTE_V_VALID) {
336 		if (realmode)
337 			rmap = real_vmalloc_addr(rmap);
338 		lock_rmap(rmap);
339 		/* Check for pending invalidations under the rmap chain lock */
340 		if (kvm->arch.using_mmu_notifiers &&
341 		    mmu_notifier_retry(kvm, mmu_seq)) {
342 			/* inval in progress, write a non-present HPTE */
343 			pteh |= HPTE_V_ABSENT;
344 			pteh &= ~HPTE_V_VALID;
345 			unlock_rmap(rmap);
346 		} else {
347 			kvmppc_add_revmap_chain(kvm, rev, rmap, pte_index,
348 						realmode);
349 			/* Only set R/C in real HPTE if already set in *rmap */
350 			rcbits = *rmap >> KVMPPC_RMAP_RC_SHIFT;
351 			ptel &= rcbits | ~(HPTE_R_R | HPTE_R_C);
352 		}
353 	}
354 
355 	hpte[1] = cpu_to_be64(ptel);
356 
357 	/* Write the first HPTE dword, unlocking the HPTE and making it valid */
358 	eieio();
359 	hpte[0] = cpu_to_be64(pteh);
360 	asm volatile("ptesync" : : : "memory");
361 
362 	*pte_idx_ret = pte_index;
363 	return H_SUCCESS;
364 }
365 EXPORT_SYMBOL_GPL(kvmppc_do_h_enter);
366 
367 long kvmppc_h_enter(struct kvm_vcpu *vcpu, unsigned long flags,
368 		    long pte_index, unsigned long pteh, unsigned long ptel)
369 {
370 	return kvmppc_do_h_enter(vcpu->kvm, flags, pte_index, pteh, ptel,
371 				 vcpu->arch.pgdir, true, &vcpu->arch.gpr[4]);
372 }
373 
374 #ifdef __BIG_ENDIAN__
375 #define LOCK_TOKEN	(*(u32 *)(&get_paca()->lock_token))
376 #else
377 #define LOCK_TOKEN	(*(u32 *)(&get_paca()->paca_index))
378 #endif
379 
380 static inline int try_lock_tlbie(unsigned int *lock)
381 {
382 	unsigned int tmp, old;
383 	unsigned int token = LOCK_TOKEN;
384 
385 	asm volatile("1:lwarx	%1,0,%2\n"
386 		     "	cmpwi	cr0,%1,0\n"
387 		     "	bne	2f\n"
388 		     "  stwcx.	%3,0,%2\n"
389 		     "	bne-	1b\n"
390 		     "  isync\n"
391 		     "2:"
392 		     : "=&r" (tmp), "=&r" (old)
393 		     : "r" (lock), "r" (token)
394 		     : "cc", "memory");
395 	return old == 0;
396 }
397 
398 /*
399  * tlbie/tlbiel is a bit different on the PPC970 compared to later
400  * processors such as POWER7; the large page bit is in the instruction
401  * not RB, and the top 16 bits and the bottom 12 bits of the VA
402  * in RB must be 0.
403  */
404 static void do_tlbies_970(struct kvm *kvm, unsigned long *rbvalues,
405 			  long npages, int global, bool need_sync)
406 {
407 	long i;
408 
409 	if (global) {
410 		while (!try_lock_tlbie(&kvm->arch.tlbie_lock))
411 			cpu_relax();
412 		if (need_sync)
413 			asm volatile("ptesync" : : : "memory");
414 		for (i = 0; i < npages; ++i) {
415 			unsigned long rb = rbvalues[i];
416 
417 			if (rb & 1)		/* large page */
418 				asm volatile("tlbie %0,1" : :
419 					     "r" (rb & 0x0000fffffffff000ul));
420 			else
421 				asm volatile("tlbie %0,0" : :
422 					     "r" (rb & 0x0000fffffffff000ul));
423 		}
424 		asm volatile("eieio; tlbsync; ptesync" : : : "memory");
425 		kvm->arch.tlbie_lock = 0;
426 	} else {
427 		if (need_sync)
428 			asm volatile("ptesync" : : : "memory");
429 		for (i = 0; i < npages; ++i) {
430 			unsigned long rb = rbvalues[i];
431 
432 			if (rb & 1)		/* large page */
433 				asm volatile("tlbiel %0,1" : :
434 					     "r" (rb & 0x0000fffffffff000ul));
435 			else
436 				asm volatile("tlbiel %0,0" : :
437 					     "r" (rb & 0x0000fffffffff000ul));
438 		}
439 		asm volatile("ptesync" : : : "memory");
440 	}
441 }
442 
443 static void do_tlbies(struct kvm *kvm, unsigned long *rbvalues,
444 		      long npages, int global, bool need_sync)
445 {
446 	long i;
447 
448 	if (cpu_has_feature(CPU_FTR_ARCH_201)) {
449 		/* PPC970 tlbie instruction is a bit different */
450 		do_tlbies_970(kvm, rbvalues, npages, global, need_sync);
451 		return;
452 	}
453 	if (global) {
454 		while (!try_lock_tlbie(&kvm->arch.tlbie_lock))
455 			cpu_relax();
456 		if (need_sync)
457 			asm volatile("ptesync" : : : "memory");
458 		for (i = 0; i < npages; ++i)
459 			asm volatile(PPC_TLBIE(%1,%0) : :
460 				     "r" (rbvalues[i]), "r" (kvm->arch.lpid));
461 		asm volatile("eieio; tlbsync; ptesync" : : : "memory");
462 		kvm->arch.tlbie_lock = 0;
463 	} else {
464 		if (need_sync)
465 			asm volatile("ptesync" : : : "memory");
466 		for (i = 0; i < npages; ++i)
467 			asm volatile("tlbiel %0" : : "r" (rbvalues[i]));
468 		asm volatile("ptesync" : : : "memory");
469 	}
470 }
471 
472 long kvmppc_do_h_remove(struct kvm *kvm, unsigned long flags,
473 			unsigned long pte_index, unsigned long avpn,
474 			unsigned long *hpret)
475 {
476 	__be64 *hpte;
477 	unsigned long v, r, rb;
478 	struct revmap_entry *rev;
479 	u64 pte;
480 
481 	if (pte_index >= kvm->arch.hpt_npte)
482 		return H_PARAMETER;
483 	hpte = (__be64 *)(kvm->arch.hpt_virt + (pte_index << 4));
484 	while (!try_lock_hpte(hpte, HPTE_V_HVLOCK))
485 		cpu_relax();
486 	pte = be64_to_cpu(hpte[0]);
487 	if ((pte & (HPTE_V_ABSENT | HPTE_V_VALID)) == 0 ||
488 	    ((flags & H_AVPN) && (pte & ~0x7fUL) != avpn) ||
489 	    ((flags & H_ANDCOND) && (pte & avpn) != 0)) {
490 		hpte[0] &= ~cpu_to_be64(HPTE_V_HVLOCK);
491 		return H_NOT_FOUND;
492 	}
493 
494 	rev = real_vmalloc_addr(&kvm->arch.revmap[pte_index]);
495 	v = pte & ~HPTE_V_HVLOCK;
496 	if (v & HPTE_V_VALID) {
497 		u64 pte1;
498 
499 		pte1 = be64_to_cpu(hpte[1]);
500 		hpte[0] &= ~cpu_to_be64(HPTE_V_VALID);
501 		rb = compute_tlbie_rb(v, pte1, pte_index);
502 		do_tlbies(kvm, &rb, 1, global_invalidates(kvm, flags), true);
503 		/* Read PTE low word after tlbie to get final R/C values */
504 		remove_revmap_chain(kvm, pte_index, rev, v, pte1);
505 	}
506 	r = rev->guest_rpte & ~HPTE_GR_RESERVED;
507 	note_hpte_modification(kvm, rev);
508 	unlock_hpte(hpte, 0);
509 
510 	hpret[0] = v;
511 	hpret[1] = r;
512 	return H_SUCCESS;
513 }
514 EXPORT_SYMBOL_GPL(kvmppc_do_h_remove);
515 
516 long kvmppc_h_remove(struct kvm_vcpu *vcpu, unsigned long flags,
517 		     unsigned long pte_index, unsigned long avpn)
518 {
519 	return kvmppc_do_h_remove(vcpu->kvm, flags, pte_index, avpn,
520 				  &vcpu->arch.gpr[4]);
521 }
522 
523 long kvmppc_h_bulk_remove(struct kvm_vcpu *vcpu)
524 {
525 	struct kvm *kvm = vcpu->kvm;
526 	unsigned long *args = &vcpu->arch.gpr[4];
527 	__be64 *hp, *hptes[4];
528 	unsigned long tlbrb[4];
529 	long int i, j, k, n, found, indexes[4];
530 	unsigned long flags, req, pte_index, rcbits;
531 	int global;
532 	long int ret = H_SUCCESS;
533 	struct revmap_entry *rev, *revs[4];
534 	u64 hp0;
535 
536 	global = global_invalidates(kvm, 0);
537 	for (i = 0; i < 4 && ret == H_SUCCESS; ) {
538 		n = 0;
539 		for (; i < 4; ++i) {
540 			j = i * 2;
541 			pte_index = args[j];
542 			flags = pte_index >> 56;
543 			pte_index &= ((1ul << 56) - 1);
544 			req = flags >> 6;
545 			flags &= 3;
546 			if (req == 3) {		/* no more requests */
547 				i = 4;
548 				break;
549 			}
550 			if (req != 1 || flags == 3 ||
551 			    pte_index >= kvm->arch.hpt_npte) {
552 				/* parameter error */
553 				args[j] = ((0xa0 | flags) << 56) + pte_index;
554 				ret = H_PARAMETER;
555 				break;
556 			}
557 			hp = (__be64 *) (kvm->arch.hpt_virt + (pte_index << 4));
558 			/* to avoid deadlock, don't spin except for first */
559 			if (!try_lock_hpte(hp, HPTE_V_HVLOCK)) {
560 				if (n)
561 					break;
562 				while (!try_lock_hpte(hp, HPTE_V_HVLOCK))
563 					cpu_relax();
564 			}
565 			found = 0;
566 			hp0 = be64_to_cpu(hp[0]);
567 			if (hp0 & (HPTE_V_ABSENT | HPTE_V_VALID)) {
568 				switch (flags & 3) {
569 				case 0:		/* absolute */
570 					found = 1;
571 					break;
572 				case 1:		/* andcond */
573 					if (!(hp0 & args[j + 1]))
574 						found = 1;
575 					break;
576 				case 2:		/* AVPN */
577 					if ((hp0 & ~0x7fUL) == args[j + 1])
578 						found = 1;
579 					break;
580 				}
581 			}
582 			if (!found) {
583 				hp[0] &= ~cpu_to_be64(HPTE_V_HVLOCK);
584 				args[j] = ((0x90 | flags) << 56) + pte_index;
585 				continue;
586 			}
587 
588 			args[j] = ((0x80 | flags) << 56) + pte_index;
589 			rev = real_vmalloc_addr(&kvm->arch.revmap[pte_index]);
590 			note_hpte_modification(kvm, rev);
591 
592 			if (!(hp0 & HPTE_V_VALID)) {
593 				/* insert R and C bits from PTE */
594 				rcbits = rev->guest_rpte & (HPTE_R_R|HPTE_R_C);
595 				args[j] |= rcbits << (56 - 5);
596 				hp[0] = 0;
597 				continue;
598 			}
599 
600 			/* leave it locked */
601 			hp[0] &= ~cpu_to_be64(HPTE_V_VALID);
602 			tlbrb[n] = compute_tlbie_rb(be64_to_cpu(hp[0]),
603 				be64_to_cpu(hp[1]), pte_index);
604 			indexes[n] = j;
605 			hptes[n] = hp;
606 			revs[n] = rev;
607 			++n;
608 		}
609 
610 		if (!n)
611 			break;
612 
613 		/* Now that we've collected a batch, do the tlbies */
614 		do_tlbies(kvm, tlbrb, n, global, true);
615 
616 		/* Read PTE low words after tlbie to get final R/C values */
617 		for (k = 0; k < n; ++k) {
618 			j = indexes[k];
619 			pte_index = args[j] & ((1ul << 56) - 1);
620 			hp = hptes[k];
621 			rev = revs[k];
622 			remove_revmap_chain(kvm, pte_index, rev,
623 				be64_to_cpu(hp[0]), be64_to_cpu(hp[1]));
624 			rcbits = rev->guest_rpte & (HPTE_R_R|HPTE_R_C);
625 			args[j] |= rcbits << (56 - 5);
626 			hp[0] = 0;
627 		}
628 	}
629 
630 	return ret;
631 }
632 
633 long kvmppc_h_protect(struct kvm_vcpu *vcpu, unsigned long flags,
634 		      unsigned long pte_index, unsigned long avpn,
635 		      unsigned long va)
636 {
637 	struct kvm *kvm = vcpu->kvm;
638 	__be64 *hpte;
639 	struct revmap_entry *rev;
640 	unsigned long v, r, rb, mask, bits;
641 	u64 pte;
642 
643 	if (pte_index >= kvm->arch.hpt_npte)
644 		return H_PARAMETER;
645 
646 	hpte = (__be64 *)(kvm->arch.hpt_virt + (pte_index << 4));
647 	while (!try_lock_hpte(hpte, HPTE_V_HVLOCK))
648 		cpu_relax();
649 	pte = be64_to_cpu(hpte[0]);
650 	if ((pte & (HPTE_V_ABSENT | HPTE_V_VALID)) == 0 ||
651 	    ((flags & H_AVPN) && (pte & ~0x7fUL) != avpn)) {
652 		hpte[0] &= ~cpu_to_be64(HPTE_V_HVLOCK);
653 		return H_NOT_FOUND;
654 	}
655 
656 	v = pte;
657 	bits = (flags << 55) & HPTE_R_PP0;
658 	bits |= (flags << 48) & HPTE_R_KEY_HI;
659 	bits |= flags & (HPTE_R_PP | HPTE_R_N | HPTE_R_KEY_LO);
660 
661 	/* Update guest view of 2nd HPTE dword */
662 	mask = HPTE_R_PP0 | HPTE_R_PP | HPTE_R_N |
663 		HPTE_R_KEY_HI | HPTE_R_KEY_LO;
664 	rev = real_vmalloc_addr(&kvm->arch.revmap[pte_index]);
665 	if (rev) {
666 		r = (rev->guest_rpte & ~mask) | bits;
667 		rev->guest_rpte = r;
668 		note_hpte_modification(kvm, rev);
669 	}
670 	r = (be64_to_cpu(hpte[1]) & ~mask) | bits;
671 
672 	/* Update HPTE */
673 	if (v & HPTE_V_VALID) {
674 		rb = compute_tlbie_rb(v, r, pte_index);
675 		hpte[0] = cpu_to_be64(v & ~HPTE_V_VALID);
676 		do_tlbies(kvm, &rb, 1, global_invalidates(kvm, flags), true);
677 		/*
678 		 * If the host has this page as readonly but the guest
679 		 * wants to make it read/write, reduce the permissions.
680 		 * Checking the host permissions involves finding the
681 		 * memslot and then the Linux PTE for the page.
682 		 */
683 		if (hpte_is_writable(r) && kvm->arch.using_mmu_notifiers) {
684 			unsigned long psize, gfn, hva;
685 			struct kvm_memory_slot *memslot;
686 			pgd_t *pgdir = vcpu->arch.pgdir;
687 			pte_t pte;
688 
689 			psize = hpte_page_size(v, r);
690 			gfn = ((r & HPTE_R_RPN) & ~(psize - 1)) >> PAGE_SHIFT;
691 			memslot = __gfn_to_memslot(kvm_memslots_raw(kvm), gfn);
692 			if (memslot) {
693 				hva = __gfn_to_hva_memslot(memslot, gfn);
694 				pte = lookup_linux_pte_and_update(pgdir, hva,
695 								  1, &psize);
696 				if (pte_present(pte) && !pte_write(pte))
697 					r = hpte_make_readonly(r);
698 			}
699 		}
700 	}
701 	hpte[1] = cpu_to_be64(r);
702 	eieio();
703 	hpte[0] = cpu_to_be64(v & ~HPTE_V_HVLOCK);
704 	asm volatile("ptesync" : : : "memory");
705 	return H_SUCCESS;
706 }
707 
708 long kvmppc_h_read(struct kvm_vcpu *vcpu, unsigned long flags,
709 		   unsigned long pte_index)
710 {
711 	struct kvm *kvm = vcpu->kvm;
712 	__be64 *hpte;
713 	unsigned long v, r;
714 	int i, n = 1;
715 	struct revmap_entry *rev = NULL;
716 
717 	if (pte_index >= kvm->arch.hpt_npte)
718 		return H_PARAMETER;
719 	if (flags & H_READ_4) {
720 		pte_index &= ~3;
721 		n = 4;
722 	}
723 	rev = real_vmalloc_addr(&kvm->arch.revmap[pte_index]);
724 	for (i = 0; i < n; ++i, ++pte_index) {
725 		hpte = (__be64 *)(kvm->arch.hpt_virt + (pte_index << 4));
726 		v = be64_to_cpu(hpte[0]) & ~HPTE_V_HVLOCK;
727 		r = be64_to_cpu(hpte[1]);
728 		if (v & HPTE_V_ABSENT) {
729 			v &= ~HPTE_V_ABSENT;
730 			v |= HPTE_V_VALID;
731 		}
732 		if (v & HPTE_V_VALID) {
733 			r = rev[i].guest_rpte | (r & (HPTE_R_R | HPTE_R_C));
734 			r &= ~HPTE_GR_RESERVED;
735 		}
736 		vcpu->arch.gpr[4 + i * 2] = v;
737 		vcpu->arch.gpr[5 + i * 2] = r;
738 	}
739 	return H_SUCCESS;
740 }
741 
742 void kvmppc_invalidate_hpte(struct kvm *kvm, __be64 *hptep,
743 			unsigned long pte_index)
744 {
745 	unsigned long rb;
746 
747 	hptep[0] &= ~cpu_to_be64(HPTE_V_VALID);
748 	rb = compute_tlbie_rb(be64_to_cpu(hptep[0]), be64_to_cpu(hptep[1]),
749 			      pte_index);
750 	do_tlbies(kvm, &rb, 1, 1, true);
751 }
752 EXPORT_SYMBOL_GPL(kvmppc_invalidate_hpte);
753 
754 void kvmppc_clear_ref_hpte(struct kvm *kvm, __be64 *hptep,
755 			   unsigned long pte_index)
756 {
757 	unsigned long rb;
758 	unsigned char rbyte;
759 
760 	rb = compute_tlbie_rb(be64_to_cpu(hptep[0]), be64_to_cpu(hptep[1]),
761 			      pte_index);
762 	rbyte = (be64_to_cpu(hptep[1]) & ~HPTE_R_R) >> 8;
763 	/* modify only the second-last byte, which contains the ref bit */
764 	*((char *)hptep + 14) = rbyte;
765 	do_tlbies(kvm, &rb, 1, 1, false);
766 }
767 EXPORT_SYMBOL_GPL(kvmppc_clear_ref_hpte);
768 
769 static int slb_base_page_shift[4] = {
770 	24,	/* 16M */
771 	16,	/* 64k */
772 	34,	/* 16G */
773 	20,	/* 1M, unsupported */
774 };
775 
776 /* When called from virtmode, this func should be protected by
777  * preempt_disable(), otherwise, the holding of HPTE_V_HVLOCK
778  * can trigger deadlock issue.
779  */
780 long kvmppc_hv_find_lock_hpte(struct kvm *kvm, gva_t eaddr, unsigned long slb_v,
781 			      unsigned long valid)
782 {
783 	unsigned int i;
784 	unsigned int pshift;
785 	unsigned long somask;
786 	unsigned long vsid, hash;
787 	unsigned long avpn;
788 	__be64 *hpte;
789 	unsigned long mask, val;
790 	unsigned long v, r;
791 
792 	/* Get page shift, work out hash and AVPN etc. */
793 	mask = SLB_VSID_B | HPTE_V_AVPN | HPTE_V_SECONDARY;
794 	val = 0;
795 	pshift = 12;
796 	if (slb_v & SLB_VSID_L) {
797 		mask |= HPTE_V_LARGE;
798 		val |= HPTE_V_LARGE;
799 		pshift = slb_base_page_shift[(slb_v & SLB_VSID_LP) >> 4];
800 	}
801 	if (slb_v & SLB_VSID_B_1T) {
802 		somask = (1UL << 40) - 1;
803 		vsid = (slb_v & ~SLB_VSID_B) >> SLB_VSID_SHIFT_1T;
804 		vsid ^= vsid << 25;
805 	} else {
806 		somask = (1UL << 28) - 1;
807 		vsid = (slb_v & ~SLB_VSID_B) >> SLB_VSID_SHIFT;
808 	}
809 	hash = (vsid ^ ((eaddr & somask) >> pshift)) & kvm->arch.hpt_mask;
810 	avpn = slb_v & ~(somask >> 16);	/* also includes B */
811 	avpn |= (eaddr & somask) >> 16;
812 
813 	if (pshift >= 24)
814 		avpn &= ~((1UL << (pshift - 16)) - 1);
815 	else
816 		avpn &= ~0x7fUL;
817 	val |= avpn;
818 
819 	for (;;) {
820 		hpte = (__be64 *)(kvm->arch.hpt_virt + (hash << 7));
821 
822 		for (i = 0; i < 16; i += 2) {
823 			/* Read the PTE racily */
824 			v = be64_to_cpu(hpte[i]) & ~HPTE_V_HVLOCK;
825 
826 			/* Check valid/absent, hash, segment size and AVPN */
827 			if (!(v & valid) || (v & mask) != val)
828 				continue;
829 
830 			/* Lock the PTE and read it under the lock */
831 			while (!try_lock_hpte(&hpte[i], HPTE_V_HVLOCK))
832 				cpu_relax();
833 			v = be64_to_cpu(hpte[i]) & ~HPTE_V_HVLOCK;
834 			r = be64_to_cpu(hpte[i+1]);
835 
836 			/*
837 			 * Check the HPTE again, including base page size
838 			 */
839 			if ((v & valid) && (v & mask) == val &&
840 			    hpte_base_page_size(v, r) == (1ul << pshift))
841 				/* Return with the HPTE still locked */
842 				return (hash << 3) + (i >> 1);
843 
844 			/* Unlock and move on */
845 			hpte[i] = cpu_to_be64(v);
846 		}
847 
848 		if (val & HPTE_V_SECONDARY)
849 			break;
850 		val |= HPTE_V_SECONDARY;
851 		hash = hash ^ kvm->arch.hpt_mask;
852 	}
853 	return -1;
854 }
855 EXPORT_SYMBOL(kvmppc_hv_find_lock_hpte);
856 
857 /*
858  * Called in real mode to check whether an HPTE not found fault
859  * is due to accessing a paged-out page or an emulated MMIO page,
860  * or if a protection fault is due to accessing a page that the
861  * guest wanted read/write access to but which we made read-only.
862  * Returns a possibly modified status (DSISR) value if not
863  * (i.e. pass the interrupt to the guest),
864  * -1 to pass the fault up to host kernel mode code, -2 to do that
865  * and also load the instruction word (for MMIO emulation),
866  * or 0 if we should make the guest retry the access.
867  */
868 long kvmppc_hpte_hv_fault(struct kvm_vcpu *vcpu, unsigned long addr,
869 			  unsigned long slb_v, unsigned int status, bool data)
870 {
871 	struct kvm *kvm = vcpu->kvm;
872 	long int index;
873 	unsigned long v, r, gr;
874 	__be64 *hpte;
875 	unsigned long valid;
876 	struct revmap_entry *rev;
877 	unsigned long pp, key;
878 
879 	/* For protection fault, expect to find a valid HPTE */
880 	valid = HPTE_V_VALID;
881 	if (status & DSISR_NOHPTE)
882 		valid |= HPTE_V_ABSENT;
883 
884 	index = kvmppc_hv_find_lock_hpte(kvm, addr, slb_v, valid);
885 	if (index < 0) {
886 		if (status & DSISR_NOHPTE)
887 			return status;	/* there really was no HPTE */
888 		return 0;		/* for prot fault, HPTE disappeared */
889 	}
890 	hpte = (__be64 *)(kvm->arch.hpt_virt + (index << 4));
891 	v = be64_to_cpu(hpte[0]) & ~HPTE_V_HVLOCK;
892 	r = be64_to_cpu(hpte[1]);
893 	rev = real_vmalloc_addr(&kvm->arch.revmap[index]);
894 	gr = rev->guest_rpte;
895 
896 	unlock_hpte(hpte, v);
897 
898 	/* For not found, if the HPTE is valid by now, retry the instruction */
899 	if ((status & DSISR_NOHPTE) && (v & HPTE_V_VALID))
900 		return 0;
901 
902 	/* Check access permissions to the page */
903 	pp = gr & (HPTE_R_PP0 | HPTE_R_PP);
904 	key = (vcpu->arch.shregs.msr & MSR_PR) ? SLB_VSID_KP : SLB_VSID_KS;
905 	status &= ~DSISR_NOHPTE;	/* DSISR_NOHPTE == SRR1_ISI_NOPT */
906 	if (!data) {
907 		if (gr & (HPTE_R_N | HPTE_R_G))
908 			return status | SRR1_ISI_N_OR_G;
909 		if (!hpte_read_permission(pp, slb_v & key))
910 			return status | SRR1_ISI_PROT;
911 	} else if (status & DSISR_ISSTORE) {
912 		/* check write permission */
913 		if (!hpte_write_permission(pp, slb_v & key))
914 			return status | DSISR_PROTFAULT;
915 	} else {
916 		if (!hpte_read_permission(pp, slb_v & key))
917 			return status | DSISR_PROTFAULT;
918 	}
919 
920 	/* Check storage key, if applicable */
921 	if (data && (vcpu->arch.shregs.msr & MSR_DR)) {
922 		unsigned int perm = hpte_get_skey_perm(gr, vcpu->arch.amr);
923 		if (status & DSISR_ISSTORE)
924 			perm >>= 1;
925 		if (perm & 1)
926 			return status | DSISR_KEYFAULT;
927 	}
928 
929 	/* Save HPTE info for virtual-mode handler */
930 	vcpu->arch.pgfault_addr = addr;
931 	vcpu->arch.pgfault_index = index;
932 	vcpu->arch.pgfault_hpte[0] = v;
933 	vcpu->arch.pgfault_hpte[1] = r;
934 
935 	/* Check the storage key to see if it is possibly emulated MMIO */
936 	if (data && (vcpu->arch.shregs.msr & MSR_IR) &&
937 	    (r & (HPTE_R_KEY_HI | HPTE_R_KEY_LO)) ==
938 	    (HPTE_R_KEY_HI | HPTE_R_KEY_LO))
939 		return -2;	/* MMIO emulation - load instr word */
940 
941 	return -1;		/* send fault up to host kernel mode */
942 }
943