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 	 * assume we don't have huge pages in vmalloc space...
31 	 * So don't worry about THP collapse/split. Called
32 	 * Only in realmode, hence won't need irq_save/restore.
33 	 */
34 	p = __find_linux_pte_or_hugepte(swapper_pg_dir, addr, NULL);
35 	if (!p || !pte_present(*p))
36 		return NULL;
37 	addr = (pte_pfn(*p) << PAGE_SHIFT) | (addr & ~PAGE_MASK);
38 	return __va(addr);
39 }
40 
41 /* Return 1 if we need to do a global tlbie, 0 if we can use tlbiel */
42 static int global_invalidates(struct kvm *kvm, unsigned long flags)
43 {
44 	int global;
45 
46 	/*
47 	 * If there is only one vcore, and it's currently running,
48 	 * as indicated by local_paca->kvm_hstate.kvm_vcpu being set,
49 	 * we can use tlbiel as long as we mark all other physical
50 	 * cores as potentially having stale TLB entries for this lpid.
51 	 * Otherwise, don't use tlbiel.
52 	 */
53 	if (kvm->arch.online_vcores == 1 && local_paca->kvm_hstate.kvm_vcpu)
54 		global = 0;
55 	else
56 		global = 1;
57 
58 	if (!global) {
59 		/* any other core might now have stale TLB entries... */
60 		smp_wmb();
61 		cpumask_setall(&kvm->arch.need_tlb_flush);
62 		cpumask_clear_cpu(local_paca->kvm_hstate.kvm_vcore->pcpu,
63 				  &kvm->arch.need_tlb_flush);
64 	}
65 
66 	return global;
67 }
68 
69 /*
70  * Add this HPTE into the chain for the real page.
71  * Must be called with the chain locked; it unlocks the chain.
72  */
73 void kvmppc_add_revmap_chain(struct kvm *kvm, struct revmap_entry *rev,
74 			     unsigned long *rmap, long pte_index, int realmode)
75 {
76 	struct revmap_entry *head, *tail;
77 	unsigned long i;
78 
79 	if (*rmap & KVMPPC_RMAP_PRESENT) {
80 		i = *rmap & KVMPPC_RMAP_INDEX;
81 		head = &kvm->arch.revmap[i];
82 		if (realmode)
83 			head = real_vmalloc_addr(head);
84 		tail = &kvm->arch.revmap[head->back];
85 		if (realmode)
86 			tail = real_vmalloc_addr(tail);
87 		rev->forw = i;
88 		rev->back = head->back;
89 		tail->forw = pte_index;
90 		head->back = pte_index;
91 	} else {
92 		rev->forw = rev->back = pte_index;
93 		*rmap = (*rmap & ~KVMPPC_RMAP_INDEX) |
94 			pte_index | KVMPPC_RMAP_PRESENT;
95 	}
96 	unlock_rmap(rmap);
97 }
98 EXPORT_SYMBOL_GPL(kvmppc_add_revmap_chain);
99 
100 /* Remove this HPTE from the chain for a real page */
101 static void remove_revmap_chain(struct kvm *kvm, long pte_index,
102 				struct revmap_entry *rev,
103 				unsigned long hpte_v, unsigned long hpte_r)
104 {
105 	struct revmap_entry *next, *prev;
106 	unsigned long gfn, ptel, head;
107 	struct kvm_memory_slot *memslot;
108 	unsigned long *rmap;
109 	unsigned long rcbits;
110 
111 	rcbits = hpte_r & (HPTE_R_R | HPTE_R_C);
112 	ptel = rev->guest_rpte |= rcbits;
113 	gfn = hpte_rpn(ptel, hpte_page_size(hpte_v, ptel));
114 	memslot = __gfn_to_memslot(kvm_memslots_raw(kvm), gfn);
115 	if (!memslot)
116 		return;
117 
118 	rmap = real_vmalloc_addr(&memslot->arch.rmap[gfn - memslot->base_gfn]);
119 	lock_rmap(rmap);
120 
121 	head = *rmap & KVMPPC_RMAP_INDEX;
122 	next = real_vmalloc_addr(&kvm->arch.revmap[rev->forw]);
123 	prev = real_vmalloc_addr(&kvm->arch.revmap[rev->back]);
124 	next->back = rev->back;
125 	prev->forw = rev->forw;
126 	if (head == pte_index) {
127 		head = rev->forw;
128 		if (head == pte_index)
129 			*rmap &= ~(KVMPPC_RMAP_PRESENT | KVMPPC_RMAP_INDEX);
130 		else
131 			*rmap = (*rmap & ~KVMPPC_RMAP_INDEX) | head;
132 	}
133 	*rmap |= rcbits << KVMPPC_RMAP_RC_SHIFT;
134 	unlock_rmap(rmap);
135 }
136 
137 long kvmppc_do_h_enter(struct kvm *kvm, unsigned long flags,
138 		       long pte_index, unsigned long pteh, unsigned long ptel,
139 		       pgd_t *pgdir, bool realmode, unsigned long *pte_idx_ret)
140 {
141 	unsigned long i, pa, gpa, gfn, psize;
142 	unsigned long slot_fn, hva;
143 	__be64 *hpte;
144 	struct revmap_entry *rev;
145 	unsigned long g_ptel;
146 	struct kvm_memory_slot *memslot;
147 	unsigned hpage_shift;
148 	unsigned long is_io;
149 	unsigned long *rmap;
150 	pte_t *ptep;
151 	unsigned int writing;
152 	unsigned long mmu_seq;
153 	unsigned long rcbits, irq_flags = 0;
154 
155 	psize = hpte_page_size(pteh, ptel);
156 	if (!psize)
157 		return H_PARAMETER;
158 	writing = hpte_is_writable(ptel);
159 	pteh &= ~(HPTE_V_HVLOCK | HPTE_V_ABSENT | HPTE_V_VALID);
160 	ptel &= ~HPTE_GR_RESERVED;
161 	g_ptel = ptel;
162 
163 	/* used later to detect if we might have been invalidated */
164 	mmu_seq = kvm->mmu_notifier_seq;
165 	smp_rmb();
166 
167 	/* Find the memslot (if any) for this address */
168 	gpa = (ptel & HPTE_R_RPN) & ~(psize - 1);
169 	gfn = gpa >> PAGE_SHIFT;
170 	memslot = __gfn_to_memslot(kvm_memslots_raw(kvm), gfn);
171 	pa = 0;
172 	is_io = ~0ul;
173 	rmap = NULL;
174 	if (!(memslot && !(memslot->flags & KVM_MEMSLOT_INVALID))) {
175 		/* Emulated MMIO - mark this with key=31 */
176 		pteh |= HPTE_V_ABSENT;
177 		ptel |= HPTE_R_KEY_HI | HPTE_R_KEY_LO;
178 		goto do_insert;
179 	}
180 
181 	/* Check if the requested page fits entirely in the memslot. */
182 	if (!slot_is_aligned(memslot, psize))
183 		return H_PARAMETER;
184 	slot_fn = gfn - memslot->base_gfn;
185 	rmap = &memslot->arch.rmap[slot_fn];
186 
187 	/* Translate to host virtual address */
188 	hva = __gfn_to_hva_memslot(memslot, gfn);
189 	/*
190 	 * If we had a page table table change after lookup, we would
191 	 * retry via mmu_notifier_retry.
192 	 */
193 	if (realmode)
194 		ptep = __find_linux_pte_or_hugepte(pgdir, hva, &hpage_shift);
195 	else {
196 		local_irq_save(irq_flags);
197 		ptep = find_linux_pte_or_hugepte(pgdir, hva, &hpage_shift);
198 	}
199 	if (ptep) {
200 		pte_t pte;
201 		unsigned int host_pte_size;
202 
203 		if (hpage_shift)
204 			host_pte_size = 1ul << hpage_shift;
205 		else
206 			host_pte_size = PAGE_SIZE;
207 		/*
208 		 * We should always find the guest page size
209 		 * to <= host page size, if host is using hugepage
210 		 */
211 		if (host_pte_size < psize) {
212 			if (!realmode)
213 				local_irq_restore(flags);
214 			return H_PARAMETER;
215 		}
216 		pte = kvmppc_read_update_linux_pte(ptep, writing);
217 		if (pte_present(pte) && !pte_protnone(pte)) {
218 			if (writing && !pte_write(pte))
219 				/* make the actual HPTE be read-only */
220 				ptel = hpte_make_readonly(ptel);
221 			is_io = hpte_cache_bits(pte_val(pte));
222 			pa = pte_pfn(pte) << PAGE_SHIFT;
223 			pa |= hva & (host_pte_size - 1);
224 			pa |= gpa & ~PAGE_MASK;
225 		}
226 	}
227 	if (!realmode)
228 		local_irq_restore(irq_flags);
229 
230 	ptel &= ~(HPTE_R_PP0 - psize);
231 	ptel |= pa;
232 
233 	if (pa)
234 		pteh |= HPTE_V_VALID;
235 	else
236 		pteh |= HPTE_V_ABSENT;
237 
238 	/* Check WIMG */
239 	if (is_io != ~0ul && !hpte_cache_flags_ok(ptel, is_io)) {
240 		if (is_io)
241 			return H_PARAMETER;
242 		/*
243 		 * Allow guest to map emulated device memory as
244 		 * uncacheable, but actually make it cacheable.
245 		 */
246 		ptel &= ~(HPTE_R_W|HPTE_R_I|HPTE_R_G);
247 		ptel |= HPTE_R_M;
248 	}
249 
250 	/* Find and lock the HPTEG slot to use */
251  do_insert:
252 	if (pte_index >= kvm->arch.hpt_npte)
253 		return H_PARAMETER;
254 	if (likely((flags & H_EXACT) == 0)) {
255 		pte_index &= ~7UL;
256 		hpte = (__be64 *)(kvm->arch.hpt_virt + (pte_index << 4));
257 		for (i = 0; i < 8; ++i) {
258 			if ((be64_to_cpu(*hpte) & HPTE_V_VALID) == 0 &&
259 			    try_lock_hpte(hpte, HPTE_V_HVLOCK | HPTE_V_VALID |
260 					  HPTE_V_ABSENT))
261 				break;
262 			hpte += 2;
263 		}
264 		if (i == 8) {
265 			/*
266 			 * Since try_lock_hpte doesn't retry (not even stdcx.
267 			 * failures), it could be that there is a free slot
268 			 * but we transiently failed to lock it.  Try again,
269 			 * actually locking each slot and checking it.
270 			 */
271 			hpte -= 16;
272 			for (i = 0; i < 8; ++i) {
273 				u64 pte;
274 				while (!try_lock_hpte(hpte, HPTE_V_HVLOCK))
275 					cpu_relax();
276 				pte = be64_to_cpu(hpte[0]);
277 				if (!(pte & (HPTE_V_VALID | HPTE_V_ABSENT)))
278 					break;
279 				__unlock_hpte(hpte, pte);
280 				hpte += 2;
281 			}
282 			if (i == 8)
283 				return H_PTEG_FULL;
284 		}
285 		pte_index += i;
286 	} else {
287 		hpte = (__be64 *)(kvm->arch.hpt_virt + (pte_index << 4));
288 		if (!try_lock_hpte(hpte, HPTE_V_HVLOCK | HPTE_V_VALID |
289 				   HPTE_V_ABSENT)) {
290 			/* Lock the slot and check again */
291 			u64 pte;
292 
293 			while (!try_lock_hpte(hpte, HPTE_V_HVLOCK))
294 				cpu_relax();
295 			pte = be64_to_cpu(hpte[0]);
296 			if (pte & (HPTE_V_VALID | HPTE_V_ABSENT)) {
297 				__unlock_hpte(hpte, pte);
298 				return H_PTEG_FULL;
299 			}
300 		}
301 	}
302 
303 	/* Save away the guest's idea of the second HPTE dword */
304 	rev = &kvm->arch.revmap[pte_index];
305 	if (realmode)
306 		rev = real_vmalloc_addr(rev);
307 	if (rev) {
308 		rev->guest_rpte = g_ptel;
309 		note_hpte_modification(kvm, rev);
310 	}
311 
312 	/* Link HPTE into reverse-map chain */
313 	if (pteh & HPTE_V_VALID) {
314 		if (realmode)
315 			rmap = real_vmalloc_addr(rmap);
316 		lock_rmap(rmap);
317 		/* Check for pending invalidations under the rmap chain lock */
318 		if (mmu_notifier_retry(kvm, mmu_seq)) {
319 			/* inval in progress, write a non-present HPTE */
320 			pteh |= HPTE_V_ABSENT;
321 			pteh &= ~HPTE_V_VALID;
322 			unlock_rmap(rmap);
323 		} else {
324 			kvmppc_add_revmap_chain(kvm, rev, rmap, pte_index,
325 						realmode);
326 			/* Only set R/C in real HPTE if already set in *rmap */
327 			rcbits = *rmap >> KVMPPC_RMAP_RC_SHIFT;
328 			ptel &= rcbits | ~(HPTE_R_R | HPTE_R_C);
329 		}
330 	}
331 
332 	hpte[1] = cpu_to_be64(ptel);
333 
334 	/* Write the first HPTE dword, unlocking the HPTE and making it valid */
335 	eieio();
336 	__unlock_hpte(hpte, pteh);
337 	asm volatile("ptesync" : : : "memory");
338 
339 	*pte_idx_ret = pte_index;
340 	return H_SUCCESS;
341 }
342 EXPORT_SYMBOL_GPL(kvmppc_do_h_enter);
343 
344 long kvmppc_h_enter(struct kvm_vcpu *vcpu, unsigned long flags,
345 		    long pte_index, unsigned long pteh, unsigned long ptel)
346 {
347 	return kvmppc_do_h_enter(vcpu->kvm, flags, pte_index, pteh, ptel,
348 				 vcpu->arch.pgdir, true, &vcpu->arch.gpr[4]);
349 }
350 
351 #ifdef __BIG_ENDIAN__
352 #define LOCK_TOKEN	(*(u32 *)(&get_paca()->lock_token))
353 #else
354 #define LOCK_TOKEN	(*(u32 *)(&get_paca()->paca_index))
355 #endif
356 
357 static inline int try_lock_tlbie(unsigned int *lock)
358 {
359 	unsigned int tmp, old;
360 	unsigned int token = LOCK_TOKEN;
361 
362 	asm volatile("1:lwarx	%1,0,%2\n"
363 		     "	cmpwi	cr0,%1,0\n"
364 		     "	bne	2f\n"
365 		     "  stwcx.	%3,0,%2\n"
366 		     "	bne-	1b\n"
367 		     "  isync\n"
368 		     "2:"
369 		     : "=&r" (tmp), "=&r" (old)
370 		     : "r" (lock), "r" (token)
371 		     : "cc", "memory");
372 	return old == 0;
373 }
374 
375 static void do_tlbies(struct kvm *kvm, unsigned long *rbvalues,
376 		      long npages, int global, bool need_sync)
377 {
378 	long i;
379 
380 	if (global) {
381 		while (!try_lock_tlbie(&kvm->arch.tlbie_lock))
382 			cpu_relax();
383 		if (need_sync)
384 			asm volatile("ptesync" : : : "memory");
385 		for (i = 0; i < npages; ++i)
386 			asm volatile(PPC_TLBIE(%1,%0) : :
387 				     "r" (rbvalues[i]), "r" (kvm->arch.lpid));
388 		asm volatile("eieio; tlbsync; ptesync" : : : "memory");
389 		kvm->arch.tlbie_lock = 0;
390 	} else {
391 		if (need_sync)
392 			asm volatile("ptesync" : : : "memory");
393 		for (i = 0; i < npages; ++i)
394 			asm volatile("tlbiel %0" : : "r" (rbvalues[i]));
395 		asm volatile("ptesync" : : : "memory");
396 	}
397 }
398 
399 long kvmppc_do_h_remove(struct kvm *kvm, unsigned long flags,
400 			unsigned long pte_index, unsigned long avpn,
401 			unsigned long *hpret)
402 {
403 	__be64 *hpte;
404 	unsigned long v, r, rb;
405 	struct revmap_entry *rev;
406 	u64 pte;
407 
408 	if (pte_index >= kvm->arch.hpt_npte)
409 		return H_PARAMETER;
410 	hpte = (__be64 *)(kvm->arch.hpt_virt + (pte_index << 4));
411 	while (!try_lock_hpte(hpte, HPTE_V_HVLOCK))
412 		cpu_relax();
413 	pte = be64_to_cpu(hpte[0]);
414 	if ((pte & (HPTE_V_ABSENT | HPTE_V_VALID)) == 0 ||
415 	    ((flags & H_AVPN) && (pte & ~0x7fUL) != avpn) ||
416 	    ((flags & H_ANDCOND) && (pte & avpn) != 0)) {
417 		__unlock_hpte(hpte, pte);
418 		return H_NOT_FOUND;
419 	}
420 
421 	rev = real_vmalloc_addr(&kvm->arch.revmap[pte_index]);
422 	v = pte & ~HPTE_V_HVLOCK;
423 	if (v & HPTE_V_VALID) {
424 		u64 pte1;
425 
426 		pte1 = be64_to_cpu(hpte[1]);
427 		hpte[0] &= ~cpu_to_be64(HPTE_V_VALID);
428 		rb = compute_tlbie_rb(v, pte1, pte_index);
429 		do_tlbies(kvm, &rb, 1, global_invalidates(kvm, flags), true);
430 		/* Read PTE low word after tlbie to get final R/C values */
431 		remove_revmap_chain(kvm, pte_index, rev, v, pte1);
432 	}
433 	r = rev->guest_rpte & ~HPTE_GR_RESERVED;
434 	note_hpte_modification(kvm, rev);
435 	unlock_hpte(hpte, 0);
436 
437 	hpret[0] = v;
438 	hpret[1] = r;
439 	return H_SUCCESS;
440 }
441 EXPORT_SYMBOL_GPL(kvmppc_do_h_remove);
442 
443 long kvmppc_h_remove(struct kvm_vcpu *vcpu, unsigned long flags,
444 		     unsigned long pte_index, unsigned long avpn)
445 {
446 	return kvmppc_do_h_remove(vcpu->kvm, flags, pte_index, avpn,
447 				  &vcpu->arch.gpr[4]);
448 }
449 
450 long kvmppc_h_bulk_remove(struct kvm_vcpu *vcpu)
451 {
452 	struct kvm *kvm = vcpu->kvm;
453 	unsigned long *args = &vcpu->arch.gpr[4];
454 	__be64 *hp, *hptes[4];
455 	unsigned long tlbrb[4];
456 	long int i, j, k, n, found, indexes[4];
457 	unsigned long flags, req, pte_index, rcbits;
458 	int global;
459 	long int ret = H_SUCCESS;
460 	struct revmap_entry *rev, *revs[4];
461 	u64 hp0;
462 
463 	global = global_invalidates(kvm, 0);
464 	for (i = 0; i < 4 && ret == H_SUCCESS; ) {
465 		n = 0;
466 		for (; i < 4; ++i) {
467 			j = i * 2;
468 			pte_index = args[j];
469 			flags = pte_index >> 56;
470 			pte_index &= ((1ul << 56) - 1);
471 			req = flags >> 6;
472 			flags &= 3;
473 			if (req == 3) {		/* no more requests */
474 				i = 4;
475 				break;
476 			}
477 			if (req != 1 || flags == 3 ||
478 			    pte_index >= kvm->arch.hpt_npte) {
479 				/* parameter error */
480 				args[j] = ((0xa0 | flags) << 56) + pte_index;
481 				ret = H_PARAMETER;
482 				break;
483 			}
484 			hp = (__be64 *) (kvm->arch.hpt_virt + (pte_index << 4));
485 			/* to avoid deadlock, don't spin except for first */
486 			if (!try_lock_hpte(hp, HPTE_V_HVLOCK)) {
487 				if (n)
488 					break;
489 				while (!try_lock_hpte(hp, HPTE_V_HVLOCK))
490 					cpu_relax();
491 			}
492 			found = 0;
493 			hp0 = be64_to_cpu(hp[0]);
494 			if (hp0 & (HPTE_V_ABSENT | HPTE_V_VALID)) {
495 				switch (flags & 3) {
496 				case 0:		/* absolute */
497 					found = 1;
498 					break;
499 				case 1:		/* andcond */
500 					if (!(hp0 & args[j + 1]))
501 						found = 1;
502 					break;
503 				case 2:		/* AVPN */
504 					if ((hp0 & ~0x7fUL) == args[j + 1])
505 						found = 1;
506 					break;
507 				}
508 			}
509 			if (!found) {
510 				hp[0] &= ~cpu_to_be64(HPTE_V_HVLOCK);
511 				args[j] = ((0x90 | flags) << 56) + pte_index;
512 				continue;
513 			}
514 
515 			args[j] = ((0x80 | flags) << 56) + pte_index;
516 			rev = real_vmalloc_addr(&kvm->arch.revmap[pte_index]);
517 			note_hpte_modification(kvm, rev);
518 
519 			if (!(hp0 & HPTE_V_VALID)) {
520 				/* insert R and C bits from PTE */
521 				rcbits = rev->guest_rpte & (HPTE_R_R|HPTE_R_C);
522 				args[j] |= rcbits << (56 - 5);
523 				hp[0] = 0;
524 				continue;
525 			}
526 
527 			/* leave it locked */
528 			hp[0] &= ~cpu_to_be64(HPTE_V_VALID);
529 			tlbrb[n] = compute_tlbie_rb(be64_to_cpu(hp[0]),
530 				be64_to_cpu(hp[1]), pte_index);
531 			indexes[n] = j;
532 			hptes[n] = hp;
533 			revs[n] = rev;
534 			++n;
535 		}
536 
537 		if (!n)
538 			break;
539 
540 		/* Now that we've collected a batch, do the tlbies */
541 		do_tlbies(kvm, tlbrb, n, global, true);
542 
543 		/* Read PTE low words after tlbie to get final R/C values */
544 		for (k = 0; k < n; ++k) {
545 			j = indexes[k];
546 			pte_index = args[j] & ((1ul << 56) - 1);
547 			hp = hptes[k];
548 			rev = revs[k];
549 			remove_revmap_chain(kvm, pte_index, rev,
550 				be64_to_cpu(hp[0]), be64_to_cpu(hp[1]));
551 			rcbits = rev->guest_rpte & (HPTE_R_R|HPTE_R_C);
552 			args[j] |= rcbits << (56 - 5);
553 			__unlock_hpte(hp, 0);
554 		}
555 	}
556 
557 	return ret;
558 }
559 
560 long kvmppc_h_protect(struct kvm_vcpu *vcpu, unsigned long flags,
561 		      unsigned long pte_index, unsigned long avpn,
562 		      unsigned long va)
563 {
564 	struct kvm *kvm = vcpu->kvm;
565 	__be64 *hpte;
566 	struct revmap_entry *rev;
567 	unsigned long v, r, rb, mask, bits;
568 	u64 pte;
569 
570 	if (pte_index >= kvm->arch.hpt_npte)
571 		return H_PARAMETER;
572 
573 	hpte = (__be64 *)(kvm->arch.hpt_virt + (pte_index << 4));
574 	while (!try_lock_hpte(hpte, HPTE_V_HVLOCK))
575 		cpu_relax();
576 	pte = be64_to_cpu(hpte[0]);
577 	if ((pte & (HPTE_V_ABSENT | HPTE_V_VALID)) == 0 ||
578 	    ((flags & H_AVPN) && (pte & ~0x7fUL) != avpn)) {
579 		__unlock_hpte(hpte, pte);
580 		return H_NOT_FOUND;
581 	}
582 
583 	v = pte;
584 	bits = (flags << 55) & HPTE_R_PP0;
585 	bits |= (flags << 48) & HPTE_R_KEY_HI;
586 	bits |= flags & (HPTE_R_PP | HPTE_R_N | HPTE_R_KEY_LO);
587 
588 	/* Update guest view of 2nd HPTE dword */
589 	mask = HPTE_R_PP0 | HPTE_R_PP | HPTE_R_N |
590 		HPTE_R_KEY_HI | HPTE_R_KEY_LO;
591 	rev = real_vmalloc_addr(&kvm->arch.revmap[pte_index]);
592 	if (rev) {
593 		r = (rev->guest_rpte & ~mask) | bits;
594 		rev->guest_rpte = r;
595 		note_hpte_modification(kvm, rev);
596 	}
597 
598 	/* Update HPTE */
599 	if (v & HPTE_V_VALID) {
600 		/*
601 		 * If the page is valid, don't let it transition from
602 		 * readonly to writable.  If it should be writable, we'll
603 		 * take a trap and let the page fault code sort it out.
604 		 */
605 		pte = be64_to_cpu(hpte[1]);
606 		r = (pte & ~mask) | bits;
607 		if (hpte_is_writable(r) && !hpte_is_writable(pte))
608 			r = hpte_make_readonly(r);
609 		/* If the PTE is changing, invalidate it first */
610 		if (r != pte) {
611 			rb = compute_tlbie_rb(v, r, pte_index);
612 			hpte[0] = cpu_to_be64((v & ~HPTE_V_VALID) |
613 					      HPTE_V_ABSENT);
614 			do_tlbies(kvm, &rb, 1, global_invalidates(kvm, flags),
615 				  true);
616 			hpte[1] = cpu_to_be64(r);
617 		}
618 	}
619 	unlock_hpte(hpte, v & ~HPTE_V_HVLOCK);
620 	asm volatile("ptesync" : : : "memory");
621 	return H_SUCCESS;
622 }
623 
624 long kvmppc_h_read(struct kvm_vcpu *vcpu, unsigned long flags,
625 		   unsigned long pte_index)
626 {
627 	struct kvm *kvm = vcpu->kvm;
628 	__be64 *hpte;
629 	unsigned long v, r;
630 	int i, n = 1;
631 	struct revmap_entry *rev = NULL;
632 
633 	if (pte_index >= kvm->arch.hpt_npte)
634 		return H_PARAMETER;
635 	if (flags & H_READ_4) {
636 		pte_index &= ~3;
637 		n = 4;
638 	}
639 	rev = real_vmalloc_addr(&kvm->arch.revmap[pte_index]);
640 	for (i = 0; i < n; ++i, ++pte_index) {
641 		hpte = (__be64 *)(kvm->arch.hpt_virt + (pte_index << 4));
642 		v = be64_to_cpu(hpte[0]) & ~HPTE_V_HVLOCK;
643 		r = be64_to_cpu(hpte[1]);
644 		if (v & HPTE_V_ABSENT) {
645 			v &= ~HPTE_V_ABSENT;
646 			v |= HPTE_V_VALID;
647 		}
648 		if (v & HPTE_V_VALID) {
649 			r = rev[i].guest_rpte | (r & (HPTE_R_R | HPTE_R_C));
650 			r &= ~HPTE_GR_RESERVED;
651 		}
652 		vcpu->arch.gpr[4 + i * 2] = v;
653 		vcpu->arch.gpr[5 + i * 2] = r;
654 	}
655 	return H_SUCCESS;
656 }
657 
658 void kvmppc_invalidate_hpte(struct kvm *kvm, __be64 *hptep,
659 			unsigned long pte_index)
660 {
661 	unsigned long rb;
662 
663 	hptep[0] &= ~cpu_to_be64(HPTE_V_VALID);
664 	rb = compute_tlbie_rb(be64_to_cpu(hptep[0]), be64_to_cpu(hptep[1]),
665 			      pte_index);
666 	do_tlbies(kvm, &rb, 1, 1, true);
667 }
668 EXPORT_SYMBOL_GPL(kvmppc_invalidate_hpte);
669 
670 void kvmppc_clear_ref_hpte(struct kvm *kvm, __be64 *hptep,
671 			   unsigned long pte_index)
672 {
673 	unsigned long rb;
674 	unsigned char rbyte;
675 
676 	rb = compute_tlbie_rb(be64_to_cpu(hptep[0]), be64_to_cpu(hptep[1]),
677 			      pte_index);
678 	rbyte = (be64_to_cpu(hptep[1]) & ~HPTE_R_R) >> 8;
679 	/* modify only the second-last byte, which contains the ref bit */
680 	*((char *)hptep + 14) = rbyte;
681 	do_tlbies(kvm, &rb, 1, 1, false);
682 }
683 EXPORT_SYMBOL_GPL(kvmppc_clear_ref_hpte);
684 
685 static int slb_base_page_shift[4] = {
686 	24,	/* 16M */
687 	16,	/* 64k */
688 	34,	/* 16G */
689 	20,	/* 1M, unsupported */
690 };
691 
692 /* When called from virtmode, this func should be protected by
693  * preempt_disable(), otherwise, the holding of HPTE_V_HVLOCK
694  * can trigger deadlock issue.
695  */
696 long kvmppc_hv_find_lock_hpte(struct kvm *kvm, gva_t eaddr, unsigned long slb_v,
697 			      unsigned long valid)
698 {
699 	unsigned int i;
700 	unsigned int pshift;
701 	unsigned long somask;
702 	unsigned long vsid, hash;
703 	unsigned long avpn;
704 	__be64 *hpte;
705 	unsigned long mask, val;
706 	unsigned long v, r;
707 
708 	/* Get page shift, work out hash and AVPN etc. */
709 	mask = SLB_VSID_B | HPTE_V_AVPN | HPTE_V_SECONDARY;
710 	val = 0;
711 	pshift = 12;
712 	if (slb_v & SLB_VSID_L) {
713 		mask |= HPTE_V_LARGE;
714 		val |= HPTE_V_LARGE;
715 		pshift = slb_base_page_shift[(slb_v & SLB_VSID_LP) >> 4];
716 	}
717 	if (slb_v & SLB_VSID_B_1T) {
718 		somask = (1UL << 40) - 1;
719 		vsid = (slb_v & ~SLB_VSID_B) >> SLB_VSID_SHIFT_1T;
720 		vsid ^= vsid << 25;
721 	} else {
722 		somask = (1UL << 28) - 1;
723 		vsid = (slb_v & ~SLB_VSID_B) >> SLB_VSID_SHIFT;
724 	}
725 	hash = (vsid ^ ((eaddr & somask) >> pshift)) & kvm->arch.hpt_mask;
726 	avpn = slb_v & ~(somask >> 16);	/* also includes B */
727 	avpn |= (eaddr & somask) >> 16;
728 
729 	if (pshift >= 24)
730 		avpn &= ~((1UL << (pshift - 16)) - 1);
731 	else
732 		avpn &= ~0x7fUL;
733 	val |= avpn;
734 
735 	for (;;) {
736 		hpte = (__be64 *)(kvm->arch.hpt_virt + (hash << 7));
737 
738 		for (i = 0; i < 16; i += 2) {
739 			/* Read the PTE racily */
740 			v = be64_to_cpu(hpte[i]) & ~HPTE_V_HVLOCK;
741 
742 			/* Check valid/absent, hash, segment size and AVPN */
743 			if (!(v & valid) || (v & mask) != val)
744 				continue;
745 
746 			/* Lock the PTE and read it under the lock */
747 			while (!try_lock_hpte(&hpte[i], HPTE_V_HVLOCK))
748 				cpu_relax();
749 			v = be64_to_cpu(hpte[i]) & ~HPTE_V_HVLOCK;
750 			r = be64_to_cpu(hpte[i+1]);
751 
752 			/*
753 			 * Check the HPTE again, including base page size
754 			 */
755 			if ((v & valid) && (v & mask) == val &&
756 			    hpte_base_page_size(v, r) == (1ul << pshift))
757 				/* Return with the HPTE still locked */
758 				return (hash << 3) + (i >> 1);
759 
760 			__unlock_hpte(&hpte[i], v);
761 		}
762 
763 		if (val & HPTE_V_SECONDARY)
764 			break;
765 		val |= HPTE_V_SECONDARY;
766 		hash = hash ^ kvm->arch.hpt_mask;
767 	}
768 	return -1;
769 }
770 EXPORT_SYMBOL(kvmppc_hv_find_lock_hpte);
771 
772 /*
773  * Called in real mode to check whether an HPTE not found fault
774  * is due to accessing a paged-out page or an emulated MMIO page,
775  * or if a protection fault is due to accessing a page that the
776  * guest wanted read/write access to but which we made read-only.
777  * Returns a possibly modified status (DSISR) value if not
778  * (i.e. pass the interrupt to the guest),
779  * -1 to pass the fault up to host kernel mode code, -2 to do that
780  * and also load the instruction word (for MMIO emulation),
781  * or 0 if we should make the guest retry the access.
782  */
783 long kvmppc_hpte_hv_fault(struct kvm_vcpu *vcpu, unsigned long addr,
784 			  unsigned long slb_v, unsigned int status, bool data)
785 {
786 	struct kvm *kvm = vcpu->kvm;
787 	long int index;
788 	unsigned long v, r, gr;
789 	__be64 *hpte;
790 	unsigned long valid;
791 	struct revmap_entry *rev;
792 	unsigned long pp, key;
793 
794 	/* For protection fault, expect to find a valid HPTE */
795 	valid = HPTE_V_VALID;
796 	if (status & DSISR_NOHPTE)
797 		valid |= HPTE_V_ABSENT;
798 
799 	index = kvmppc_hv_find_lock_hpte(kvm, addr, slb_v, valid);
800 	if (index < 0) {
801 		if (status & DSISR_NOHPTE)
802 			return status;	/* there really was no HPTE */
803 		return 0;		/* for prot fault, HPTE disappeared */
804 	}
805 	hpte = (__be64 *)(kvm->arch.hpt_virt + (index << 4));
806 	v = be64_to_cpu(hpte[0]) & ~HPTE_V_HVLOCK;
807 	r = be64_to_cpu(hpte[1]);
808 	rev = real_vmalloc_addr(&kvm->arch.revmap[index]);
809 	gr = rev->guest_rpte;
810 
811 	unlock_hpte(hpte, v);
812 
813 	/* For not found, if the HPTE is valid by now, retry the instruction */
814 	if ((status & DSISR_NOHPTE) && (v & HPTE_V_VALID))
815 		return 0;
816 
817 	/* Check access permissions to the page */
818 	pp = gr & (HPTE_R_PP0 | HPTE_R_PP);
819 	key = (vcpu->arch.shregs.msr & MSR_PR) ? SLB_VSID_KP : SLB_VSID_KS;
820 	status &= ~DSISR_NOHPTE;	/* DSISR_NOHPTE == SRR1_ISI_NOPT */
821 	if (!data) {
822 		if (gr & (HPTE_R_N | HPTE_R_G))
823 			return status | SRR1_ISI_N_OR_G;
824 		if (!hpte_read_permission(pp, slb_v & key))
825 			return status | SRR1_ISI_PROT;
826 	} else if (status & DSISR_ISSTORE) {
827 		/* check write permission */
828 		if (!hpte_write_permission(pp, slb_v & key))
829 			return status | DSISR_PROTFAULT;
830 	} else {
831 		if (!hpte_read_permission(pp, slb_v & key))
832 			return status | DSISR_PROTFAULT;
833 	}
834 
835 	/* Check storage key, if applicable */
836 	if (data && (vcpu->arch.shregs.msr & MSR_DR)) {
837 		unsigned int perm = hpte_get_skey_perm(gr, vcpu->arch.amr);
838 		if (status & DSISR_ISSTORE)
839 			perm >>= 1;
840 		if (perm & 1)
841 			return status | DSISR_KEYFAULT;
842 	}
843 
844 	/* Save HPTE info for virtual-mode handler */
845 	vcpu->arch.pgfault_addr = addr;
846 	vcpu->arch.pgfault_index = index;
847 	vcpu->arch.pgfault_hpte[0] = v;
848 	vcpu->arch.pgfault_hpte[1] = r;
849 
850 	/* Check the storage key to see if it is possibly emulated MMIO */
851 	if (data && (vcpu->arch.shregs.msr & MSR_IR) &&
852 	    (r & (HPTE_R_KEY_HI | HPTE_R_KEY_LO)) ==
853 	    (HPTE_R_KEY_HI | HPTE_R_KEY_LO))
854 		return -2;	/* MMIO emulation - load instr word */
855 
856 	return -1;		/* send fault up to host kernel mode */
857 }
858