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