1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  *
4  * Copyright 2010-2011 Paul Mackerras, IBM Corp. <paulus@au1.ibm.com>
5  */
6 
7 #include <linux/types.h>
8 #include <linux/string.h>
9 #include <linux/kvm.h>
10 #include <linux/kvm_host.h>
11 #include <linux/hugetlb.h>
12 #include <linux/module.h>
13 #include <linux/log2.h>
14 #include <linux/sizes.h>
15 
16 #include <asm/trace.h>
17 #include <asm/kvm_ppc.h>
18 #include <asm/kvm_book3s.h>
19 #include <asm/book3s/64/mmu-hash.h>
20 #include <asm/hvcall.h>
21 #include <asm/synch.h>
22 #include <asm/ppc-opcode.h>
23 #include <asm/pte-walk.h>
24 
25 /* Translate address of a vmalloc'd thing to a linear map address */
26 static void *real_vmalloc_addr(void *addr)
27 {
28 	return __va(ppc_find_vmap_phys((unsigned long)addr));
29 }
30 
31 /* Return 1 if we need to do a global tlbie, 0 if we can use tlbiel */
32 static int global_invalidates(struct kvm *kvm)
33 {
34 	int global;
35 	int cpu;
36 
37 	/*
38 	 * If there is only one vcore, and it's currently running,
39 	 * as indicated by local_paca->kvm_hstate.kvm_vcpu being set,
40 	 * we can use tlbiel as long as we mark all other physical
41 	 * cores as potentially having stale TLB entries for this lpid.
42 	 * Otherwise, don't use tlbiel.
43 	 */
44 	if (kvm->arch.online_vcores == 1 && local_paca->kvm_hstate.kvm_vcpu)
45 		global = 0;
46 	else
47 		global = 1;
48 
49 	/* LPID has been switched to host if in virt mode so can't do local */
50 	if (!global && (mfmsr() & (MSR_IR|MSR_DR)))
51 		global = 1;
52 
53 	if (!global) {
54 		/* any other core might now have stale TLB entries... */
55 		smp_wmb();
56 		cpumask_setall(&kvm->arch.need_tlb_flush);
57 		cpu = local_paca->kvm_hstate.kvm_vcore->pcpu;
58 		cpumask_clear_cpu(cpu, &kvm->arch.need_tlb_flush);
59 	}
60 
61 	return global;
62 }
63 
64 /*
65  * Add this HPTE into the chain for the real page.
66  * Must be called with the chain locked; it unlocks the chain.
67  */
68 void kvmppc_add_revmap_chain(struct kvm *kvm, struct revmap_entry *rev,
69 			     unsigned long *rmap, long pte_index, int realmode)
70 {
71 	struct revmap_entry *head, *tail;
72 	unsigned long i;
73 
74 	if (*rmap & KVMPPC_RMAP_PRESENT) {
75 		i = *rmap & KVMPPC_RMAP_INDEX;
76 		head = &kvm->arch.hpt.rev[i];
77 		if (realmode)
78 			head = real_vmalloc_addr(head);
79 		tail = &kvm->arch.hpt.rev[head->back];
80 		if (realmode)
81 			tail = real_vmalloc_addr(tail);
82 		rev->forw = i;
83 		rev->back = head->back;
84 		tail->forw = pte_index;
85 		head->back = pte_index;
86 	} else {
87 		rev->forw = rev->back = pte_index;
88 		*rmap = (*rmap & ~KVMPPC_RMAP_INDEX) |
89 			pte_index | KVMPPC_RMAP_PRESENT | KVMPPC_RMAP_HPT;
90 	}
91 	unlock_rmap(rmap);
92 }
93 EXPORT_SYMBOL_GPL(kvmppc_add_revmap_chain);
94 
95 /* Update the dirty bitmap of a memslot */
96 void kvmppc_update_dirty_map(const struct kvm_memory_slot *memslot,
97 			     unsigned long gfn, unsigned long psize)
98 {
99 	unsigned long npages;
100 
101 	if (!psize || !memslot->dirty_bitmap)
102 		return;
103 	npages = (psize + PAGE_SIZE - 1) / PAGE_SIZE;
104 	gfn -= memslot->base_gfn;
105 	set_dirty_bits_atomic(memslot->dirty_bitmap, gfn, npages);
106 }
107 EXPORT_SYMBOL_GPL(kvmppc_update_dirty_map);
108 
109 static void kvmppc_set_dirty_from_hpte(struct kvm *kvm,
110 				unsigned long hpte_v, unsigned long hpte_gr)
111 {
112 	struct kvm_memory_slot *memslot;
113 	unsigned long gfn;
114 	unsigned long psize;
115 
116 	psize = kvmppc_actual_pgsz(hpte_v, hpte_gr);
117 	gfn = hpte_rpn(hpte_gr, psize);
118 	memslot = __gfn_to_memslot(kvm_memslots_raw(kvm), gfn);
119 	if (memslot && memslot->dirty_bitmap)
120 		kvmppc_update_dirty_map(memslot, gfn, psize);
121 }
122 
123 /* Returns a pointer to the revmap entry for the page mapped by a HPTE */
124 static unsigned long *revmap_for_hpte(struct kvm *kvm, unsigned long hpte_v,
125 				      unsigned long hpte_gr,
126 				      struct kvm_memory_slot **memslotp,
127 				      unsigned long *gfnp)
128 {
129 	struct kvm_memory_slot *memslot;
130 	unsigned long *rmap;
131 	unsigned long gfn;
132 
133 	gfn = hpte_rpn(hpte_gr, kvmppc_actual_pgsz(hpte_v, hpte_gr));
134 	memslot = __gfn_to_memslot(kvm_memslots_raw(kvm), gfn);
135 	if (memslotp)
136 		*memslotp = memslot;
137 	if (gfnp)
138 		*gfnp = gfn;
139 	if (!memslot)
140 		return NULL;
141 
142 	rmap = real_vmalloc_addr(&memslot->arch.rmap[gfn - memslot->base_gfn]);
143 	return rmap;
144 }
145 
146 /* Remove this HPTE from the chain for a real page */
147 static void remove_revmap_chain(struct kvm *kvm, long pte_index,
148 				struct revmap_entry *rev,
149 				unsigned long hpte_v, unsigned long hpte_r)
150 {
151 	struct revmap_entry *next, *prev;
152 	unsigned long ptel, head;
153 	unsigned long *rmap;
154 	unsigned long rcbits;
155 	struct kvm_memory_slot *memslot;
156 	unsigned long gfn;
157 
158 	rcbits = hpte_r & (HPTE_R_R | HPTE_R_C);
159 	ptel = rev->guest_rpte |= rcbits;
160 	rmap = revmap_for_hpte(kvm, hpte_v, ptel, &memslot, &gfn);
161 	if (!rmap)
162 		return;
163 	lock_rmap(rmap);
164 
165 	head = *rmap & KVMPPC_RMAP_INDEX;
166 	next = real_vmalloc_addr(&kvm->arch.hpt.rev[rev->forw]);
167 	prev = real_vmalloc_addr(&kvm->arch.hpt.rev[rev->back]);
168 	next->back = rev->back;
169 	prev->forw = rev->forw;
170 	if (head == pte_index) {
171 		head = rev->forw;
172 		if (head == pte_index)
173 			*rmap &= ~(KVMPPC_RMAP_PRESENT | KVMPPC_RMAP_INDEX);
174 		else
175 			*rmap = (*rmap & ~KVMPPC_RMAP_INDEX) | head;
176 	}
177 	*rmap |= rcbits << KVMPPC_RMAP_RC_SHIFT;
178 	if (rcbits & HPTE_R_C)
179 		kvmppc_update_dirty_map(memslot, gfn,
180 					kvmppc_actual_pgsz(hpte_v, hpte_r));
181 	unlock_rmap(rmap);
182 }
183 
184 long kvmppc_do_h_enter(struct kvm *kvm, unsigned long flags,
185 		       long pte_index, unsigned long pteh, unsigned long ptel,
186 		       pgd_t *pgdir, bool realmode, unsigned long *pte_idx_ret)
187 {
188 	unsigned long i, pa, gpa, gfn, psize;
189 	unsigned long slot_fn, hva;
190 	__be64 *hpte;
191 	struct revmap_entry *rev;
192 	unsigned long g_ptel;
193 	struct kvm_memory_slot *memslot;
194 	unsigned hpage_shift;
195 	bool is_ci;
196 	unsigned long *rmap;
197 	pte_t *ptep;
198 	unsigned int writing;
199 	unsigned long mmu_seq;
200 	unsigned long rcbits;
201 
202 	if (kvm_is_radix(kvm))
203 		return H_FUNCTION;
204 	/*
205 	 * The HPTE gets used by compute_tlbie_rb() to set TLBIE bits, so
206 	 * these functions should work together -- must ensure a guest can not
207 	 * cause problems with the TLBIE that KVM executes.
208 	 */
209 	if ((pteh >> HPTE_V_SSIZE_SHIFT) & 0x2) {
210 		/* B=0b1x is a reserved value, disallow it. */
211 		return H_PARAMETER;
212 	}
213 	psize = kvmppc_actual_pgsz(pteh, ptel);
214 	if (!psize)
215 		return H_PARAMETER;
216 	writing = hpte_is_writable(ptel);
217 	pteh &= ~(HPTE_V_HVLOCK | HPTE_V_ABSENT | HPTE_V_VALID);
218 	ptel &= ~HPTE_GR_RESERVED;
219 	g_ptel = ptel;
220 
221 	/* used later to detect if we might have been invalidated */
222 	mmu_seq = kvm->mmu_invalidate_seq;
223 	smp_rmb();
224 
225 	/* Find the memslot (if any) for this address */
226 	gpa = (ptel & HPTE_R_RPN) & ~(psize - 1);
227 	gfn = gpa >> PAGE_SHIFT;
228 	memslot = __gfn_to_memslot(kvm_memslots_raw(kvm), gfn);
229 	pa = 0;
230 	is_ci = false;
231 	rmap = NULL;
232 	if (!(memslot && !(memslot->flags & KVM_MEMSLOT_INVALID))) {
233 		/* Emulated MMIO - mark this with key=31 */
234 		pteh |= HPTE_V_ABSENT;
235 		ptel |= HPTE_R_KEY_HI | HPTE_R_KEY_LO;
236 		goto do_insert;
237 	}
238 
239 	/* Check if the requested page fits entirely in the memslot. */
240 	if (!slot_is_aligned(memslot, psize))
241 		return H_PARAMETER;
242 	slot_fn = gfn - memslot->base_gfn;
243 	rmap = &memslot->arch.rmap[slot_fn];
244 
245 	/* Translate to host virtual address */
246 	hva = __gfn_to_hva_memslot(memslot, gfn);
247 
248 	arch_spin_lock(&kvm->mmu_lock.rlock.raw_lock);
249 	ptep = find_kvm_host_pte(kvm, mmu_seq, hva, &hpage_shift);
250 	if (ptep) {
251 		pte_t pte;
252 		unsigned int host_pte_size;
253 
254 		if (hpage_shift)
255 			host_pte_size = 1ul << hpage_shift;
256 		else
257 			host_pte_size = PAGE_SIZE;
258 		/*
259 		 * We should always find the guest page size
260 		 * to <= host page size, if host is using hugepage
261 		 */
262 		if (host_pte_size < psize) {
263 			arch_spin_unlock(&kvm->mmu_lock.rlock.raw_lock);
264 			return H_PARAMETER;
265 		}
266 		pte = kvmppc_read_update_linux_pte(ptep, writing);
267 		if (pte_present(pte) && !pte_protnone(pte)) {
268 			if (writing && !pte_write(pte))
269 				/* make the actual HPTE be read-only */
270 				ptel = hpte_make_readonly(ptel);
271 			is_ci = pte_ci(pte);
272 			pa = pte_pfn(pte) << PAGE_SHIFT;
273 			pa |= hva & (host_pte_size - 1);
274 			pa |= gpa & ~PAGE_MASK;
275 		}
276 	}
277 	arch_spin_unlock(&kvm->mmu_lock.rlock.raw_lock);
278 
279 	ptel &= HPTE_R_KEY | HPTE_R_PP0 | (psize-1);
280 	ptel |= pa;
281 
282 	if (pa)
283 		pteh |= HPTE_V_VALID;
284 	else {
285 		pteh |= HPTE_V_ABSENT;
286 		ptel &= ~(HPTE_R_KEY_HI | HPTE_R_KEY_LO);
287 	}
288 
289 	/*If we had host pte mapping then  Check WIMG */
290 	if (ptep && !hpte_cache_flags_ok(ptel, is_ci)) {
291 		if (is_ci)
292 			return H_PARAMETER;
293 		/*
294 		 * Allow guest to map emulated device memory as
295 		 * uncacheable, but actually make it cacheable.
296 		 */
297 		ptel &= ~(HPTE_R_W|HPTE_R_I|HPTE_R_G);
298 		ptel |= HPTE_R_M;
299 	}
300 
301 	/* Find and lock the HPTEG slot to use */
302  do_insert:
303 	if (pte_index >= kvmppc_hpt_npte(&kvm->arch.hpt))
304 		return H_PARAMETER;
305 	if (likely((flags & H_EXACT) == 0)) {
306 		pte_index &= ~7UL;
307 		hpte = (__be64 *)(kvm->arch.hpt.virt + (pte_index << 4));
308 		for (i = 0; i < 8; ++i) {
309 			if ((be64_to_cpu(*hpte) & HPTE_V_VALID) == 0 &&
310 			    try_lock_hpte(hpte, HPTE_V_HVLOCK | HPTE_V_VALID |
311 					  HPTE_V_ABSENT))
312 				break;
313 			hpte += 2;
314 		}
315 		if (i == 8) {
316 			/*
317 			 * Since try_lock_hpte doesn't retry (not even stdcx.
318 			 * failures), it could be that there is a free slot
319 			 * but we transiently failed to lock it.  Try again,
320 			 * actually locking each slot and checking it.
321 			 */
322 			hpte -= 16;
323 			for (i = 0; i < 8; ++i) {
324 				u64 pte;
325 				while (!try_lock_hpte(hpte, HPTE_V_HVLOCK))
326 					cpu_relax();
327 				pte = be64_to_cpu(hpte[0]);
328 				if (!(pte & (HPTE_V_VALID | HPTE_V_ABSENT)))
329 					break;
330 				__unlock_hpte(hpte, pte);
331 				hpte += 2;
332 			}
333 			if (i == 8)
334 				return H_PTEG_FULL;
335 		}
336 		pte_index += i;
337 	} else {
338 		hpte = (__be64 *)(kvm->arch.hpt.virt + (pte_index << 4));
339 		if (!try_lock_hpte(hpte, HPTE_V_HVLOCK | HPTE_V_VALID |
340 				   HPTE_V_ABSENT)) {
341 			/* Lock the slot and check again */
342 			u64 pte;
343 
344 			while (!try_lock_hpte(hpte, HPTE_V_HVLOCK))
345 				cpu_relax();
346 			pte = be64_to_cpu(hpte[0]);
347 			if (pte & (HPTE_V_VALID | HPTE_V_ABSENT)) {
348 				__unlock_hpte(hpte, pte);
349 				return H_PTEG_FULL;
350 			}
351 		}
352 	}
353 
354 	/* Save away the guest's idea of the second HPTE dword */
355 	rev = &kvm->arch.hpt.rev[pte_index];
356 	if (realmode)
357 		rev = real_vmalloc_addr(rev);
358 	if (rev) {
359 		rev->guest_rpte = g_ptel;
360 		note_hpte_modification(kvm, rev);
361 	}
362 
363 	/* Link HPTE into reverse-map chain */
364 	if (pteh & HPTE_V_VALID) {
365 		if (realmode)
366 			rmap = real_vmalloc_addr(rmap);
367 		lock_rmap(rmap);
368 		/* Check for pending invalidations under the rmap chain lock */
369 		if (mmu_invalidate_retry(kvm, mmu_seq)) {
370 			/* inval in progress, write a non-present HPTE */
371 			pteh |= HPTE_V_ABSENT;
372 			pteh &= ~HPTE_V_VALID;
373 			ptel &= ~(HPTE_R_KEY_HI | HPTE_R_KEY_LO);
374 			unlock_rmap(rmap);
375 		} else {
376 			kvmppc_add_revmap_chain(kvm, rev, rmap, pte_index,
377 						realmode);
378 			/* Only set R/C in real HPTE if already set in *rmap */
379 			rcbits = *rmap >> KVMPPC_RMAP_RC_SHIFT;
380 			ptel &= rcbits | ~(HPTE_R_R | HPTE_R_C);
381 		}
382 	}
383 
384 	/* Convert to new format on P9 */
385 	if (cpu_has_feature(CPU_FTR_ARCH_300)) {
386 		ptel = hpte_old_to_new_r(pteh, ptel);
387 		pteh = hpte_old_to_new_v(pteh);
388 	}
389 	hpte[1] = cpu_to_be64(ptel);
390 
391 	/* Write the first HPTE dword, unlocking the HPTE and making it valid */
392 	eieio();
393 	__unlock_hpte(hpte, pteh);
394 	asm volatile("ptesync" : : : "memory");
395 
396 	*pte_idx_ret = pte_index;
397 	return H_SUCCESS;
398 }
399 EXPORT_SYMBOL_GPL(kvmppc_do_h_enter);
400 
401 long kvmppc_h_enter(struct kvm_vcpu *vcpu, unsigned long flags,
402 		    long pte_index, unsigned long pteh, unsigned long ptel)
403 {
404 	return kvmppc_do_h_enter(vcpu->kvm, flags, pte_index, pteh, ptel,
405 				 vcpu->arch.pgdir, true,
406 				 &vcpu->arch.regs.gpr[4]);
407 }
408 EXPORT_SYMBOL_GPL(kvmppc_h_enter);
409 
410 #ifdef __BIG_ENDIAN__
411 #define LOCK_TOKEN	(*(u32 *)(&get_paca()->lock_token))
412 #else
413 #define LOCK_TOKEN	(*(u32 *)(&get_paca()->paca_index))
414 #endif
415 
416 static inline int is_mmio_hpte(unsigned long v, unsigned long r)
417 {
418 	return ((v & HPTE_V_ABSENT) &&
419 		(r & (HPTE_R_KEY_HI | HPTE_R_KEY_LO)) ==
420 		(HPTE_R_KEY_HI | HPTE_R_KEY_LO));
421 }
422 
423 static inline void fixup_tlbie_lpid(unsigned long rb_value, unsigned long lpid)
424 {
425 
426 	if (cpu_has_feature(CPU_FTR_P9_TLBIE_ERAT_BUG)) {
427 		/* Radix flush for a hash guest */
428 
429 		unsigned long rb,rs,prs,r,ric;
430 
431 		rb = PPC_BIT(52); /* IS = 2 */
432 		rs = 0;  /* lpid = 0 */
433 		prs = 0; /* partition scoped */
434 		r = 1;   /* radix format */
435 		ric = 0; /* RIC_FLSUH_TLB */
436 
437 		/*
438 		 * Need the extra ptesync to make sure we don't
439 		 * re-order the tlbie
440 		 */
441 		asm volatile("ptesync": : :"memory");
442 		asm volatile(PPC_TLBIE_5(%0, %4, %3, %2, %1)
443 			     : : "r"(rb), "i"(r), "i"(prs),
444 			       "i"(ric), "r"(rs) : "memory");
445 	}
446 
447 	if (cpu_has_feature(CPU_FTR_P9_TLBIE_STQ_BUG)) {
448 		asm volatile("ptesync": : :"memory");
449 		asm volatile(PPC_TLBIE_5(%0,%1,0,0,0) : :
450 			     "r" (rb_value), "r" (lpid));
451 	}
452 }
453 
454 static void do_tlbies(struct kvm *kvm, unsigned long *rbvalues,
455 		      long npages, int global, bool need_sync)
456 {
457 	long i;
458 
459 	/*
460 	 * We use the POWER9 5-operand versions of tlbie and tlbiel here.
461 	 * Since we are using RIC=0 PRS=0 R=0, and P7/P8 tlbiel ignores
462 	 * the RS field, this is backwards-compatible with P7 and P8.
463 	 */
464 	if (global) {
465 		if (need_sync)
466 			asm volatile("ptesync" : : : "memory");
467 		for (i = 0; i < npages; ++i) {
468 			asm volatile(PPC_TLBIE_5(%0,%1,0,0,0) : :
469 				     "r" (rbvalues[i]), "r" (kvm->arch.lpid));
470 		}
471 
472 		fixup_tlbie_lpid(rbvalues[i - 1], kvm->arch.lpid);
473 		asm volatile("eieio; tlbsync; ptesync" : : : "memory");
474 	} else {
475 		if (need_sync)
476 			asm volatile("ptesync" : : : "memory");
477 		for (i = 0; i < npages; ++i) {
478 			asm volatile(PPC_TLBIEL(%0,%1,0,0,0) : :
479 				     "r" (rbvalues[i]), "r" (0));
480 		}
481 		asm volatile("ptesync" : : : "memory");
482 	}
483 }
484 
485 long kvmppc_do_h_remove(struct kvm *kvm, unsigned long flags,
486 			unsigned long pte_index, unsigned long avpn,
487 			unsigned long *hpret)
488 {
489 	__be64 *hpte;
490 	unsigned long v, r, rb;
491 	struct revmap_entry *rev;
492 	u64 pte, orig_pte, pte_r;
493 
494 	if (kvm_is_radix(kvm))
495 		return H_FUNCTION;
496 	if (pte_index >= kvmppc_hpt_npte(&kvm->arch.hpt))
497 		return H_PARAMETER;
498 	hpte = (__be64 *)(kvm->arch.hpt.virt + (pte_index << 4));
499 	while (!try_lock_hpte(hpte, HPTE_V_HVLOCK))
500 		cpu_relax();
501 	pte = orig_pte = be64_to_cpu(hpte[0]);
502 	pte_r = be64_to_cpu(hpte[1]);
503 	if (cpu_has_feature(CPU_FTR_ARCH_300)) {
504 		pte = hpte_new_to_old_v(pte, pte_r);
505 		pte_r = hpte_new_to_old_r(pte_r);
506 	}
507 	if ((pte & (HPTE_V_ABSENT | HPTE_V_VALID)) == 0 ||
508 	    ((flags & H_AVPN) && (pte & ~0x7fUL) != avpn) ||
509 	    ((flags & H_ANDCOND) && (pte & avpn) != 0)) {
510 		__unlock_hpte(hpte, orig_pte);
511 		return H_NOT_FOUND;
512 	}
513 
514 	rev = real_vmalloc_addr(&kvm->arch.hpt.rev[pte_index]);
515 	v = pte & ~HPTE_V_HVLOCK;
516 	if (v & HPTE_V_VALID) {
517 		hpte[0] &= ~cpu_to_be64(HPTE_V_VALID);
518 		rb = compute_tlbie_rb(v, pte_r, pte_index);
519 		do_tlbies(kvm, &rb, 1, global_invalidates(kvm), true);
520 		/*
521 		 * The reference (R) and change (C) bits in a HPT
522 		 * entry can be set by hardware at any time up until
523 		 * the HPTE is invalidated and the TLB invalidation
524 		 * sequence has completed.  This means that when
525 		 * removing a HPTE, we need to re-read the HPTE after
526 		 * the invalidation sequence has completed in order to
527 		 * obtain reliable values of R and C.
528 		 */
529 		remove_revmap_chain(kvm, pte_index, rev, v,
530 				    be64_to_cpu(hpte[1]));
531 	}
532 	r = rev->guest_rpte & ~HPTE_GR_RESERVED;
533 	note_hpte_modification(kvm, rev);
534 	unlock_hpte(hpte, 0);
535 
536 	if (is_mmio_hpte(v, pte_r))
537 		atomic64_inc(&kvm->arch.mmio_update);
538 
539 	if (v & HPTE_V_ABSENT)
540 		v = (v & ~HPTE_V_ABSENT) | HPTE_V_VALID;
541 	hpret[0] = v;
542 	hpret[1] = r;
543 	return H_SUCCESS;
544 }
545 EXPORT_SYMBOL_GPL(kvmppc_do_h_remove);
546 
547 long kvmppc_h_remove(struct kvm_vcpu *vcpu, unsigned long flags,
548 		     unsigned long pte_index, unsigned long avpn)
549 {
550 	return kvmppc_do_h_remove(vcpu->kvm, flags, pte_index, avpn,
551 				  &vcpu->arch.regs.gpr[4]);
552 }
553 EXPORT_SYMBOL_GPL(kvmppc_h_remove);
554 
555 long kvmppc_h_bulk_remove(struct kvm_vcpu *vcpu)
556 {
557 	struct kvm *kvm = vcpu->kvm;
558 	unsigned long *args = &vcpu->arch.regs.gpr[4];
559 	__be64 *hp, *hptes[4];
560 	unsigned long tlbrb[4];
561 	long int i, j, k, n, found, indexes[4];
562 	unsigned long flags, req, pte_index, rcbits;
563 	int global;
564 	long int ret = H_SUCCESS;
565 	struct revmap_entry *rev, *revs[4];
566 	u64 hp0, hp1;
567 
568 	if (kvm_is_radix(kvm))
569 		return H_FUNCTION;
570 	global = global_invalidates(kvm);
571 	for (i = 0; i < 4 && ret == H_SUCCESS; ) {
572 		n = 0;
573 		for (; i < 4; ++i) {
574 			j = i * 2;
575 			pte_index = args[j];
576 			flags = pte_index >> 56;
577 			pte_index &= ((1ul << 56) - 1);
578 			req = flags >> 6;
579 			flags &= 3;
580 			if (req == 3) {		/* no more requests */
581 				i = 4;
582 				break;
583 			}
584 			if (req != 1 || flags == 3 ||
585 			    pte_index >= kvmppc_hpt_npte(&kvm->arch.hpt)) {
586 				/* parameter error */
587 				args[j] = ((0xa0 | flags) << 56) + pte_index;
588 				ret = H_PARAMETER;
589 				break;
590 			}
591 			hp = (__be64 *) (kvm->arch.hpt.virt + (pte_index << 4));
592 			/* to avoid deadlock, don't spin except for first */
593 			if (!try_lock_hpte(hp, HPTE_V_HVLOCK)) {
594 				if (n)
595 					break;
596 				while (!try_lock_hpte(hp, HPTE_V_HVLOCK))
597 					cpu_relax();
598 			}
599 			found = 0;
600 			hp0 = be64_to_cpu(hp[0]);
601 			hp1 = be64_to_cpu(hp[1]);
602 			if (cpu_has_feature(CPU_FTR_ARCH_300)) {
603 				hp0 = hpte_new_to_old_v(hp0, hp1);
604 				hp1 = hpte_new_to_old_r(hp1);
605 			}
606 			if (hp0 & (HPTE_V_ABSENT | HPTE_V_VALID)) {
607 				switch (flags & 3) {
608 				case 0:		/* absolute */
609 					found = 1;
610 					break;
611 				case 1:		/* andcond */
612 					if (!(hp0 & args[j + 1]))
613 						found = 1;
614 					break;
615 				case 2:		/* AVPN */
616 					if ((hp0 & ~0x7fUL) == args[j + 1])
617 						found = 1;
618 					break;
619 				}
620 			}
621 			if (!found) {
622 				hp[0] &= ~cpu_to_be64(HPTE_V_HVLOCK);
623 				args[j] = ((0x90 | flags) << 56) + pte_index;
624 				continue;
625 			}
626 
627 			args[j] = ((0x80 | flags) << 56) + pte_index;
628 			rev = real_vmalloc_addr(&kvm->arch.hpt.rev[pte_index]);
629 			note_hpte_modification(kvm, rev);
630 
631 			if (!(hp0 & HPTE_V_VALID)) {
632 				/* insert R and C bits from PTE */
633 				rcbits = rev->guest_rpte & (HPTE_R_R|HPTE_R_C);
634 				args[j] |= rcbits << (56 - 5);
635 				hp[0] = 0;
636 				if (is_mmio_hpte(hp0, hp1))
637 					atomic64_inc(&kvm->arch.mmio_update);
638 				continue;
639 			}
640 
641 			/* leave it locked */
642 			hp[0] &= ~cpu_to_be64(HPTE_V_VALID);
643 			tlbrb[n] = compute_tlbie_rb(hp0, hp1, pte_index);
644 			indexes[n] = j;
645 			hptes[n] = hp;
646 			revs[n] = rev;
647 			++n;
648 		}
649 
650 		if (!n)
651 			break;
652 
653 		/* Now that we've collected a batch, do the tlbies */
654 		do_tlbies(kvm, tlbrb, n, global, true);
655 
656 		/* Read PTE low words after tlbie to get final R/C values */
657 		for (k = 0; k < n; ++k) {
658 			j = indexes[k];
659 			pte_index = args[j] & ((1ul << 56) - 1);
660 			hp = hptes[k];
661 			rev = revs[k];
662 			remove_revmap_chain(kvm, pte_index, rev,
663 				be64_to_cpu(hp[0]), be64_to_cpu(hp[1]));
664 			rcbits = rev->guest_rpte & (HPTE_R_R|HPTE_R_C);
665 			args[j] |= rcbits << (56 - 5);
666 			__unlock_hpte(hp, 0);
667 		}
668 	}
669 
670 	return ret;
671 }
672 EXPORT_SYMBOL_GPL(kvmppc_h_bulk_remove);
673 
674 long kvmppc_h_protect(struct kvm_vcpu *vcpu, unsigned long flags,
675 		      unsigned long pte_index, unsigned long avpn)
676 {
677 	struct kvm *kvm = vcpu->kvm;
678 	__be64 *hpte;
679 	struct revmap_entry *rev;
680 	unsigned long v, r, rb, mask, bits;
681 	u64 pte_v, pte_r;
682 
683 	if (kvm_is_radix(kvm))
684 		return H_FUNCTION;
685 	if (pte_index >= kvmppc_hpt_npte(&kvm->arch.hpt))
686 		return H_PARAMETER;
687 
688 	hpte = (__be64 *)(kvm->arch.hpt.virt + (pte_index << 4));
689 	while (!try_lock_hpte(hpte, HPTE_V_HVLOCK))
690 		cpu_relax();
691 	v = pte_v = be64_to_cpu(hpte[0]);
692 	if (cpu_has_feature(CPU_FTR_ARCH_300))
693 		v = hpte_new_to_old_v(v, be64_to_cpu(hpte[1]));
694 	if ((v & (HPTE_V_ABSENT | HPTE_V_VALID)) == 0 ||
695 	    ((flags & H_AVPN) && (v & ~0x7fUL) != avpn)) {
696 		__unlock_hpte(hpte, pte_v);
697 		return H_NOT_FOUND;
698 	}
699 
700 	pte_r = be64_to_cpu(hpte[1]);
701 	bits = (flags << 55) & HPTE_R_PP0;
702 	bits |= (flags << 48) & HPTE_R_KEY_HI;
703 	bits |= flags & (HPTE_R_PP | HPTE_R_N | HPTE_R_KEY_LO);
704 
705 	/* Update guest view of 2nd HPTE dword */
706 	mask = HPTE_R_PP0 | HPTE_R_PP | HPTE_R_N |
707 		HPTE_R_KEY_HI | HPTE_R_KEY_LO;
708 	rev = real_vmalloc_addr(&kvm->arch.hpt.rev[pte_index]);
709 	if (rev) {
710 		r = (rev->guest_rpte & ~mask) | bits;
711 		rev->guest_rpte = r;
712 		note_hpte_modification(kvm, rev);
713 	}
714 
715 	/* Update HPTE */
716 	if (v & HPTE_V_VALID) {
717 		/*
718 		 * If the page is valid, don't let it transition from
719 		 * readonly to writable.  If it should be writable, we'll
720 		 * take a trap and let the page fault code sort it out.
721 		 */
722 		r = (pte_r & ~mask) | bits;
723 		if (hpte_is_writable(r) && !hpte_is_writable(pte_r))
724 			r = hpte_make_readonly(r);
725 		/* If the PTE is changing, invalidate it first */
726 		if (r != pte_r) {
727 			rb = compute_tlbie_rb(v, r, pte_index);
728 			hpte[0] = cpu_to_be64((pte_v & ~HPTE_V_VALID) |
729 					      HPTE_V_ABSENT);
730 			do_tlbies(kvm, &rb, 1, global_invalidates(kvm), true);
731 			/* Don't lose R/C bit updates done by hardware */
732 			r |= be64_to_cpu(hpte[1]) & (HPTE_R_R | HPTE_R_C);
733 			hpte[1] = cpu_to_be64(r);
734 		}
735 	}
736 	unlock_hpte(hpte, pte_v & ~HPTE_V_HVLOCK);
737 	asm volatile("ptesync" : : : "memory");
738 	if (is_mmio_hpte(v, pte_r))
739 		atomic64_inc(&kvm->arch.mmio_update);
740 
741 	return H_SUCCESS;
742 }
743 EXPORT_SYMBOL_GPL(kvmppc_h_protect);
744 
745 long kvmppc_h_read(struct kvm_vcpu *vcpu, unsigned long flags,
746 		   unsigned long pte_index)
747 {
748 	struct kvm *kvm = vcpu->kvm;
749 	__be64 *hpte;
750 	unsigned long v, r;
751 	int i, n = 1;
752 	struct revmap_entry *rev = NULL;
753 
754 	if (kvm_is_radix(kvm))
755 		return H_FUNCTION;
756 	if (pte_index >= kvmppc_hpt_npte(&kvm->arch.hpt))
757 		return H_PARAMETER;
758 	if (flags & H_READ_4) {
759 		pte_index &= ~3;
760 		n = 4;
761 	}
762 	rev = real_vmalloc_addr(&kvm->arch.hpt.rev[pte_index]);
763 	for (i = 0; i < n; ++i, ++pte_index) {
764 		hpte = (__be64 *)(kvm->arch.hpt.virt + (pte_index << 4));
765 		v = be64_to_cpu(hpte[0]) & ~HPTE_V_HVLOCK;
766 		r = be64_to_cpu(hpte[1]);
767 		if (cpu_has_feature(CPU_FTR_ARCH_300)) {
768 			v = hpte_new_to_old_v(v, r);
769 			r = hpte_new_to_old_r(r);
770 		}
771 		if (v & HPTE_V_ABSENT) {
772 			v &= ~HPTE_V_ABSENT;
773 			v |= HPTE_V_VALID;
774 		}
775 		if (v & HPTE_V_VALID) {
776 			r = rev[i].guest_rpte | (r & (HPTE_R_R | HPTE_R_C));
777 			r &= ~HPTE_GR_RESERVED;
778 		}
779 		vcpu->arch.regs.gpr[4 + i * 2] = v;
780 		vcpu->arch.regs.gpr[5 + i * 2] = r;
781 	}
782 	return H_SUCCESS;
783 }
784 EXPORT_SYMBOL_GPL(kvmppc_h_read);
785 
786 long kvmppc_h_clear_ref(struct kvm_vcpu *vcpu, unsigned long flags,
787 			unsigned long pte_index)
788 {
789 	struct kvm *kvm = vcpu->kvm;
790 	__be64 *hpte;
791 	unsigned long v, r, gr;
792 	struct revmap_entry *rev;
793 	unsigned long *rmap;
794 	long ret = H_NOT_FOUND;
795 
796 	if (kvm_is_radix(kvm))
797 		return H_FUNCTION;
798 	if (pte_index >= kvmppc_hpt_npte(&kvm->arch.hpt))
799 		return H_PARAMETER;
800 
801 	rev = real_vmalloc_addr(&kvm->arch.hpt.rev[pte_index]);
802 	hpte = (__be64 *)(kvm->arch.hpt.virt + (pte_index << 4));
803 	while (!try_lock_hpte(hpte, HPTE_V_HVLOCK))
804 		cpu_relax();
805 	v = be64_to_cpu(hpte[0]);
806 	r = be64_to_cpu(hpte[1]);
807 	if (!(v & (HPTE_V_VALID | HPTE_V_ABSENT)))
808 		goto out;
809 
810 	gr = rev->guest_rpte;
811 	if (rev->guest_rpte & HPTE_R_R) {
812 		rev->guest_rpte &= ~HPTE_R_R;
813 		note_hpte_modification(kvm, rev);
814 	}
815 	if (v & HPTE_V_VALID) {
816 		gr |= r & (HPTE_R_R | HPTE_R_C);
817 		if (r & HPTE_R_R) {
818 			kvmppc_clear_ref_hpte(kvm, hpte, pte_index);
819 			rmap = revmap_for_hpte(kvm, v, gr, NULL, NULL);
820 			if (rmap) {
821 				lock_rmap(rmap);
822 				*rmap |= KVMPPC_RMAP_REFERENCED;
823 				unlock_rmap(rmap);
824 			}
825 		}
826 	}
827 	vcpu->arch.regs.gpr[4] = gr;
828 	ret = H_SUCCESS;
829  out:
830 	unlock_hpte(hpte, v & ~HPTE_V_HVLOCK);
831 	return ret;
832 }
833 EXPORT_SYMBOL_GPL(kvmppc_h_clear_ref);
834 
835 long kvmppc_h_clear_mod(struct kvm_vcpu *vcpu, unsigned long flags,
836 			unsigned long pte_index)
837 {
838 	struct kvm *kvm = vcpu->kvm;
839 	__be64 *hpte;
840 	unsigned long v, r, gr;
841 	struct revmap_entry *rev;
842 	long ret = H_NOT_FOUND;
843 
844 	if (kvm_is_radix(kvm))
845 		return H_FUNCTION;
846 	if (pte_index >= kvmppc_hpt_npte(&kvm->arch.hpt))
847 		return H_PARAMETER;
848 
849 	rev = real_vmalloc_addr(&kvm->arch.hpt.rev[pte_index]);
850 	hpte = (__be64 *)(kvm->arch.hpt.virt + (pte_index << 4));
851 	while (!try_lock_hpte(hpte, HPTE_V_HVLOCK))
852 		cpu_relax();
853 	v = be64_to_cpu(hpte[0]);
854 	r = be64_to_cpu(hpte[1]);
855 	if (!(v & (HPTE_V_VALID | HPTE_V_ABSENT)))
856 		goto out;
857 
858 	gr = rev->guest_rpte;
859 	if (gr & HPTE_R_C) {
860 		rev->guest_rpte &= ~HPTE_R_C;
861 		note_hpte_modification(kvm, rev);
862 	}
863 	if (v & HPTE_V_VALID) {
864 		/* need to make it temporarily absent so C is stable */
865 		hpte[0] |= cpu_to_be64(HPTE_V_ABSENT);
866 		kvmppc_invalidate_hpte(kvm, hpte, pte_index);
867 		r = be64_to_cpu(hpte[1]);
868 		gr |= r & (HPTE_R_R | HPTE_R_C);
869 		if (r & HPTE_R_C) {
870 			hpte[1] = cpu_to_be64(r & ~HPTE_R_C);
871 			eieio();
872 			kvmppc_set_dirty_from_hpte(kvm, v, gr);
873 		}
874 	}
875 	vcpu->arch.regs.gpr[4] = gr;
876 	ret = H_SUCCESS;
877  out:
878 	unlock_hpte(hpte, v & ~HPTE_V_HVLOCK);
879 	return ret;
880 }
881 EXPORT_SYMBOL_GPL(kvmppc_h_clear_mod);
882 
883 static int kvmppc_get_hpa(struct kvm_vcpu *vcpu, unsigned long mmu_seq,
884 			  unsigned long gpa, int writing, unsigned long *hpa,
885 			  struct kvm_memory_slot **memslot_p)
886 {
887 	struct kvm *kvm = vcpu->kvm;
888 	struct kvm_memory_slot *memslot;
889 	unsigned long gfn, hva, pa, psize = PAGE_SHIFT;
890 	unsigned int shift;
891 	pte_t *ptep, pte;
892 
893 	/* Find the memslot for this address */
894 	gfn = gpa >> PAGE_SHIFT;
895 	memslot = __gfn_to_memslot(kvm_memslots_raw(kvm), gfn);
896 	if (!memslot || (memslot->flags & KVM_MEMSLOT_INVALID))
897 		return H_PARAMETER;
898 
899 	/* Translate to host virtual address */
900 	hva = __gfn_to_hva_memslot(memslot, gfn);
901 
902 	/* Try to find the host pte for that virtual address */
903 	ptep = find_kvm_host_pte(kvm, mmu_seq, hva, &shift);
904 	if (!ptep)
905 		return H_TOO_HARD;
906 	pte = kvmppc_read_update_linux_pte(ptep, writing);
907 	if (!pte_present(pte))
908 		return H_TOO_HARD;
909 
910 	/* Convert to a physical address */
911 	if (shift)
912 		psize = 1UL << shift;
913 	pa = pte_pfn(pte) << PAGE_SHIFT;
914 	pa |= hva & (psize - 1);
915 	pa |= gpa & ~PAGE_MASK;
916 
917 	if (hpa)
918 		*hpa = pa;
919 	if (memslot_p)
920 		*memslot_p = memslot;
921 
922 	return H_SUCCESS;
923 }
924 
925 static long kvmppc_do_h_page_init_zero(struct kvm_vcpu *vcpu,
926 				       unsigned long dest)
927 {
928 	struct kvm_memory_slot *memslot;
929 	struct kvm *kvm = vcpu->kvm;
930 	unsigned long pa, mmu_seq;
931 	long ret = H_SUCCESS;
932 	int i;
933 
934 	/* Used later to detect if we might have been invalidated */
935 	mmu_seq = kvm->mmu_invalidate_seq;
936 	smp_rmb();
937 
938 	arch_spin_lock(&kvm->mmu_lock.rlock.raw_lock);
939 
940 	ret = kvmppc_get_hpa(vcpu, mmu_seq, dest, 1, &pa, &memslot);
941 	if (ret != H_SUCCESS)
942 		goto out_unlock;
943 
944 	/* Zero the page */
945 	for (i = 0; i < SZ_4K; i += L1_CACHE_BYTES, pa += L1_CACHE_BYTES)
946 		dcbz((void *)pa);
947 	kvmppc_update_dirty_map(memslot, dest >> PAGE_SHIFT, PAGE_SIZE);
948 
949 out_unlock:
950 	arch_spin_unlock(&kvm->mmu_lock.rlock.raw_lock);
951 	return ret;
952 }
953 
954 static long kvmppc_do_h_page_init_copy(struct kvm_vcpu *vcpu,
955 				       unsigned long dest, unsigned long src)
956 {
957 	unsigned long dest_pa, src_pa, mmu_seq;
958 	struct kvm_memory_slot *dest_memslot;
959 	struct kvm *kvm = vcpu->kvm;
960 	long ret = H_SUCCESS;
961 
962 	/* Used later to detect if we might have been invalidated */
963 	mmu_seq = kvm->mmu_invalidate_seq;
964 	smp_rmb();
965 
966 	arch_spin_lock(&kvm->mmu_lock.rlock.raw_lock);
967 	ret = kvmppc_get_hpa(vcpu, mmu_seq, dest, 1, &dest_pa, &dest_memslot);
968 	if (ret != H_SUCCESS)
969 		goto out_unlock;
970 
971 	ret = kvmppc_get_hpa(vcpu, mmu_seq, src, 0, &src_pa, NULL);
972 	if (ret != H_SUCCESS)
973 		goto out_unlock;
974 
975 	/* Copy the page */
976 	memcpy((void *)dest_pa, (void *)src_pa, SZ_4K);
977 
978 	kvmppc_update_dirty_map(dest_memslot, dest >> PAGE_SHIFT, PAGE_SIZE);
979 
980 out_unlock:
981 	arch_spin_unlock(&kvm->mmu_lock.rlock.raw_lock);
982 	return ret;
983 }
984 
985 long kvmppc_rm_h_page_init(struct kvm_vcpu *vcpu, unsigned long flags,
986 			   unsigned long dest, unsigned long src)
987 {
988 	struct kvm *kvm = vcpu->kvm;
989 	u64 pg_mask = SZ_4K - 1;	/* 4K page size */
990 	long ret = H_SUCCESS;
991 
992 	/* Don't handle radix mode here, go up to the virtual mode handler */
993 	if (kvm_is_radix(kvm))
994 		return H_TOO_HARD;
995 
996 	/* Check for invalid flags (H_PAGE_SET_LOANED covers all CMO flags) */
997 	if (flags & ~(H_ICACHE_INVALIDATE | H_ICACHE_SYNCHRONIZE |
998 		      H_ZERO_PAGE | H_COPY_PAGE | H_PAGE_SET_LOANED))
999 		return H_PARAMETER;
1000 
1001 	/* dest (and src if copy_page flag set) must be page aligned */
1002 	if ((dest & pg_mask) || ((flags & H_COPY_PAGE) && (src & pg_mask)))
1003 		return H_PARAMETER;
1004 
1005 	/* zero and/or copy the page as determined by the flags */
1006 	if (flags & H_COPY_PAGE)
1007 		ret = kvmppc_do_h_page_init_copy(vcpu, dest, src);
1008 	else if (flags & H_ZERO_PAGE)
1009 		ret = kvmppc_do_h_page_init_zero(vcpu, dest);
1010 
1011 	/* We can ignore the other flags */
1012 
1013 	return ret;
1014 }
1015 
1016 void kvmppc_invalidate_hpte(struct kvm *kvm, __be64 *hptep,
1017 			unsigned long pte_index)
1018 {
1019 	unsigned long rb;
1020 	u64 hp0, hp1;
1021 
1022 	hptep[0] &= ~cpu_to_be64(HPTE_V_VALID);
1023 	hp0 = be64_to_cpu(hptep[0]);
1024 	hp1 = be64_to_cpu(hptep[1]);
1025 	if (cpu_has_feature(CPU_FTR_ARCH_300)) {
1026 		hp0 = hpte_new_to_old_v(hp0, hp1);
1027 		hp1 = hpte_new_to_old_r(hp1);
1028 	}
1029 	rb = compute_tlbie_rb(hp0, hp1, pte_index);
1030 	do_tlbies(kvm, &rb, 1, 1, true);
1031 }
1032 EXPORT_SYMBOL_GPL(kvmppc_invalidate_hpte);
1033 
1034 void kvmppc_clear_ref_hpte(struct kvm *kvm, __be64 *hptep,
1035 			   unsigned long pte_index)
1036 {
1037 	unsigned long rb;
1038 	unsigned char rbyte;
1039 	u64 hp0, hp1;
1040 
1041 	hp0 = be64_to_cpu(hptep[0]);
1042 	hp1 = be64_to_cpu(hptep[1]);
1043 	if (cpu_has_feature(CPU_FTR_ARCH_300)) {
1044 		hp0 = hpte_new_to_old_v(hp0, hp1);
1045 		hp1 = hpte_new_to_old_r(hp1);
1046 	}
1047 	rb = compute_tlbie_rb(hp0, hp1, pte_index);
1048 	rbyte = (be64_to_cpu(hptep[1]) & ~HPTE_R_R) >> 8;
1049 	/* modify only the second-last byte, which contains the ref bit */
1050 	*((char *)hptep + 14) = rbyte;
1051 	do_tlbies(kvm, &rb, 1, 1, false);
1052 }
1053 EXPORT_SYMBOL_GPL(kvmppc_clear_ref_hpte);
1054 
1055 static int slb_base_page_shift[4] = {
1056 	24,	/* 16M */
1057 	16,	/* 64k */
1058 	34,	/* 16G */
1059 	20,	/* 1M, unsupported */
1060 };
1061 
1062 static struct mmio_hpte_cache_entry *mmio_cache_search(struct kvm_vcpu *vcpu,
1063 		unsigned long eaddr, unsigned long slb_v, long mmio_update)
1064 {
1065 	struct mmio_hpte_cache_entry *entry = NULL;
1066 	unsigned int pshift;
1067 	unsigned int i;
1068 
1069 	for (i = 0; i < MMIO_HPTE_CACHE_SIZE; i++) {
1070 		entry = &vcpu->arch.mmio_cache.entry[i];
1071 		if (entry->mmio_update == mmio_update) {
1072 			pshift = entry->slb_base_pshift;
1073 			if ((entry->eaddr >> pshift) == (eaddr >> pshift) &&
1074 			    entry->slb_v == slb_v)
1075 				return entry;
1076 		}
1077 	}
1078 	return NULL;
1079 }
1080 
1081 static struct mmio_hpte_cache_entry *
1082 			next_mmio_cache_entry(struct kvm_vcpu *vcpu)
1083 {
1084 	unsigned int index = vcpu->arch.mmio_cache.index;
1085 
1086 	vcpu->arch.mmio_cache.index++;
1087 	if (vcpu->arch.mmio_cache.index == MMIO_HPTE_CACHE_SIZE)
1088 		vcpu->arch.mmio_cache.index = 0;
1089 
1090 	return &vcpu->arch.mmio_cache.entry[index];
1091 }
1092 
1093 /* When called from virtmode, this func should be protected by
1094  * preempt_disable(), otherwise, the holding of HPTE_V_HVLOCK
1095  * can trigger deadlock issue.
1096  */
1097 long kvmppc_hv_find_lock_hpte(struct kvm *kvm, gva_t eaddr, unsigned long slb_v,
1098 			      unsigned long valid)
1099 {
1100 	unsigned int i;
1101 	unsigned int pshift;
1102 	unsigned long somask;
1103 	unsigned long vsid, hash;
1104 	unsigned long avpn;
1105 	__be64 *hpte;
1106 	unsigned long mask, val;
1107 	unsigned long v, r, orig_v;
1108 
1109 	/* Get page shift, work out hash and AVPN etc. */
1110 	mask = SLB_VSID_B | HPTE_V_AVPN | HPTE_V_SECONDARY;
1111 	val = 0;
1112 	pshift = 12;
1113 	if (slb_v & SLB_VSID_L) {
1114 		mask |= HPTE_V_LARGE;
1115 		val |= HPTE_V_LARGE;
1116 		pshift = slb_base_page_shift[(slb_v & SLB_VSID_LP) >> 4];
1117 	}
1118 	if (slb_v & SLB_VSID_B_1T) {
1119 		somask = (1UL << 40) - 1;
1120 		vsid = (slb_v & ~SLB_VSID_B) >> SLB_VSID_SHIFT_1T;
1121 		vsid ^= vsid << 25;
1122 	} else {
1123 		somask = (1UL << 28) - 1;
1124 		vsid = (slb_v & ~SLB_VSID_B) >> SLB_VSID_SHIFT;
1125 	}
1126 	hash = (vsid ^ ((eaddr & somask) >> pshift)) & kvmppc_hpt_mask(&kvm->arch.hpt);
1127 	avpn = slb_v & ~(somask >> 16);	/* also includes B */
1128 	avpn |= (eaddr & somask) >> 16;
1129 
1130 	if (pshift >= 24)
1131 		avpn &= ~((1UL << (pshift - 16)) - 1);
1132 	else
1133 		avpn &= ~0x7fUL;
1134 	val |= avpn;
1135 
1136 	for (;;) {
1137 		hpte = (__be64 *)(kvm->arch.hpt.virt + (hash << 7));
1138 
1139 		for (i = 0; i < 16; i += 2) {
1140 			/* Read the PTE racily */
1141 			v = be64_to_cpu(hpte[i]) & ~HPTE_V_HVLOCK;
1142 			if (cpu_has_feature(CPU_FTR_ARCH_300))
1143 				v = hpte_new_to_old_v(v, be64_to_cpu(hpte[i+1]));
1144 
1145 			/* Check valid/absent, hash, segment size and AVPN */
1146 			if (!(v & valid) || (v & mask) != val)
1147 				continue;
1148 
1149 			/* Lock the PTE and read it under the lock */
1150 			while (!try_lock_hpte(&hpte[i], HPTE_V_HVLOCK))
1151 				cpu_relax();
1152 			v = orig_v = be64_to_cpu(hpte[i]) & ~HPTE_V_HVLOCK;
1153 			r = be64_to_cpu(hpte[i+1]);
1154 			if (cpu_has_feature(CPU_FTR_ARCH_300)) {
1155 				v = hpte_new_to_old_v(v, r);
1156 				r = hpte_new_to_old_r(r);
1157 			}
1158 
1159 			/*
1160 			 * Check the HPTE again, including base page size
1161 			 */
1162 			if ((v & valid) && (v & mask) == val &&
1163 			    kvmppc_hpte_base_page_shift(v, r) == pshift)
1164 				/* Return with the HPTE still locked */
1165 				return (hash << 3) + (i >> 1);
1166 
1167 			__unlock_hpte(&hpte[i], orig_v);
1168 		}
1169 
1170 		if (val & HPTE_V_SECONDARY)
1171 			break;
1172 		val |= HPTE_V_SECONDARY;
1173 		hash = hash ^ kvmppc_hpt_mask(&kvm->arch.hpt);
1174 	}
1175 	return -1;
1176 }
1177 EXPORT_SYMBOL(kvmppc_hv_find_lock_hpte);
1178 
1179 /*
1180  * Called in real mode to check whether an HPTE not found fault
1181  * is due to accessing a paged-out page or an emulated MMIO page,
1182  * or if a protection fault is due to accessing a page that the
1183  * guest wanted read/write access to but which we made read-only.
1184  * Returns a possibly modified status (DSISR) value if not
1185  * (i.e. pass the interrupt to the guest),
1186  * -1 to pass the fault up to host kernel mode code, -2 to do that
1187  * and also load the instruction word (for MMIO emulation),
1188  * or 0 if we should make the guest retry the access.
1189  */
1190 long kvmppc_hpte_hv_fault(struct kvm_vcpu *vcpu, unsigned long addr,
1191 			  unsigned long slb_v, unsigned int status, bool data)
1192 {
1193 	struct kvm *kvm = vcpu->kvm;
1194 	long int index;
1195 	unsigned long v, r, gr, orig_v;
1196 	__be64 *hpte;
1197 	unsigned long valid;
1198 	struct revmap_entry *rev;
1199 	unsigned long pp, key;
1200 	struct mmio_hpte_cache_entry *cache_entry = NULL;
1201 	long mmio_update = 0;
1202 
1203 	/* For protection fault, expect to find a valid HPTE */
1204 	valid = HPTE_V_VALID;
1205 	if (status & DSISR_NOHPTE) {
1206 		valid |= HPTE_V_ABSENT;
1207 		mmio_update = atomic64_read(&kvm->arch.mmio_update);
1208 		cache_entry = mmio_cache_search(vcpu, addr, slb_v, mmio_update);
1209 	}
1210 	if (cache_entry) {
1211 		index = cache_entry->pte_index;
1212 		v = cache_entry->hpte_v;
1213 		r = cache_entry->hpte_r;
1214 		gr = cache_entry->rpte;
1215 	} else {
1216 		index = kvmppc_hv_find_lock_hpte(kvm, addr, slb_v, valid);
1217 		if (index < 0) {
1218 			if (status & DSISR_NOHPTE)
1219 				return status;	/* there really was no HPTE */
1220 			return 0;	/* for prot fault, HPTE disappeared */
1221 		}
1222 		hpte = (__be64 *)(kvm->arch.hpt.virt + (index << 4));
1223 		v = orig_v = be64_to_cpu(hpte[0]) & ~HPTE_V_HVLOCK;
1224 		r = be64_to_cpu(hpte[1]);
1225 		if (cpu_has_feature(CPU_FTR_ARCH_300)) {
1226 			v = hpte_new_to_old_v(v, r);
1227 			r = hpte_new_to_old_r(r);
1228 		}
1229 		rev = real_vmalloc_addr(&kvm->arch.hpt.rev[index]);
1230 		gr = rev->guest_rpte;
1231 
1232 		unlock_hpte(hpte, orig_v);
1233 	}
1234 
1235 	/* For not found, if the HPTE is valid by now, retry the instruction */
1236 	if ((status & DSISR_NOHPTE) && (v & HPTE_V_VALID))
1237 		return 0;
1238 
1239 	/* Check access permissions to the page */
1240 	pp = gr & (HPTE_R_PP0 | HPTE_R_PP);
1241 	key = (vcpu->arch.shregs.msr & MSR_PR) ? SLB_VSID_KP : SLB_VSID_KS;
1242 	status &= ~DSISR_NOHPTE;	/* DSISR_NOHPTE == SRR1_ISI_NOPT */
1243 	if (!data) {
1244 		if (gr & (HPTE_R_N | HPTE_R_G))
1245 			return status | SRR1_ISI_N_G_OR_CIP;
1246 		if (!hpte_read_permission(pp, slb_v & key))
1247 			return status | SRR1_ISI_PROT;
1248 	} else if (status & DSISR_ISSTORE) {
1249 		/* check write permission */
1250 		if (!hpte_write_permission(pp, slb_v & key))
1251 			return status | DSISR_PROTFAULT;
1252 	} else {
1253 		if (!hpte_read_permission(pp, slb_v & key))
1254 			return status | DSISR_PROTFAULT;
1255 	}
1256 
1257 	/* Check storage key, if applicable */
1258 	if (data && (vcpu->arch.shregs.msr & MSR_DR)) {
1259 		unsigned int perm = hpte_get_skey_perm(gr, vcpu->arch.amr);
1260 		if (status & DSISR_ISSTORE)
1261 			perm >>= 1;
1262 		if (perm & 1)
1263 			return status | DSISR_KEYFAULT;
1264 	}
1265 
1266 	/* Save HPTE info for virtual-mode handler */
1267 	vcpu->arch.pgfault_addr = addr;
1268 	vcpu->arch.pgfault_index = index;
1269 	vcpu->arch.pgfault_hpte[0] = v;
1270 	vcpu->arch.pgfault_hpte[1] = r;
1271 	vcpu->arch.pgfault_cache = cache_entry;
1272 
1273 	/* Check the storage key to see if it is possibly emulated MMIO */
1274 	if ((r & (HPTE_R_KEY_HI | HPTE_R_KEY_LO)) ==
1275 	    (HPTE_R_KEY_HI | HPTE_R_KEY_LO)) {
1276 		if (!cache_entry) {
1277 			unsigned int pshift = 12;
1278 			unsigned int pshift_index;
1279 
1280 			if (slb_v & SLB_VSID_L) {
1281 				pshift_index = ((slb_v & SLB_VSID_LP) >> 4);
1282 				pshift = slb_base_page_shift[pshift_index];
1283 			}
1284 			cache_entry = next_mmio_cache_entry(vcpu);
1285 			cache_entry->eaddr = addr;
1286 			cache_entry->slb_base_pshift = pshift;
1287 			cache_entry->pte_index = index;
1288 			cache_entry->hpte_v = v;
1289 			cache_entry->hpte_r = r;
1290 			cache_entry->rpte = gr;
1291 			cache_entry->slb_v = slb_v;
1292 			cache_entry->mmio_update = mmio_update;
1293 		}
1294 		if (data && (vcpu->arch.shregs.msr & MSR_IR))
1295 			return -2;	/* MMIO emulation - load instr word */
1296 	}
1297 
1298 	return -1;		/* send fault up to host kernel mode */
1299 }
1300 EXPORT_SYMBOL_GPL(kvmppc_hpte_hv_fault);
1301