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 *x)
27 {
28 	unsigned long addr = (unsigned long) x;
29 	pte_t *p;
30 	/*
31 	 * assume we don't have huge pages in vmalloc space...
32 	 * So don't worry about THP collapse/split. Called
33 	 * Only in realmode with MSR_EE = 0, hence won't need irq_save/restore.
34 	 */
35 	p = find_init_mm_pte(addr, NULL);
36 	if (!p || !pte_present(*p))
37 		return NULL;
38 	addr = (pte_pfn(*p) << PAGE_SHIFT) | (addr & ~PAGE_MASK);
39 	return __va(addr);
40 }
41 
42 /* Return 1 if we need to do a global tlbie, 0 if we can use tlbiel */
43 static int global_invalidates(struct kvm *kvm)
44 {
45 	int global;
46 	int cpu;
47 
48 	/*
49 	 * If there is only one vcore, and it's currently running,
50 	 * as indicated by local_paca->kvm_hstate.kvm_vcpu being set,
51 	 * we can use tlbiel as long as we mark all other physical
52 	 * cores as potentially having stale TLB entries for this lpid.
53 	 * Otherwise, don't use tlbiel.
54 	 */
55 	if (kvm->arch.online_vcores == 1 && local_paca->kvm_hstate.kvm_vcpu)
56 		global = 0;
57 	else
58 		global = 1;
59 
60 	if (!global) {
61 		/* any other core might now have stale TLB entries... */
62 		smp_wmb();
63 		cpumask_setall(&kvm->arch.need_tlb_flush);
64 		cpu = local_paca->kvm_hstate.kvm_vcore->pcpu;
65 		/*
66 		 * On POWER9, threads are independent but the TLB is shared,
67 		 * so use the bit for the first thread to represent the core.
68 		 */
69 		if (cpu_has_feature(CPU_FTR_ARCH_300))
70 			cpu = cpu_first_thread_sibling(cpu);
71 		cpumask_clear_cpu(cpu, &kvm->arch.need_tlb_flush);
72 	}
73 
74 	return global;
75 }
76 
77 /*
78  * Add this HPTE into the chain for the real page.
79  * Must be called with the chain locked; it unlocks the chain.
80  */
81 void kvmppc_add_revmap_chain(struct kvm *kvm, struct revmap_entry *rev,
82 			     unsigned long *rmap, long pte_index, int realmode)
83 {
84 	struct revmap_entry *head, *tail;
85 	unsigned long i;
86 
87 	if (*rmap & KVMPPC_RMAP_PRESENT) {
88 		i = *rmap & KVMPPC_RMAP_INDEX;
89 		head = &kvm->arch.hpt.rev[i];
90 		if (realmode)
91 			head = real_vmalloc_addr(head);
92 		tail = &kvm->arch.hpt.rev[head->back];
93 		if (realmode)
94 			tail = real_vmalloc_addr(tail);
95 		rev->forw = i;
96 		rev->back = head->back;
97 		tail->forw = pte_index;
98 		head->back = pte_index;
99 	} else {
100 		rev->forw = rev->back = pte_index;
101 		*rmap = (*rmap & ~KVMPPC_RMAP_INDEX) |
102 			pte_index | KVMPPC_RMAP_PRESENT;
103 	}
104 	unlock_rmap(rmap);
105 }
106 EXPORT_SYMBOL_GPL(kvmppc_add_revmap_chain);
107 
108 /* Update the dirty bitmap of a memslot */
109 void kvmppc_update_dirty_map(const struct kvm_memory_slot *memslot,
110 			     unsigned long gfn, unsigned long psize)
111 {
112 	unsigned long npages;
113 
114 	if (!psize || !memslot->dirty_bitmap)
115 		return;
116 	npages = (psize + PAGE_SIZE - 1) / PAGE_SIZE;
117 	gfn -= memslot->base_gfn;
118 	set_dirty_bits_atomic(memslot->dirty_bitmap, gfn, npages);
119 }
120 EXPORT_SYMBOL_GPL(kvmppc_update_dirty_map);
121 
122 static void kvmppc_set_dirty_from_hpte(struct kvm *kvm,
123 				unsigned long hpte_v, unsigned long hpte_gr)
124 {
125 	struct kvm_memory_slot *memslot;
126 	unsigned long gfn;
127 	unsigned long psize;
128 
129 	psize = kvmppc_actual_pgsz(hpte_v, hpte_gr);
130 	gfn = hpte_rpn(hpte_gr, psize);
131 	memslot = __gfn_to_memslot(kvm_memslots_raw(kvm), gfn);
132 	if (memslot && memslot->dirty_bitmap)
133 		kvmppc_update_dirty_map(memslot, gfn, psize);
134 }
135 
136 /* Returns a pointer to the revmap entry for the page mapped by a HPTE */
137 static unsigned long *revmap_for_hpte(struct kvm *kvm, unsigned long hpte_v,
138 				      unsigned long hpte_gr,
139 				      struct kvm_memory_slot **memslotp,
140 				      unsigned long *gfnp)
141 {
142 	struct kvm_memory_slot *memslot;
143 	unsigned long *rmap;
144 	unsigned long gfn;
145 
146 	gfn = hpte_rpn(hpte_gr, kvmppc_actual_pgsz(hpte_v, hpte_gr));
147 	memslot = __gfn_to_memslot(kvm_memslots_raw(kvm), gfn);
148 	if (memslotp)
149 		*memslotp = memslot;
150 	if (gfnp)
151 		*gfnp = gfn;
152 	if (!memslot)
153 		return NULL;
154 
155 	rmap = real_vmalloc_addr(&memslot->arch.rmap[gfn - memslot->base_gfn]);
156 	return rmap;
157 }
158 
159 /* Remove this HPTE from the chain for a real page */
160 static void remove_revmap_chain(struct kvm *kvm, long pte_index,
161 				struct revmap_entry *rev,
162 				unsigned long hpte_v, unsigned long hpte_r)
163 {
164 	struct revmap_entry *next, *prev;
165 	unsigned long ptel, head;
166 	unsigned long *rmap;
167 	unsigned long rcbits;
168 	struct kvm_memory_slot *memslot;
169 	unsigned long gfn;
170 
171 	rcbits = hpte_r & (HPTE_R_R | HPTE_R_C);
172 	ptel = rev->guest_rpte |= rcbits;
173 	rmap = revmap_for_hpte(kvm, hpte_v, ptel, &memslot, &gfn);
174 	if (!rmap)
175 		return;
176 	lock_rmap(rmap);
177 
178 	head = *rmap & KVMPPC_RMAP_INDEX;
179 	next = real_vmalloc_addr(&kvm->arch.hpt.rev[rev->forw]);
180 	prev = real_vmalloc_addr(&kvm->arch.hpt.rev[rev->back]);
181 	next->back = rev->back;
182 	prev->forw = rev->forw;
183 	if (head == pte_index) {
184 		head = rev->forw;
185 		if (head == pte_index)
186 			*rmap &= ~(KVMPPC_RMAP_PRESENT | KVMPPC_RMAP_INDEX);
187 		else
188 			*rmap = (*rmap & ~KVMPPC_RMAP_INDEX) | head;
189 	}
190 	*rmap |= rcbits << KVMPPC_RMAP_RC_SHIFT;
191 	if (rcbits & HPTE_R_C)
192 		kvmppc_update_dirty_map(memslot, gfn,
193 					kvmppc_actual_pgsz(hpte_v, hpte_r));
194 	unlock_rmap(rmap);
195 }
196 
197 long kvmppc_do_h_enter(struct kvm *kvm, unsigned long flags,
198 		       long pte_index, unsigned long pteh, unsigned long ptel,
199 		       pgd_t *pgdir, bool realmode, unsigned long *pte_idx_ret)
200 {
201 	unsigned long i, pa, gpa, gfn, psize;
202 	unsigned long slot_fn, hva;
203 	__be64 *hpte;
204 	struct revmap_entry *rev;
205 	unsigned long g_ptel;
206 	struct kvm_memory_slot *memslot;
207 	unsigned hpage_shift;
208 	bool is_ci;
209 	unsigned long *rmap;
210 	pte_t *ptep;
211 	unsigned int writing;
212 	unsigned long mmu_seq;
213 	unsigned long rcbits, irq_flags = 0;
214 
215 	if (kvm_is_radix(kvm))
216 		return H_FUNCTION;
217 	psize = kvmppc_actual_pgsz(pteh, ptel);
218 	if (!psize)
219 		return H_PARAMETER;
220 	writing = hpte_is_writable(ptel);
221 	pteh &= ~(HPTE_V_HVLOCK | HPTE_V_ABSENT | HPTE_V_VALID);
222 	ptel &= ~HPTE_GR_RESERVED;
223 	g_ptel = ptel;
224 
225 	/* used later to detect if we might have been invalidated */
226 	mmu_seq = kvm->mmu_notifier_seq;
227 	smp_rmb();
228 
229 	/* Find the memslot (if any) for this address */
230 	gpa = (ptel & HPTE_R_RPN) & ~(psize - 1);
231 	gfn = gpa >> PAGE_SHIFT;
232 	memslot = __gfn_to_memslot(kvm_memslots_raw(kvm), gfn);
233 	pa = 0;
234 	is_ci = false;
235 	rmap = NULL;
236 	if (!(memslot && !(memslot->flags & KVM_MEMSLOT_INVALID))) {
237 		/* Emulated MMIO - mark this with key=31 */
238 		pteh |= HPTE_V_ABSENT;
239 		ptel |= HPTE_R_KEY_HI | HPTE_R_KEY_LO;
240 		goto do_insert;
241 	}
242 
243 	/* Check if the requested page fits entirely in the memslot. */
244 	if (!slot_is_aligned(memslot, psize))
245 		return H_PARAMETER;
246 	slot_fn = gfn - memslot->base_gfn;
247 	rmap = &memslot->arch.rmap[slot_fn];
248 
249 	/* Translate to host virtual address */
250 	hva = __gfn_to_hva_memslot(memslot, gfn);
251 	/*
252 	 * If we had a page table table change after lookup, we would
253 	 * retry via mmu_notifier_retry.
254 	 */
255 	if (!realmode)
256 		local_irq_save(irq_flags);
257 	/*
258 	 * If called in real mode we have MSR_EE = 0. Otherwise
259 	 * we disable irq above.
260 	 */
261 	ptep = __find_linux_pte(pgdir, hva, NULL, &hpage_shift);
262 	if (ptep) {
263 		pte_t pte;
264 		unsigned int host_pte_size;
265 
266 		if (hpage_shift)
267 			host_pte_size = 1ul << hpage_shift;
268 		else
269 			host_pte_size = PAGE_SIZE;
270 		/*
271 		 * We should always find the guest page size
272 		 * to <= host page size, if host is using hugepage
273 		 */
274 		if (host_pte_size < psize) {
275 			if (!realmode)
276 				local_irq_restore(flags);
277 			return H_PARAMETER;
278 		}
279 		pte = kvmppc_read_update_linux_pte(ptep, writing);
280 		if (pte_present(pte) && !pte_protnone(pte)) {
281 			if (writing && !__pte_write(pte))
282 				/* make the actual HPTE be read-only */
283 				ptel = hpte_make_readonly(ptel);
284 			is_ci = pte_ci(pte);
285 			pa = pte_pfn(pte) << PAGE_SHIFT;
286 			pa |= hva & (host_pte_size - 1);
287 			pa |= gpa & ~PAGE_MASK;
288 		}
289 	}
290 	if (!realmode)
291 		local_irq_restore(irq_flags);
292 
293 	ptel &= HPTE_R_KEY | HPTE_R_PP0 | (psize-1);
294 	ptel |= pa;
295 
296 	if (pa)
297 		pteh |= HPTE_V_VALID;
298 	else {
299 		pteh |= HPTE_V_ABSENT;
300 		ptel &= ~(HPTE_R_KEY_HI | HPTE_R_KEY_LO);
301 	}
302 
303 	/*If we had host pte mapping then  Check WIMG */
304 	if (ptep && !hpte_cache_flags_ok(ptel, is_ci)) {
305 		if (is_ci)
306 			return H_PARAMETER;
307 		/*
308 		 * Allow guest to map emulated device memory as
309 		 * uncacheable, but actually make it cacheable.
310 		 */
311 		ptel &= ~(HPTE_R_W|HPTE_R_I|HPTE_R_G);
312 		ptel |= HPTE_R_M;
313 	}
314 
315 	/* Find and lock the HPTEG slot to use */
316  do_insert:
317 	if (pte_index >= kvmppc_hpt_npte(&kvm->arch.hpt))
318 		return H_PARAMETER;
319 	if (likely((flags & H_EXACT) == 0)) {
320 		pte_index &= ~7UL;
321 		hpte = (__be64 *)(kvm->arch.hpt.virt + (pte_index << 4));
322 		for (i = 0; i < 8; ++i) {
323 			if ((be64_to_cpu(*hpte) & HPTE_V_VALID) == 0 &&
324 			    try_lock_hpte(hpte, HPTE_V_HVLOCK | HPTE_V_VALID |
325 					  HPTE_V_ABSENT))
326 				break;
327 			hpte += 2;
328 		}
329 		if (i == 8) {
330 			/*
331 			 * Since try_lock_hpte doesn't retry (not even stdcx.
332 			 * failures), it could be that there is a free slot
333 			 * but we transiently failed to lock it.  Try again,
334 			 * actually locking each slot and checking it.
335 			 */
336 			hpte -= 16;
337 			for (i = 0; i < 8; ++i) {
338 				u64 pte;
339 				while (!try_lock_hpte(hpte, HPTE_V_HVLOCK))
340 					cpu_relax();
341 				pte = be64_to_cpu(hpte[0]);
342 				if (!(pte & (HPTE_V_VALID | HPTE_V_ABSENT)))
343 					break;
344 				__unlock_hpte(hpte, pte);
345 				hpte += 2;
346 			}
347 			if (i == 8)
348 				return H_PTEG_FULL;
349 		}
350 		pte_index += i;
351 	} else {
352 		hpte = (__be64 *)(kvm->arch.hpt.virt + (pte_index << 4));
353 		if (!try_lock_hpte(hpte, HPTE_V_HVLOCK | HPTE_V_VALID |
354 				   HPTE_V_ABSENT)) {
355 			/* Lock the slot and check again */
356 			u64 pte;
357 
358 			while (!try_lock_hpte(hpte, HPTE_V_HVLOCK))
359 				cpu_relax();
360 			pte = be64_to_cpu(hpte[0]);
361 			if (pte & (HPTE_V_VALID | HPTE_V_ABSENT)) {
362 				__unlock_hpte(hpte, pte);
363 				return H_PTEG_FULL;
364 			}
365 		}
366 	}
367 
368 	/* Save away the guest's idea of the second HPTE dword */
369 	rev = &kvm->arch.hpt.rev[pte_index];
370 	if (realmode)
371 		rev = real_vmalloc_addr(rev);
372 	if (rev) {
373 		rev->guest_rpte = g_ptel;
374 		note_hpte_modification(kvm, rev);
375 	}
376 
377 	/* Link HPTE into reverse-map chain */
378 	if (pteh & HPTE_V_VALID) {
379 		if (realmode)
380 			rmap = real_vmalloc_addr(rmap);
381 		lock_rmap(rmap);
382 		/* Check for pending invalidations under the rmap chain lock */
383 		if (mmu_notifier_retry(kvm, mmu_seq)) {
384 			/* inval in progress, write a non-present HPTE */
385 			pteh |= HPTE_V_ABSENT;
386 			pteh &= ~HPTE_V_VALID;
387 			ptel &= ~(HPTE_R_KEY_HI | HPTE_R_KEY_LO);
388 			unlock_rmap(rmap);
389 		} else {
390 			kvmppc_add_revmap_chain(kvm, rev, rmap, pte_index,
391 						realmode);
392 			/* Only set R/C in real HPTE if already set in *rmap */
393 			rcbits = *rmap >> KVMPPC_RMAP_RC_SHIFT;
394 			ptel &= rcbits | ~(HPTE_R_R | HPTE_R_C);
395 		}
396 	}
397 
398 	/* Convert to new format on P9 */
399 	if (cpu_has_feature(CPU_FTR_ARCH_300)) {
400 		ptel = hpte_old_to_new_r(pteh, ptel);
401 		pteh = hpte_old_to_new_v(pteh);
402 	}
403 	hpte[1] = cpu_to_be64(ptel);
404 
405 	/* Write the first HPTE dword, unlocking the HPTE and making it valid */
406 	eieio();
407 	__unlock_hpte(hpte, pteh);
408 	asm volatile("ptesync" : : : "memory");
409 
410 	*pte_idx_ret = pte_index;
411 	return H_SUCCESS;
412 }
413 EXPORT_SYMBOL_GPL(kvmppc_do_h_enter);
414 
415 long kvmppc_h_enter(struct kvm_vcpu *vcpu, unsigned long flags,
416 		    long pte_index, unsigned long pteh, unsigned long ptel)
417 {
418 	return kvmppc_do_h_enter(vcpu->kvm, flags, pte_index, pteh, ptel,
419 				 vcpu->arch.pgdir, true,
420 				 &vcpu->arch.regs.gpr[4]);
421 }
422 
423 #ifdef __BIG_ENDIAN__
424 #define LOCK_TOKEN	(*(u32 *)(&get_paca()->lock_token))
425 #else
426 #define LOCK_TOKEN	(*(u32 *)(&get_paca()->paca_index))
427 #endif
428 
429 static inline int is_mmio_hpte(unsigned long v, unsigned long r)
430 {
431 	return ((v & HPTE_V_ABSENT) &&
432 		(r & (HPTE_R_KEY_HI | HPTE_R_KEY_LO)) ==
433 		(HPTE_R_KEY_HI | HPTE_R_KEY_LO));
434 }
435 
436 static void do_tlbies(struct kvm *kvm, unsigned long *rbvalues,
437 		      long npages, int global, bool need_sync)
438 {
439 	long i;
440 
441 	/*
442 	 * We use the POWER9 5-operand versions of tlbie and tlbiel here.
443 	 * Since we are using RIC=0 PRS=0 R=0, and P7/P8 tlbiel ignores
444 	 * the RS field, this is backwards-compatible with P7 and P8.
445 	 */
446 	if (global) {
447 		if (need_sync)
448 			asm volatile("ptesync" : : : "memory");
449 		for (i = 0; i < npages; ++i) {
450 			asm volatile(PPC_TLBIE_5(%0,%1,0,0,0) : :
451 				     "r" (rbvalues[i]), "r" (kvm->arch.lpid));
452 		}
453 
454 		if (cpu_has_feature(CPU_FTR_P9_TLBIE_BUG)) {
455 			/*
456 			 * Need the extra ptesync to make sure we don't
457 			 * re-order the tlbie
458 			 */
459 			asm volatile("ptesync": : :"memory");
460 			asm volatile(PPC_TLBIE_5(%0,%1,0,0,0) : :
461 				     "r" (rbvalues[0]), "r" (kvm->arch.lpid));
462 		}
463 
464 		asm volatile("eieio; tlbsync; ptesync" : : : "memory");
465 	} else {
466 		if (need_sync)
467 			asm volatile("ptesync" : : : "memory");
468 		for (i = 0; i < npages; ++i) {
469 			asm volatile(PPC_TLBIEL(%0,%1,0,0,0) : :
470 				     "r" (rbvalues[i]), "r" (0));
471 		}
472 		asm volatile("ptesync" : : : "memory");
473 	}
474 }
475 
476 long kvmppc_do_h_remove(struct kvm *kvm, unsigned long flags,
477 			unsigned long pte_index, unsigned long avpn,
478 			unsigned long *hpret)
479 {
480 	__be64 *hpte;
481 	unsigned long v, r, rb;
482 	struct revmap_entry *rev;
483 	u64 pte, orig_pte, pte_r;
484 
485 	if (kvm_is_radix(kvm))
486 		return H_FUNCTION;
487 	if (pte_index >= kvmppc_hpt_npte(&kvm->arch.hpt))
488 		return H_PARAMETER;
489 	hpte = (__be64 *)(kvm->arch.hpt.virt + (pte_index << 4));
490 	while (!try_lock_hpte(hpte, HPTE_V_HVLOCK))
491 		cpu_relax();
492 	pte = orig_pte = be64_to_cpu(hpte[0]);
493 	pte_r = be64_to_cpu(hpte[1]);
494 	if (cpu_has_feature(CPU_FTR_ARCH_300)) {
495 		pte = hpte_new_to_old_v(pte, pte_r);
496 		pte_r = hpte_new_to_old_r(pte_r);
497 	}
498 	if ((pte & (HPTE_V_ABSENT | HPTE_V_VALID)) == 0 ||
499 	    ((flags & H_AVPN) && (pte & ~0x7fUL) != avpn) ||
500 	    ((flags & H_ANDCOND) && (pte & avpn) != 0)) {
501 		__unlock_hpte(hpte, orig_pte);
502 		return H_NOT_FOUND;
503 	}
504 
505 	rev = real_vmalloc_addr(&kvm->arch.hpt.rev[pte_index]);
506 	v = pte & ~HPTE_V_HVLOCK;
507 	if (v & HPTE_V_VALID) {
508 		hpte[0] &= ~cpu_to_be64(HPTE_V_VALID);
509 		rb = compute_tlbie_rb(v, pte_r, pte_index);
510 		do_tlbies(kvm, &rb, 1, global_invalidates(kvm), true);
511 		/*
512 		 * The reference (R) and change (C) bits in a HPT
513 		 * entry can be set by hardware at any time up until
514 		 * the HPTE is invalidated and the TLB invalidation
515 		 * sequence has completed.  This means that when
516 		 * removing a HPTE, we need to re-read the HPTE after
517 		 * the invalidation sequence has completed in order to
518 		 * obtain reliable values of R and C.
519 		 */
520 		remove_revmap_chain(kvm, pte_index, rev, v,
521 				    be64_to_cpu(hpte[1]));
522 	}
523 	r = rev->guest_rpte & ~HPTE_GR_RESERVED;
524 	note_hpte_modification(kvm, rev);
525 	unlock_hpte(hpte, 0);
526 
527 	if (is_mmio_hpte(v, pte_r))
528 		atomic64_inc(&kvm->arch.mmio_update);
529 
530 	if (v & HPTE_V_ABSENT)
531 		v = (v & ~HPTE_V_ABSENT) | HPTE_V_VALID;
532 	hpret[0] = v;
533 	hpret[1] = r;
534 	return H_SUCCESS;
535 }
536 EXPORT_SYMBOL_GPL(kvmppc_do_h_remove);
537 
538 long kvmppc_h_remove(struct kvm_vcpu *vcpu, unsigned long flags,
539 		     unsigned long pte_index, unsigned long avpn)
540 {
541 	return kvmppc_do_h_remove(vcpu->kvm, flags, pte_index, avpn,
542 				  &vcpu->arch.regs.gpr[4]);
543 }
544 
545 long kvmppc_h_bulk_remove(struct kvm_vcpu *vcpu)
546 {
547 	struct kvm *kvm = vcpu->kvm;
548 	unsigned long *args = &vcpu->arch.regs.gpr[4];
549 	__be64 *hp, *hptes[4];
550 	unsigned long tlbrb[4];
551 	long int i, j, k, n, found, indexes[4];
552 	unsigned long flags, req, pte_index, rcbits;
553 	int global;
554 	long int ret = H_SUCCESS;
555 	struct revmap_entry *rev, *revs[4];
556 	u64 hp0, hp1;
557 
558 	if (kvm_is_radix(kvm))
559 		return H_FUNCTION;
560 	global = global_invalidates(kvm);
561 	for (i = 0; i < 4 && ret == H_SUCCESS; ) {
562 		n = 0;
563 		for (; i < 4; ++i) {
564 			j = i * 2;
565 			pte_index = args[j];
566 			flags = pte_index >> 56;
567 			pte_index &= ((1ul << 56) - 1);
568 			req = flags >> 6;
569 			flags &= 3;
570 			if (req == 3) {		/* no more requests */
571 				i = 4;
572 				break;
573 			}
574 			if (req != 1 || flags == 3 ||
575 			    pte_index >= kvmppc_hpt_npte(&kvm->arch.hpt)) {
576 				/* parameter error */
577 				args[j] = ((0xa0 | flags) << 56) + pte_index;
578 				ret = H_PARAMETER;
579 				break;
580 			}
581 			hp = (__be64 *) (kvm->arch.hpt.virt + (pte_index << 4));
582 			/* to avoid deadlock, don't spin except for first */
583 			if (!try_lock_hpte(hp, HPTE_V_HVLOCK)) {
584 				if (n)
585 					break;
586 				while (!try_lock_hpte(hp, HPTE_V_HVLOCK))
587 					cpu_relax();
588 			}
589 			found = 0;
590 			hp0 = be64_to_cpu(hp[0]);
591 			hp1 = be64_to_cpu(hp[1]);
592 			if (cpu_has_feature(CPU_FTR_ARCH_300)) {
593 				hp0 = hpte_new_to_old_v(hp0, hp1);
594 				hp1 = hpte_new_to_old_r(hp1);
595 			}
596 			if (hp0 & (HPTE_V_ABSENT | HPTE_V_VALID)) {
597 				switch (flags & 3) {
598 				case 0:		/* absolute */
599 					found = 1;
600 					break;
601 				case 1:		/* andcond */
602 					if (!(hp0 & args[j + 1]))
603 						found = 1;
604 					break;
605 				case 2:		/* AVPN */
606 					if ((hp0 & ~0x7fUL) == args[j + 1])
607 						found = 1;
608 					break;
609 				}
610 			}
611 			if (!found) {
612 				hp[0] &= ~cpu_to_be64(HPTE_V_HVLOCK);
613 				args[j] = ((0x90 | flags) << 56) + pte_index;
614 				continue;
615 			}
616 
617 			args[j] = ((0x80 | flags) << 56) + pte_index;
618 			rev = real_vmalloc_addr(&kvm->arch.hpt.rev[pte_index]);
619 			note_hpte_modification(kvm, rev);
620 
621 			if (!(hp0 & HPTE_V_VALID)) {
622 				/* insert R and C bits from PTE */
623 				rcbits = rev->guest_rpte & (HPTE_R_R|HPTE_R_C);
624 				args[j] |= rcbits << (56 - 5);
625 				hp[0] = 0;
626 				if (is_mmio_hpte(hp0, hp1))
627 					atomic64_inc(&kvm->arch.mmio_update);
628 				continue;
629 			}
630 
631 			/* leave it locked */
632 			hp[0] &= ~cpu_to_be64(HPTE_V_VALID);
633 			tlbrb[n] = compute_tlbie_rb(hp0, hp1, pte_index);
634 			indexes[n] = j;
635 			hptes[n] = hp;
636 			revs[n] = rev;
637 			++n;
638 		}
639 
640 		if (!n)
641 			break;
642 
643 		/* Now that we've collected a batch, do the tlbies */
644 		do_tlbies(kvm, tlbrb, n, global, true);
645 
646 		/* Read PTE low words after tlbie to get final R/C values */
647 		for (k = 0; k < n; ++k) {
648 			j = indexes[k];
649 			pte_index = args[j] & ((1ul << 56) - 1);
650 			hp = hptes[k];
651 			rev = revs[k];
652 			remove_revmap_chain(kvm, pte_index, rev,
653 				be64_to_cpu(hp[0]), be64_to_cpu(hp[1]));
654 			rcbits = rev->guest_rpte & (HPTE_R_R|HPTE_R_C);
655 			args[j] |= rcbits << (56 - 5);
656 			__unlock_hpte(hp, 0);
657 		}
658 	}
659 
660 	return ret;
661 }
662 
663 long kvmppc_h_protect(struct kvm_vcpu *vcpu, unsigned long flags,
664 		      unsigned long pte_index, unsigned long avpn,
665 		      unsigned long va)
666 {
667 	struct kvm *kvm = vcpu->kvm;
668 	__be64 *hpte;
669 	struct revmap_entry *rev;
670 	unsigned long v, r, rb, mask, bits;
671 	u64 pte_v, pte_r;
672 
673 	if (kvm_is_radix(kvm))
674 		return H_FUNCTION;
675 	if (pte_index >= kvmppc_hpt_npte(&kvm->arch.hpt))
676 		return H_PARAMETER;
677 
678 	hpte = (__be64 *)(kvm->arch.hpt.virt + (pte_index << 4));
679 	while (!try_lock_hpte(hpte, HPTE_V_HVLOCK))
680 		cpu_relax();
681 	v = pte_v = be64_to_cpu(hpte[0]);
682 	if (cpu_has_feature(CPU_FTR_ARCH_300))
683 		v = hpte_new_to_old_v(v, be64_to_cpu(hpte[1]));
684 	if ((v & (HPTE_V_ABSENT | HPTE_V_VALID)) == 0 ||
685 	    ((flags & H_AVPN) && (v & ~0x7fUL) != avpn)) {
686 		__unlock_hpte(hpte, pte_v);
687 		return H_NOT_FOUND;
688 	}
689 
690 	pte_r = be64_to_cpu(hpte[1]);
691 	bits = (flags << 55) & HPTE_R_PP0;
692 	bits |= (flags << 48) & HPTE_R_KEY_HI;
693 	bits |= flags & (HPTE_R_PP | HPTE_R_N | HPTE_R_KEY_LO);
694 
695 	/* Update guest view of 2nd HPTE dword */
696 	mask = HPTE_R_PP0 | HPTE_R_PP | HPTE_R_N |
697 		HPTE_R_KEY_HI | HPTE_R_KEY_LO;
698 	rev = real_vmalloc_addr(&kvm->arch.hpt.rev[pte_index]);
699 	if (rev) {
700 		r = (rev->guest_rpte & ~mask) | bits;
701 		rev->guest_rpte = r;
702 		note_hpte_modification(kvm, rev);
703 	}
704 
705 	/* Update HPTE */
706 	if (v & HPTE_V_VALID) {
707 		/*
708 		 * If the page is valid, don't let it transition from
709 		 * readonly to writable.  If it should be writable, we'll
710 		 * take a trap and let the page fault code sort it out.
711 		 */
712 		r = (pte_r & ~mask) | bits;
713 		if (hpte_is_writable(r) && !hpte_is_writable(pte_r))
714 			r = hpte_make_readonly(r);
715 		/* If the PTE is changing, invalidate it first */
716 		if (r != pte_r) {
717 			rb = compute_tlbie_rb(v, r, pte_index);
718 			hpte[0] = cpu_to_be64((pte_v & ~HPTE_V_VALID) |
719 					      HPTE_V_ABSENT);
720 			do_tlbies(kvm, &rb, 1, global_invalidates(kvm), true);
721 			/* Don't lose R/C bit updates done by hardware */
722 			r |= be64_to_cpu(hpte[1]) & (HPTE_R_R | HPTE_R_C);
723 			hpte[1] = cpu_to_be64(r);
724 		}
725 	}
726 	unlock_hpte(hpte, pte_v & ~HPTE_V_HVLOCK);
727 	asm volatile("ptesync" : : : "memory");
728 	if (is_mmio_hpte(v, pte_r))
729 		atomic64_inc(&kvm->arch.mmio_update);
730 
731 	return H_SUCCESS;
732 }
733 
734 long kvmppc_h_read(struct kvm_vcpu *vcpu, unsigned long flags,
735 		   unsigned long pte_index)
736 {
737 	struct kvm *kvm = vcpu->kvm;
738 	__be64 *hpte;
739 	unsigned long v, r;
740 	int i, n = 1;
741 	struct revmap_entry *rev = NULL;
742 
743 	if (kvm_is_radix(kvm))
744 		return H_FUNCTION;
745 	if (pte_index >= kvmppc_hpt_npte(&kvm->arch.hpt))
746 		return H_PARAMETER;
747 	if (flags & H_READ_4) {
748 		pte_index &= ~3;
749 		n = 4;
750 	}
751 	rev = real_vmalloc_addr(&kvm->arch.hpt.rev[pte_index]);
752 	for (i = 0; i < n; ++i, ++pte_index) {
753 		hpte = (__be64 *)(kvm->arch.hpt.virt + (pte_index << 4));
754 		v = be64_to_cpu(hpte[0]) & ~HPTE_V_HVLOCK;
755 		r = be64_to_cpu(hpte[1]);
756 		if (cpu_has_feature(CPU_FTR_ARCH_300)) {
757 			v = hpte_new_to_old_v(v, r);
758 			r = hpte_new_to_old_r(r);
759 		}
760 		if (v & HPTE_V_ABSENT) {
761 			v &= ~HPTE_V_ABSENT;
762 			v |= HPTE_V_VALID;
763 		}
764 		if (v & HPTE_V_VALID) {
765 			r = rev[i].guest_rpte | (r & (HPTE_R_R | HPTE_R_C));
766 			r &= ~HPTE_GR_RESERVED;
767 		}
768 		vcpu->arch.regs.gpr[4 + i * 2] = v;
769 		vcpu->arch.regs.gpr[5 + i * 2] = r;
770 	}
771 	return H_SUCCESS;
772 }
773 
774 long kvmppc_h_clear_ref(struct kvm_vcpu *vcpu, unsigned long flags,
775 			unsigned long pte_index)
776 {
777 	struct kvm *kvm = vcpu->kvm;
778 	__be64 *hpte;
779 	unsigned long v, r, gr;
780 	struct revmap_entry *rev;
781 	unsigned long *rmap;
782 	long ret = H_NOT_FOUND;
783 
784 	if (kvm_is_radix(kvm))
785 		return H_FUNCTION;
786 	if (pte_index >= kvmppc_hpt_npte(&kvm->arch.hpt))
787 		return H_PARAMETER;
788 
789 	rev = real_vmalloc_addr(&kvm->arch.hpt.rev[pte_index]);
790 	hpte = (__be64 *)(kvm->arch.hpt.virt + (pte_index << 4));
791 	while (!try_lock_hpte(hpte, HPTE_V_HVLOCK))
792 		cpu_relax();
793 	v = be64_to_cpu(hpte[0]);
794 	r = be64_to_cpu(hpte[1]);
795 	if (!(v & (HPTE_V_VALID | HPTE_V_ABSENT)))
796 		goto out;
797 
798 	gr = rev->guest_rpte;
799 	if (rev->guest_rpte & HPTE_R_R) {
800 		rev->guest_rpte &= ~HPTE_R_R;
801 		note_hpte_modification(kvm, rev);
802 	}
803 	if (v & HPTE_V_VALID) {
804 		gr |= r & (HPTE_R_R | HPTE_R_C);
805 		if (r & HPTE_R_R) {
806 			kvmppc_clear_ref_hpte(kvm, hpte, pte_index);
807 			rmap = revmap_for_hpte(kvm, v, gr, NULL, NULL);
808 			if (rmap) {
809 				lock_rmap(rmap);
810 				*rmap |= KVMPPC_RMAP_REFERENCED;
811 				unlock_rmap(rmap);
812 			}
813 		}
814 	}
815 	vcpu->arch.regs.gpr[4] = gr;
816 	ret = H_SUCCESS;
817  out:
818 	unlock_hpte(hpte, v & ~HPTE_V_HVLOCK);
819 	return ret;
820 }
821 
822 long kvmppc_h_clear_mod(struct kvm_vcpu *vcpu, unsigned long flags,
823 			unsigned long pte_index)
824 {
825 	struct kvm *kvm = vcpu->kvm;
826 	__be64 *hpte;
827 	unsigned long v, r, gr;
828 	struct revmap_entry *rev;
829 	long ret = H_NOT_FOUND;
830 
831 	if (kvm_is_radix(kvm))
832 		return H_FUNCTION;
833 	if (pte_index >= kvmppc_hpt_npte(&kvm->arch.hpt))
834 		return H_PARAMETER;
835 
836 	rev = real_vmalloc_addr(&kvm->arch.hpt.rev[pte_index]);
837 	hpte = (__be64 *)(kvm->arch.hpt.virt + (pte_index << 4));
838 	while (!try_lock_hpte(hpte, HPTE_V_HVLOCK))
839 		cpu_relax();
840 	v = be64_to_cpu(hpte[0]);
841 	r = be64_to_cpu(hpte[1]);
842 	if (!(v & (HPTE_V_VALID | HPTE_V_ABSENT)))
843 		goto out;
844 
845 	gr = rev->guest_rpte;
846 	if (gr & HPTE_R_C) {
847 		rev->guest_rpte &= ~HPTE_R_C;
848 		note_hpte_modification(kvm, rev);
849 	}
850 	if (v & HPTE_V_VALID) {
851 		/* need to make it temporarily absent so C is stable */
852 		hpte[0] |= cpu_to_be64(HPTE_V_ABSENT);
853 		kvmppc_invalidate_hpte(kvm, hpte, pte_index);
854 		r = be64_to_cpu(hpte[1]);
855 		gr |= r & (HPTE_R_R | HPTE_R_C);
856 		if (r & HPTE_R_C) {
857 			hpte[1] = cpu_to_be64(r & ~HPTE_R_C);
858 			eieio();
859 			kvmppc_set_dirty_from_hpte(kvm, v, gr);
860 		}
861 	}
862 	vcpu->arch.regs.gpr[4] = gr;
863 	ret = H_SUCCESS;
864  out:
865 	unlock_hpte(hpte, v & ~HPTE_V_HVLOCK);
866 	return ret;
867 }
868 
869 static int kvmppc_get_hpa(struct kvm_vcpu *vcpu, unsigned long gpa,
870 			  int writing, unsigned long *hpa,
871 			  struct kvm_memory_slot **memslot_p)
872 {
873 	struct kvm *kvm = vcpu->kvm;
874 	struct kvm_memory_slot *memslot;
875 	unsigned long gfn, hva, pa, psize = PAGE_SHIFT;
876 	unsigned int shift;
877 	pte_t *ptep, pte;
878 
879 	/* Find the memslot for this address */
880 	gfn = gpa >> PAGE_SHIFT;
881 	memslot = __gfn_to_memslot(kvm_memslots_raw(kvm), gfn);
882 	if (!memslot || (memslot->flags & KVM_MEMSLOT_INVALID))
883 		return H_PARAMETER;
884 
885 	/* Translate to host virtual address */
886 	hva = __gfn_to_hva_memslot(memslot, gfn);
887 
888 	/* Try to find the host pte for that virtual address */
889 	ptep = __find_linux_pte(vcpu->arch.pgdir, hva, NULL, &shift);
890 	if (!ptep)
891 		return H_TOO_HARD;
892 	pte = kvmppc_read_update_linux_pte(ptep, writing);
893 	if (!pte_present(pte))
894 		return H_TOO_HARD;
895 
896 	/* Convert to a physical address */
897 	if (shift)
898 		psize = 1UL << shift;
899 	pa = pte_pfn(pte) << PAGE_SHIFT;
900 	pa |= hva & (psize - 1);
901 	pa |= gpa & ~PAGE_MASK;
902 
903 	if (hpa)
904 		*hpa = pa;
905 	if (memslot_p)
906 		*memslot_p = memslot;
907 
908 	return H_SUCCESS;
909 }
910 
911 static long kvmppc_do_h_page_init_zero(struct kvm_vcpu *vcpu,
912 				       unsigned long dest)
913 {
914 	struct kvm_memory_slot *memslot;
915 	struct kvm *kvm = vcpu->kvm;
916 	unsigned long pa, mmu_seq;
917 	long ret = H_SUCCESS;
918 	int i;
919 
920 	/* Used later to detect if we might have been invalidated */
921 	mmu_seq = kvm->mmu_notifier_seq;
922 	smp_rmb();
923 
924 	ret = kvmppc_get_hpa(vcpu, dest, 1, &pa, &memslot);
925 	if (ret != H_SUCCESS)
926 		return ret;
927 
928 	/* Check if we've been invalidated */
929 	raw_spin_lock(&kvm->mmu_lock.rlock);
930 	if (mmu_notifier_retry(kvm, mmu_seq)) {
931 		ret = H_TOO_HARD;
932 		goto out_unlock;
933 	}
934 
935 	/* Zero the page */
936 	for (i = 0; i < SZ_4K; i += L1_CACHE_BYTES, pa += L1_CACHE_BYTES)
937 		dcbz((void *)pa);
938 	kvmppc_update_dirty_map(memslot, dest >> PAGE_SHIFT, PAGE_SIZE);
939 
940 out_unlock:
941 	raw_spin_unlock(&kvm->mmu_lock.rlock);
942 	return ret;
943 }
944 
945 static long kvmppc_do_h_page_init_copy(struct kvm_vcpu *vcpu,
946 				       unsigned long dest, unsigned long src)
947 {
948 	unsigned long dest_pa, src_pa, mmu_seq;
949 	struct kvm_memory_slot *dest_memslot;
950 	struct kvm *kvm = vcpu->kvm;
951 	long ret = H_SUCCESS;
952 
953 	/* Used later to detect if we might have been invalidated */
954 	mmu_seq = kvm->mmu_notifier_seq;
955 	smp_rmb();
956 
957 	ret = kvmppc_get_hpa(vcpu, dest, 1, &dest_pa, &dest_memslot);
958 	if (ret != H_SUCCESS)
959 		return ret;
960 	ret = kvmppc_get_hpa(vcpu, src, 0, &src_pa, NULL);
961 	if (ret != H_SUCCESS)
962 		return ret;
963 
964 	/* Check if we've been invalidated */
965 	raw_spin_lock(&kvm->mmu_lock.rlock);
966 	if (mmu_notifier_retry(kvm, mmu_seq)) {
967 		ret = H_TOO_HARD;
968 		goto out_unlock;
969 	}
970 
971 	/* Copy the page */
972 	memcpy((void *)dest_pa, (void *)src_pa, SZ_4K);
973 
974 	kvmppc_update_dirty_map(dest_memslot, dest >> PAGE_SHIFT, PAGE_SIZE);
975 
976 out_unlock:
977 	raw_spin_unlock(&kvm->mmu_lock.rlock);
978 	return ret;
979 }
980 
981 long kvmppc_rm_h_page_init(struct kvm_vcpu *vcpu, unsigned long flags,
982 			   unsigned long dest, unsigned long src)
983 {
984 	struct kvm *kvm = vcpu->kvm;
985 	u64 pg_mask = SZ_4K - 1;	/* 4K page size */
986 	long ret = H_SUCCESS;
987 
988 	/* Don't handle radix mode here, go up to the virtual mode handler */
989 	if (kvm_is_radix(kvm))
990 		return H_TOO_HARD;
991 
992 	/* Check for invalid flags (H_PAGE_SET_LOANED covers all CMO flags) */
993 	if (flags & ~(H_ICACHE_INVALIDATE | H_ICACHE_SYNCHRONIZE |
994 		      H_ZERO_PAGE | H_COPY_PAGE | H_PAGE_SET_LOANED))
995 		return H_PARAMETER;
996 
997 	/* dest (and src if copy_page flag set) must be page aligned */
998 	if ((dest & pg_mask) || ((flags & H_COPY_PAGE) && (src & pg_mask)))
999 		return H_PARAMETER;
1000 
1001 	/* zero and/or copy the page as determined by the flags */
1002 	if (flags & H_COPY_PAGE)
1003 		ret = kvmppc_do_h_page_init_copy(vcpu, dest, src);
1004 	else if (flags & H_ZERO_PAGE)
1005 		ret = kvmppc_do_h_page_init_zero(vcpu, dest);
1006 
1007 	/* We can ignore the other flags */
1008 
1009 	return ret;
1010 }
1011 
1012 void kvmppc_invalidate_hpte(struct kvm *kvm, __be64 *hptep,
1013 			unsigned long pte_index)
1014 {
1015 	unsigned long rb;
1016 	u64 hp0, hp1;
1017 
1018 	hptep[0] &= ~cpu_to_be64(HPTE_V_VALID);
1019 	hp0 = be64_to_cpu(hptep[0]);
1020 	hp1 = be64_to_cpu(hptep[1]);
1021 	if (cpu_has_feature(CPU_FTR_ARCH_300)) {
1022 		hp0 = hpte_new_to_old_v(hp0, hp1);
1023 		hp1 = hpte_new_to_old_r(hp1);
1024 	}
1025 	rb = compute_tlbie_rb(hp0, hp1, pte_index);
1026 	do_tlbies(kvm, &rb, 1, 1, true);
1027 }
1028 EXPORT_SYMBOL_GPL(kvmppc_invalidate_hpte);
1029 
1030 void kvmppc_clear_ref_hpte(struct kvm *kvm, __be64 *hptep,
1031 			   unsigned long pte_index)
1032 {
1033 	unsigned long rb;
1034 	unsigned char rbyte;
1035 	u64 hp0, hp1;
1036 
1037 	hp0 = be64_to_cpu(hptep[0]);
1038 	hp1 = be64_to_cpu(hptep[1]);
1039 	if (cpu_has_feature(CPU_FTR_ARCH_300)) {
1040 		hp0 = hpte_new_to_old_v(hp0, hp1);
1041 		hp1 = hpte_new_to_old_r(hp1);
1042 	}
1043 	rb = compute_tlbie_rb(hp0, hp1, pte_index);
1044 	rbyte = (be64_to_cpu(hptep[1]) & ~HPTE_R_R) >> 8;
1045 	/* modify only the second-last byte, which contains the ref bit */
1046 	*((char *)hptep + 14) = rbyte;
1047 	do_tlbies(kvm, &rb, 1, 1, false);
1048 }
1049 EXPORT_SYMBOL_GPL(kvmppc_clear_ref_hpte);
1050 
1051 static int slb_base_page_shift[4] = {
1052 	24,	/* 16M */
1053 	16,	/* 64k */
1054 	34,	/* 16G */
1055 	20,	/* 1M, unsupported */
1056 };
1057 
1058 static struct mmio_hpte_cache_entry *mmio_cache_search(struct kvm_vcpu *vcpu,
1059 		unsigned long eaddr, unsigned long slb_v, long mmio_update)
1060 {
1061 	struct mmio_hpte_cache_entry *entry = NULL;
1062 	unsigned int pshift;
1063 	unsigned int i;
1064 
1065 	for (i = 0; i < MMIO_HPTE_CACHE_SIZE; i++) {
1066 		entry = &vcpu->arch.mmio_cache.entry[i];
1067 		if (entry->mmio_update == mmio_update) {
1068 			pshift = entry->slb_base_pshift;
1069 			if ((entry->eaddr >> pshift) == (eaddr >> pshift) &&
1070 			    entry->slb_v == slb_v)
1071 				return entry;
1072 		}
1073 	}
1074 	return NULL;
1075 }
1076 
1077 static struct mmio_hpte_cache_entry *
1078 			next_mmio_cache_entry(struct kvm_vcpu *vcpu)
1079 {
1080 	unsigned int index = vcpu->arch.mmio_cache.index;
1081 
1082 	vcpu->arch.mmio_cache.index++;
1083 	if (vcpu->arch.mmio_cache.index == MMIO_HPTE_CACHE_SIZE)
1084 		vcpu->arch.mmio_cache.index = 0;
1085 
1086 	return &vcpu->arch.mmio_cache.entry[index];
1087 }
1088 
1089 /* When called from virtmode, this func should be protected by
1090  * preempt_disable(), otherwise, the holding of HPTE_V_HVLOCK
1091  * can trigger deadlock issue.
1092  */
1093 long kvmppc_hv_find_lock_hpte(struct kvm *kvm, gva_t eaddr, unsigned long slb_v,
1094 			      unsigned long valid)
1095 {
1096 	unsigned int i;
1097 	unsigned int pshift;
1098 	unsigned long somask;
1099 	unsigned long vsid, hash;
1100 	unsigned long avpn;
1101 	__be64 *hpte;
1102 	unsigned long mask, val;
1103 	unsigned long v, r, orig_v;
1104 
1105 	/* Get page shift, work out hash and AVPN etc. */
1106 	mask = SLB_VSID_B | HPTE_V_AVPN | HPTE_V_SECONDARY;
1107 	val = 0;
1108 	pshift = 12;
1109 	if (slb_v & SLB_VSID_L) {
1110 		mask |= HPTE_V_LARGE;
1111 		val |= HPTE_V_LARGE;
1112 		pshift = slb_base_page_shift[(slb_v & SLB_VSID_LP) >> 4];
1113 	}
1114 	if (slb_v & SLB_VSID_B_1T) {
1115 		somask = (1UL << 40) - 1;
1116 		vsid = (slb_v & ~SLB_VSID_B) >> SLB_VSID_SHIFT_1T;
1117 		vsid ^= vsid << 25;
1118 	} else {
1119 		somask = (1UL << 28) - 1;
1120 		vsid = (slb_v & ~SLB_VSID_B) >> SLB_VSID_SHIFT;
1121 	}
1122 	hash = (vsid ^ ((eaddr & somask) >> pshift)) & kvmppc_hpt_mask(&kvm->arch.hpt);
1123 	avpn = slb_v & ~(somask >> 16);	/* also includes B */
1124 	avpn |= (eaddr & somask) >> 16;
1125 
1126 	if (pshift >= 24)
1127 		avpn &= ~((1UL << (pshift - 16)) - 1);
1128 	else
1129 		avpn &= ~0x7fUL;
1130 	val |= avpn;
1131 
1132 	for (;;) {
1133 		hpte = (__be64 *)(kvm->arch.hpt.virt + (hash << 7));
1134 
1135 		for (i = 0; i < 16; i += 2) {
1136 			/* Read the PTE racily */
1137 			v = be64_to_cpu(hpte[i]) & ~HPTE_V_HVLOCK;
1138 			if (cpu_has_feature(CPU_FTR_ARCH_300))
1139 				v = hpte_new_to_old_v(v, be64_to_cpu(hpte[i+1]));
1140 
1141 			/* Check valid/absent, hash, segment size and AVPN */
1142 			if (!(v & valid) || (v & mask) != val)
1143 				continue;
1144 
1145 			/* Lock the PTE and read it under the lock */
1146 			while (!try_lock_hpte(&hpte[i], HPTE_V_HVLOCK))
1147 				cpu_relax();
1148 			v = orig_v = be64_to_cpu(hpte[i]) & ~HPTE_V_HVLOCK;
1149 			r = be64_to_cpu(hpte[i+1]);
1150 			if (cpu_has_feature(CPU_FTR_ARCH_300)) {
1151 				v = hpte_new_to_old_v(v, r);
1152 				r = hpte_new_to_old_r(r);
1153 			}
1154 
1155 			/*
1156 			 * Check the HPTE again, including base page size
1157 			 */
1158 			if ((v & valid) && (v & mask) == val &&
1159 			    kvmppc_hpte_base_page_shift(v, r) == pshift)
1160 				/* Return with the HPTE still locked */
1161 				return (hash << 3) + (i >> 1);
1162 
1163 			__unlock_hpte(&hpte[i], orig_v);
1164 		}
1165 
1166 		if (val & HPTE_V_SECONDARY)
1167 			break;
1168 		val |= HPTE_V_SECONDARY;
1169 		hash = hash ^ kvmppc_hpt_mask(&kvm->arch.hpt);
1170 	}
1171 	return -1;
1172 }
1173 EXPORT_SYMBOL(kvmppc_hv_find_lock_hpte);
1174 
1175 /*
1176  * Called in real mode to check whether an HPTE not found fault
1177  * is due to accessing a paged-out page or an emulated MMIO page,
1178  * or if a protection fault is due to accessing a page that the
1179  * guest wanted read/write access to but which we made read-only.
1180  * Returns a possibly modified status (DSISR) value if not
1181  * (i.e. pass the interrupt to the guest),
1182  * -1 to pass the fault up to host kernel mode code, -2 to do that
1183  * and also load the instruction word (for MMIO emulation),
1184  * or 0 if we should make the guest retry the access.
1185  */
1186 long kvmppc_hpte_hv_fault(struct kvm_vcpu *vcpu, unsigned long addr,
1187 			  unsigned long slb_v, unsigned int status, bool data)
1188 {
1189 	struct kvm *kvm = vcpu->kvm;
1190 	long int index;
1191 	unsigned long v, r, gr, orig_v;
1192 	__be64 *hpte;
1193 	unsigned long valid;
1194 	struct revmap_entry *rev;
1195 	unsigned long pp, key;
1196 	struct mmio_hpte_cache_entry *cache_entry = NULL;
1197 	long mmio_update = 0;
1198 
1199 	/* For protection fault, expect to find a valid HPTE */
1200 	valid = HPTE_V_VALID;
1201 	if (status & DSISR_NOHPTE) {
1202 		valid |= HPTE_V_ABSENT;
1203 		mmio_update = atomic64_read(&kvm->arch.mmio_update);
1204 		cache_entry = mmio_cache_search(vcpu, addr, slb_v, mmio_update);
1205 	}
1206 	if (cache_entry) {
1207 		index = cache_entry->pte_index;
1208 		v = cache_entry->hpte_v;
1209 		r = cache_entry->hpte_r;
1210 		gr = cache_entry->rpte;
1211 	} else {
1212 		index = kvmppc_hv_find_lock_hpte(kvm, addr, slb_v, valid);
1213 		if (index < 0) {
1214 			if (status & DSISR_NOHPTE)
1215 				return status;	/* there really was no HPTE */
1216 			return 0;	/* for prot fault, HPTE disappeared */
1217 		}
1218 		hpte = (__be64 *)(kvm->arch.hpt.virt + (index << 4));
1219 		v = orig_v = be64_to_cpu(hpte[0]) & ~HPTE_V_HVLOCK;
1220 		r = be64_to_cpu(hpte[1]);
1221 		if (cpu_has_feature(CPU_FTR_ARCH_300)) {
1222 			v = hpte_new_to_old_v(v, r);
1223 			r = hpte_new_to_old_r(r);
1224 		}
1225 		rev = real_vmalloc_addr(&kvm->arch.hpt.rev[index]);
1226 		gr = rev->guest_rpte;
1227 
1228 		unlock_hpte(hpte, orig_v);
1229 	}
1230 
1231 	/* For not found, if the HPTE is valid by now, retry the instruction */
1232 	if ((status & DSISR_NOHPTE) && (v & HPTE_V_VALID))
1233 		return 0;
1234 
1235 	/* Check access permissions to the page */
1236 	pp = gr & (HPTE_R_PP0 | HPTE_R_PP);
1237 	key = (vcpu->arch.shregs.msr & MSR_PR) ? SLB_VSID_KP : SLB_VSID_KS;
1238 	status &= ~DSISR_NOHPTE;	/* DSISR_NOHPTE == SRR1_ISI_NOPT */
1239 	if (!data) {
1240 		if (gr & (HPTE_R_N | HPTE_R_G))
1241 			return status | SRR1_ISI_N_OR_G;
1242 		if (!hpte_read_permission(pp, slb_v & key))
1243 			return status | SRR1_ISI_PROT;
1244 	} else if (status & DSISR_ISSTORE) {
1245 		/* check write permission */
1246 		if (!hpte_write_permission(pp, slb_v & key))
1247 			return status | DSISR_PROTFAULT;
1248 	} else {
1249 		if (!hpte_read_permission(pp, slb_v & key))
1250 			return status | DSISR_PROTFAULT;
1251 	}
1252 
1253 	/* Check storage key, if applicable */
1254 	if (data && (vcpu->arch.shregs.msr & MSR_DR)) {
1255 		unsigned int perm = hpte_get_skey_perm(gr, vcpu->arch.amr);
1256 		if (status & DSISR_ISSTORE)
1257 			perm >>= 1;
1258 		if (perm & 1)
1259 			return status | DSISR_KEYFAULT;
1260 	}
1261 
1262 	/* Save HPTE info for virtual-mode handler */
1263 	vcpu->arch.pgfault_addr = addr;
1264 	vcpu->arch.pgfault_index = index;
1265 	vcpu->arch.pgfault_hpte[0] = v;
1266 	vcpu->arch.pgfault_hpte[1] = r;
1267 	vcpu->arch.pgfault_cache = cache_entry;
1268 
1269 	/* Check the storage key to see if it is possibly emulated MMIO */
1270 	if ((r & (HPTE_R_KEY_HI | HPTE_R_KEY_LO)) ==
1271 	    (HPTE_R_KEY_HI | HPTE_R_KEY_LO)) {
1272 		if (!cache_entry) {
1273 			unsigned int pshift = 12;
1274 			unsigned int pshift_index;
1275 
1276 			if (slb_v & SLB_VSID_L) {
1277 				pshift_index = ((slb_v & SLB_VSID_LP) >> 4);
1278 				pshift = slb_base_page_shift[pshift_index];
1279 			}
1280 			cache_entry = next_mmio_cache_entry(vcpu);
1281 			cache_entry->eaddr = addr;
1282 			cache_entry->slb_base_pshift = pshift;
1283 			cache_entry->pte_index = index;
1284 			cache_entry->hpte_v = v;
1285 			cache_entry->hpte_r = r;
1286 			cache_entry->rpte = gr;
1287 			cache_entry->slb_v = slb_v;
1288 			cache_entry->mmio_update = mmio_update;
1289 		}
1290 		if (data && (vcpu->arch.shregs.msr & MSR_IR))
1291 			return -2;	/* MMIO emulation - load instr word */
1292 	}
1293 
1294 	return -1;		/* send fault up to host kernel mode */
1295 }
1296