1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  *
4  * Copyright 2016 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/anon_inodes.h>
12 #include <linux/file.h>
13 #include <linux/debugfs.h>
14 
15 #include <asm/kvm_ppc.h>
16 #include <asm/kvm_book3s.h>
17 #include <asm/page.h>
18 #include <asm/mmu.h>
19 #include <asm/pgtable.h>
20 #include <asm/pgalloc.h>
21 #include <asm/pte-walk.h>
22 #include <asm/ultravisor.h>
23 #include <asm/kvm_book3s_uvmem.h>
24 
25 /*
26  * Supported radix tree geometry.
27  * Like p9, we support either 5 or 9 bits at the first (lowest) level,
28  * for a page size of 64k or 4k.
29  */
30 static int p9_supported_radix_bits[4] = { 5, 9, 9, 13 };
31 
32 unsigned long __kvmhv_copy_tofrom_guest_radix(int lpid, int pid,
33 					      gva_t eaddr, void *to, void *from,
34 					      unsigned long n)
35 {
36 	int uninitialized_var(old_pid), old_lpid;
37 	unsigned long quadrant, ret = n;
38 	bool is_load = !!to;
39 
40 	/* Can't access quadrants 1 or 2 in non-HV mode, call the HV to do it */
41 	if (kvmhv_on_pseries())
42 		return plpar_hcall_norets(H_COPY_TOFROM_GUEST, lpid, pid, eaddr,
43 					  __pa(to), __pa(from), n);
44 
45 	quadrant = 1;
46 	if (!pid)
47 		quadrant = 2;
48 	if (is_load)
49 		from = (void *) (eaddr | (quadrant << 62));
50 	else
51 		to = (void *) (eaddr | (quadrant << 62));
52 
53 	preempt_disable();
54 
55 	/* switch the lpid first to avoid running host with unallocated pid */
56 	old_lpid = mfspr(SPRN_LPID);
57 	if (old_lpid != lpid)
58 		mtspr(SPRN_LPID, lpid);
59 	if (quadrant == 1) {
60 		old_pid = mfspr(SPRN_PID);
61 		if (old_pid != pid)
62 			mtspr(SPRN_PID, pid);
63 	}
64 	isync();
65 
66 	if (is_load)
67 		ret = probe_user_read(to, (const void __user *)from, n);
68 	else
69 		ret = probe_user_write((void __user *)to, from, n);
70 
71 	/* switch the pid first to avoid running host with unallocated pid */
72 	if (quadrant == 1 && pid != old_pid)
73 		mtspr(SPRN_PID, old_pid);
74 	if (lpid != old_lpid)
75 		mtspr(SPRN_LPID, old_lpid);
76 	isync();
77 
78 	preempt_enable();
79 
80 	return ret;
81 }
82 EXPORT_SYMBOL_GPL(__kvmhv_copy_tofrom_guest_radix);
83 
84 static long kvmhv_copy_tofrom_guest_radix(struct kvm_vcpu *vcpu, gva_t eaddr,
85 					  void *to, void *from, unsigned long n)
86 {
87 	int lpid = vcpu->kvm->arch.lpid;
88 	int pid = vcpu->arch.pid;
89 
90 	/* This would cause a data segment intr so don't allow the access */
91 	if (eaddr & (0x3FFUL << 52))
92 		return -EINVAL;
93 
94 	/* Should we be using the nested lpid */
95 	if (vcpu->arch.nested)
96 		lpid = vcpu->arch.nested->shadow_lpid;
97 
98 	/* If accessing quadrant 3 then pid is expected to be 0 */
99 	if (((eaddr >> 62) & 0x3) == 0x3)
100 		pid = 0;
101 
102 	eaddr &= ~(0xFFFUL << 52);
103 
104 	return __kvmhv_copy_tofrom_guest_radix(lpid, pid, eaddr, to, from, n);
105 }
106 
107 long kvmhv_copy_from_guest_radix(struct kvm_vcpu *vcpu, gva_t eaddr, void *to,
108 				 unsigned long n)
109 {
110 	long ret;
111 
112 	ret = kvmhv_copy_tofrom_guest_radix(vcpu, eaddr, to, NULL, n);
113 	if (ret > 0)
114 		memset(to + (n - ret), 0, ret);
115 
116 	return ret;
117 }
118 EXPORT_SYMBOL_GPL(kvmhv_copy_from_guest_radix);
119 
120 long kvmhv_copy_to_guest_radix(struct kvm_vcpu *vcpu, gva_t eaddr, void *from,
121 			       unsigned long n)
122 {
123 	return kvmhv_copy_tofrom_guest_radix(vcpu, eaddr, NULL, from, n);
124 }
125 EXPORT_SYMBOL_GPL(kvmhv_copy_to_guest_radix);
126 
127 int kvmppc_mmu_walk_radix_tree(struct kvm_vcpu *vcpu, gva_t eaddr,
128 			       struct kvmppc_pte *gpte, u64 root,
129 			       u64 *pte_ret_p)
130 {
131 	struct kvm *kvm = vcpu->kvm;
132 	int ret, level, ps;
133 	unsigned long rts, bits, offset, index;
134 	u64 pte, base, gpa;
135 	__be64 rpte;
136 
137 	rts = ((root & RTS1_MASK) >> (RTS1_SHIFT - 3)) |
138 		((root & RTS2_MASK) >> RTS2_SHIFT);
139 	bits = root & RPDS_MASK;
140 	base = root & RPDB_MASK;
141 
142 	offset = rts + 31;
143 
144 	/* Current implementations only support 52-bit space */
145 	if (offset != 52)
146 		return -EINVAL;
147 
148 	/* Walk each level of the radix tree */
149 	for (level = 3; level >= 0; --level) {
150 		u64 addr;
151 		/* Check a valid size */
152 		if (level && bits != p9_supported_radix_bits[level])
153 			return -EINVAL;
154 		if (level == 0 && !(bits == 5 || bits == 9))
155 			return -EINVAL;
156 		offset -= bits;
157 		index = (eaddr >> offset) & ((1UL << bits) - 1);
158 		/* Check that low bits of page table base are zero */
159 		if (base & ((1UL << (bits + 3)) - 1))
160 			return -EINVAL;
161 		/* Read the entry from guest memory */
162 		addr = base + (index * sizeof(rpte));
163 		ret = kvm_read_guest(kvm, addr, &rpte, sizeof(rpte));
164 		if (ret) {
165 			if (pte_ret_p)
166 				*pte_ret_p = addr;
167 			return ret;
168 		}
169 		pte = __be64_to_cpu(rpte);
170 		if (!(pte & _PAGE_PRESENT))
171 			return -ENOENT;
172 		/* Check if a leaf entry */
173 		if (pte & _PAGE_PTE)
174 			break;
175 		/* Get ready to walk the next level */
176 		base = pte & RPDB_MASK;
177 		bits = pte & RPDS_MASK;
178 	}
179 
180 	/* Need a leaf at lowest level; 512GB pages not supported */
181 	if (level < 0 || level == 3)
182 		return -EINVAL;
183 
184 	/* We found a valid leaf PTE */
185 	/* Offset is now log base 2 of the page size */
186 	gpa = pte & 0x01fffffffffff000ul;
187 	if (gpa & ((1ul << offset) - 1))
188 		return -EINVAL;
189 	gpa |= eaddr & ((1ul << offset) - 1);
190 	for (ps = MMU_PAGE_4K; ps < MMU_PAGE_COUNT; ++ps)
191 		if (offset == mmu_psize_defs[ps].shift)
192 			break;
193 	gpte->page_size = ps;
194 	gpte->page_shift = offset;
195 
196 	gpte->eaddr = eaddr;
197 	gpte->raddr = gpa;
198 
199 	/* Work out permissions */
200 	gpte->may_read = !!(pte & _PAGE_READ);
201 	gpte->may_write = !!(pte & _PAGE_WRITE);
202 	gpte->may_execute = !!(pte & _PAGE_EXEC);
203 
204 	gpte->rc = pte & (_PAGE_ACCESSED | _PAGE_DIRTY);
205 
206 	if (pte_ret_p)
207 		*pte_ret_p = pte;
208 
209 	return 0;
210 }
211 
212 /*
213  * Used to walk a partition or process table radix tree in guest memory
214  * Note: We exploit the fact that a partition table and a process
215  * table have the same layout, a partition-scoped page table and a
216  * process-scoped page table have the same layout, and the 2nd
217  * doubleword of a partition table entry has the same layout as
218  * the PTCR register.
219  */
220 int kvmppc_mmu_radix_translate_table(struct kvm_vcpu *vcpu, gva_t eaddr,
221 				     struct kvmppc_pte *gpte, u64 table,
222 				     int table_index, u64 *pte_ret_p)
223 {
224 	struct kvm *kvm = vcpu->kvm;
225 	int ret;
226 	unsigned long size, ptbl, root;
227 	struct prtb_entry entry;
228 
229 	if ((table & PRTS_MASK) > 24)
230 		return -EINVAL;
231 	size = 1ul << ((table & PRTS_MASK) + 12);
232 
233 	/* Is the table big enough to contain this entry? */
234 	if ((table_index * sizeof(entry)) >= size)
235 		return -EINVAL;
236 
237 	/* Read the table to find the root of the radix tree */
238 	ptbl = (table & PRTB_MASK) + (table_index * sizeof(entry));
239 	ret = kvm_read_guest(kvm, ptbl, &entry, sizeof(entry));
240 	if (ret)
241 		return ret;
242 
243 	/* Root is stored in the first double word */
244 	root = be64_to_cpu(entry.prtb0);
245 
246 	return kvmppc_mmu_walk_radix_tree(vcpu, eaddr, gpte, root, pte_ret_p);
247 }
248 
249 int kvmppc_mmu_radix_xlate(struct kvm_vcpu *vcpu, gva_t eaddr,
250 			   struct kvmppc_pte *gpte, bool data, bool iswrite)
251 {
252 	u32 pid;
253 	u64 pte;
254 	int ret;
255 
256 	/* Work out effective PID */
257 	switch (eaddr >> 62) {
258 	case 0:
259 		pid = vcpu->arch.pid;
260 		break;
261 	case 3:
262 		pid = 0;
263 		break;
264 	default:
265 		return -EINVAL;
266 	}
267 
268 	ret = kvmppc_mmu_radix_translate_table(vcpu, eaddr, gpte,
269 				vcpu->kvm->arch.process_table, pid, &pte);
270 	if (ret)
271 		return ret;
272 
273 	/* Check privilege (applies only to process scoped translations) */
274 	if (kvmppc_get_msr(vcpu) & MSR_PR) {
275 		if (pte & _PAGE_PRIVILEGED) {
276 			gpte->may_read = 0;
277 			gpte->may_write = 0;
278 			gpte->may_execute = 0;
279 		}
280 	} else {
281 		if (!(pte & _PAGE_PRIVILEGED)) {
282 			/* Check AMR/IAMR to see if strict mode is in force */
283 			if (vcpu->arch.amr & (1ul << 62))
284 				gpte->may_read = 0;
285 			if (vcpu->arch.amr & (1ul << 63))
286 				gpte->may_write = 0;
287 			if (vcpu->arch.iamr & (1ul << 62))
288 				gpte->may_execute = 0;
289 		}
290 	}
291 
292 	return 0;
293 }
294 
295 void kvmppc_radix_tlbie_page(struct kvm *kvm, unsigned long addr,
296 			     unsigned int pshift, unsigned int lpid)
297 {
298 	unsigned long psize = PAGE_SIZE;
299 	int psi;
300 	long rc;
301 	unsigned long rb;
302 
303 	if (pshift)
304 		psize = 1UL << pshift;
305 	else
306 		pshift = PAGE_SHIFT;
307 
308 	addr &= ~(psize - 1);
309 
310 	if (!kvmhv_on_pseries()) {
311 		radix__flush_tlb_lpid_page(lpid, addr, psize);
312 		return;
313 	}
314 
315 	psi = shift_to_mmu_psize(pshift);
316 	rb = addr | (mmu_get_ap(psi) << PPC_BITLSHIFT(58));
317 	rc = plpar_hcall_norets(H_TLB_INVALIDATE, H_TLBIE_P1_ENC(0, 0, 1),
318 				lpid, rb);
319 	if (rc)
320 		pr_err("KVM: TLB page invalidation hcall failed, rc=%ld\n", rc);
321 }
322 
323 static void kvmppc_radix_flush_pwc(struct kvm *kvm, unsigned int lpid)
324 {
325 	long rc;
326 
327 	if (!kvmhv_on_pseries()) {
328 		radix__flush_pwc_lpid(lpid);
329 		return;
330 	}
331 
332 	rc = plpar_hcall_norets(H_TLB_INVALIDATE, H_TLBIE_P1_ENC(1, 0, 1),
333 				lpid, TLBIEL_INVAL_SET_LPID);
334 	if (rc)
335 		pr_err("KVM: TLB PWC invalidation hcall failed, rc=%ld\n", rc);
336 }
337 
338 static unsigned long kvmppc_radix_update_pte(struct kvm *kvm, pte_t *ptep,
339 				      unsigned long clr, unsigned long set,
340 				      unsigned long addr, unsigned int shift)
341 {
342 	return __radix_pte_update(ptep, clr, set);
343 }
344 
345 void kvmppc_radix_set_pte_at(struct kvm *kvm, unsigned long addr,
346 			     pte_t *ptep, pte_t pte)
347 {
348 	radix__set_pte_at(kvm->mm, addr, ptep, pte, 0);
349 }
350 
351 static struct kmem_cache *kvm_pte_cache;
352 static struct kmem_cache *kvm_pmd_cache;
353 
354 static pte_t *kvmppc_pte_alloc(void)
355 {
356 	return kmem_cache_alloc(kvm_pte_cache, GFP_KERNEL);
357 }
358 
359 static void kvmppc_pte_free(pte_t *ptep)
360 {
361 	kmem_cache_free(kvm_pte_cache, ptep);
362 }
363 
364 static pmd_t *kvmppc_pmd_alloc(void)
365 {
366 	return kmem_cache_alloc(kvm_pmd_cache, GFP_KERNEL);
367 }
368 
369 static void kvmppc_pmd_free(pmd_t *pmdp)
370 {
371 	kmem_cache_free(kvm_pmd_cache, pmdp);
372 }
373 
374 /* Called with kvm->mmu_lock held */
375 void kvmppc_unmap_pte(struct kvm *kvm, pte_t *pte, unsigned long gpa,
376 		      unsigned int shift,
377 		      const struct kvm_memory_slot *memslot,
378 		      unsigned int lpid)
379 
380 {
381 	unsigned long old;
382 	unsigned long gfn = gpa >> PAGE_SHIFT;
383 	unsigned long page_size = PAGE_SIZE;
384 	unsigned long hpa;
385 
386 	old = kvmppc_radix_update_pte(kvm, pte, ~0UL, 0, gpa, shift);
387 	kvmppc_radix_tlbie_page(kvm, gpa, shift, lpid);
388 
389 	/* The following only applies to L1 entries */
390 	if (lpid != kvm->arch.lpid)
391 		return;
392 
393 	if (!memslot) {
394 		memslot = gfn_to_memslot(kvm, gfn);
395 		if (!memslot)
396 			return;
397 	}
398 	if (shift) { /* 1GB or 2MB page */
399 		page_size = 1ul << shift;
400 		if (shift == PMD_SHIFT)
401 			kvm->stat.num_2M_pages--;
402 		else if (shift == PUD_SHIFT)
403 			kvm->stat.num_1G_pages--;
404 	}
405 
406 	gpa &= ~(page_size - 1);
407 	hpa = old & PTE_RPN_MASK;
408 	kvmhv_remove_nest_rmap_range(kvm, memslot, gpa, hpa, page_size);
409 
410 	if ((old & _PAGE_DIRTY) && memslot->dirty_bitmap)
411 		kvmppc_update_dirty_map(memslot, gfn, page_size);
412 }
413 
414 /*
415  * kvmppc_free_p?d are used to free existing page tables, and recursively
416  * descend and clear and free children.
417  * Callers are responsible for flushing the PWC.
418  *
419  * When page tables are being unmapped/freed as part of page fault path
420  * (full == false), ptes are not expected. There is code to unmap them
421  * and emit a warning if encountered, but there may already be data
422  * corruption due to the unexpected mappings.
423  */
424 static void kvmppc_unmap_free_pte(struct kvm *kvm, pte_t *pte, bool full,
425 				  unsigned int lpid)
426 {
427 	if (full) {
428 		memset(pte, 0, sizeof(long) << RADIX_PTE_INDEX_SIZE);
429 	} else {
430 		pte_t *p = pte;
431 		unsigned long it;
432 
433 		for (it = 0; it < PTRS_PER_PTE; ++it, ++p) {
434 			if (pte_val(*p) == 0)
435 				continue;
436 			WARN_ON_ONCE(1);
437 			kvmppc_unmap_pte(kvm, p,
438 					 pte_pfn(*p) << PAGE_SHIFT,
439 					 PAGE_SHIFT, NULL, lpid);
440 		}
441 	}
442 
443 	kvmppc_pte_free(pte);
444 }
445 
446 static void kvmppc_unmap_free_pmd(struct kvm *kvm, pmd_t *pmd, bool full,
447 				  unsigned int lpid)
448 {
449 	unsigned long im;
450 	pmd_t *p = pmd;
451 
452 	for (im = 0; im < PTRS_PER_PMD; ++im, ++p) {
453 		if (!pmd_present(*p))
454 			continue;
455 		if (pmd_is_leaf(*p)) {
456 			if (full) {
457 				pmd_clear(p);
458 			} else {
459 				WARN_ON_ONCE(1);
460 				kvmppc_unmap_pte(kvm, (pte_t *)p,
461 					 pte_pfn(*(pte_t *)p) << PAGE_SHIFT,
462 					 PMD_SHIFT, NULL, lpid);
463 			}
464 		} else {
465 			pte_t *pte;
466 
467 			pte = pte_offset_map(p, 0);
468 			kvmppc_unmap_free_pte(kvm, pte, full, lpid);
469 			pmd_clear(p);
470 		}
471 	}
472 	kvmppc_pmd_free(pmd);
473 }
474 
475 static void kvmppc_unmap_free_pud(struct kvm *kvm, pud_t *pud,
476 				  unsigned int lpid)
477 {
478 	unsigned long iu;
479 	pud_t *p = pud;
480 
481 	for (iu = 0; iu < PTRS_PER_PUD; ++iu, ++p) {
482 		if (!pud_present(*p))
483 			continue;
484 		if (pud_is_leaf(*p)) {
485 			pud_clear(p);
486 		} else {
487 			pmd_t *pmd;
488 
489 			pmd = pmd_offset(p, 0);
490 			kvmppc_unmap_free_pmd(kvm, pmd, true, lpid);
491 			pud_clear(p);
492 		}
493 	}
494 	pud_free(kvm->mm, pud);
495 }
496 
497 void kvmppc_free_pgtable_radix(struct kvm *kvm, pgd_t *pgd, unsigned int lpid)
498 {
499 	unsigned long ig;
500 
501 	for (ig = 0; ig < PTRS_PER_PGD; ++ig, ++pgd) {
502 		pud_t *pud;
503 
504 		if (!pgd_present(*pgd))
505 			continue;
506 		pud = pud_offset(pgd, 0);
507 		kvmppc_unmap_free_pud(kvm, pud, lpid);
508 		pgd_clear(pgd);
509 	}
510 }
511 
512 void kvmppc_free_radix(struct kvm *kvm)
513 {
514 	if (kvm->arch.pgtable) {
515 		kvmppc_free_pgtable_radix(kvm, kvm->arch.pgtable,
516 					  kvm->arch.lpid);
517 		pgd_free(kvm->mm, kvm->arch.pgtable);
518 		kvm->arch.pgtable = NULL;
519 	}
520 }
521 
522 static void kvmppc_unmap_free_pmd_entry_table(struct kvm *kvm, pmd_t *pmd,
523 					unsigned long gpa, unsigned int lpid)
524 {
525 	pte_t *pte = pte_offset_kernel(pmd, 0);
526 
527 	/*
528 	 * Clearing the pmd entry then flushing the PWC ensures that the pte
529 	 * page no longer be cached by the MMU, so can be freed without
530 	 * flushing the PWC again.
531 	 */
532 	pmd_clear(pmd);
533 	kvmppc_radix_flush_pwc(kvm, lpid);
534 
535 	kvmppc_unmap_free_pte(kvm, pte, false, lpid);
536 }
537 
538 static void kvmppc_unmap_free_pud_entry_table(struct kvm *kvm, pud_t *pud,
539 					unsigned long gpa, unsigned int lpid)
540 {
541 	pmd_t *pmd = pmd_offset(pud, 0);
542 
543 	/*
544 	 * Clearing the pud entry then flushing the PWC ensures that the pmd
545 	 * page and any children pte pages will no longer be cached by the MMU,
546 	 * so can be freed without flushing the PWC again.
547 	 */
548 	pud_clear(pud);
549 	kvmppc_radix_flush_pwc(kvm, lpid);
550 
551 	kvmppc_unmap_free_pmd(kvm, pmd, false, lpid);
552 }
553 
554 /*
555  * There are a number of bits which may differ between different faults to
556  * the same partition scope entry. RC bits, in the course of cleaning and
557  * aging. And the write bit can change, either the access could have been
558  * upgraded, or a read fault could happen concurrently with a write fault
559  * that sets those bits first.
560  */
561 #define PTE_BITS_MUST_MATCH (~(_PAGE_WRITE | _PAGE_DIRTY | _PAGE_ACCESSED))
562 
563 int kvmppc_create_pte(struct kvm *kvm, pgd_t *pgtable, pte_t pte,
564 		      unsigned long gpa, unsigned int level,
565 		      unsigned long mmu_seq, unsigned int lpid,
566 		      unsigned long *rmapp, struct rmap_nested **n_rmap)
567 {
568 	pgd_t *pgd;
569 	pud_t *pud, *new_pud = NULL;
570 	pmd_t *pmd, *new_pmd = NULL;
571 	pte_t *ptep, *new_ptep = NULL;
572 	int ret;
573 
574 	/* Traverse the guest's 2nd-level tree, allocate new levels needed */
575 	pgd = pgtable + pgd_index(gpa);
576 	pud = NULL;
577 	if (pgd_present(*pgd))
578 		pud = pud_offset(pgd, gpa);
579 	else
580 		new_pud = pud_alloc_one(kvm->mm, gpa);
581 
582 	pmd = NULL;
583 	if (pud && pud_present(*pud) && !pud_is_leaf(*pud))
584 		pmd = pmd_offset(pud, gpa);
585 	else if (level <= 1)
586 		new_pmd = kvmppc_pmd_alloc();
587 
588 	if (level == 0 && !(pmd && pmd_present(*pmd) && !pmd_is_leaf(*pmd)))
589 		new_ptep = kvmppc_pte_alloc();
590 
591 	/* Check if we might have been invalidated; let the guest retry if so */
592 	spin_lock(&kvm->mmu_lock);
593 	ret = -EAGAIN;
594 	if (mmu_notifier_retry(kvm, mmu_seq))
595 		goto out_unlock;
596 
597 	/* Now traverse again under the lock and change the tree */
598 	ret = -ENOMEM;
599 	if (pgd_none(*pgd)) {
600 		if (!new_pud)
601 			goto out_unlock;
602 		pgd_populate(kvm->mm, pgd, new_pud);
603 		new_pud = NULL;
604 	}
605 	pud = pud_offset(pgd, gpa);
606 	if (pud_is_leaf(*pud)) {
607 		unsigned long hgpa = gpa & PUD_MASK;
608 
609 		/* Check if we raced and someone else has set the same thing */
610 		if (level == 2) {
611 			if (pud_raw(*pud) == pte_raw(pte)) {
612 				ret = 0;
613 				goto out_unlock;
614 			}
615 			/* Valid 1GB page here already, add our extra bits */
616 			WARN_ON_ONCE((pud_val(*pud) ^ pte_val(pte)) &
617 							PTE_BITS_MUST_MATCH);
618 			kvmppc_radix_update_pte(kvm, (pte_t *)pud,
619 					      0, pte_val(pte), hgpa, PUD_SHIFT);
620 			ret = 0;
621 			goto out_unlock;
622 		}
623 		/*
624 		 * If we raced with another CPU which has just put
625 		 * a 1GB pte in after we saw a pmd page, try again.
626 		 */
627 		if (!new_pmd) {
628 			ret = -EAGAIN;
629 			goto out_unlock;
630 		}
631 		/* Valid 1GB page here already, remove it */
632 		kvmppc_unmap_pte(kvm, (pte_t *)pud, hgpa, PUD_SHIFT, NULL,
633 				 lpid);
634 	}
635 	if (level == 2) {
636 		if (!pud_none(*pud)) {
637 			/*
638 			 * There's a page table page here, but we wanted to
639 			 * install a large page, so remove and free the page
640 			 * table page.
641 			 */
642 			kvmppc_unmap_free_pud_entry_table(kvm, pud, gpa, lpid);
643 		}
644 		kvmppc_radix_set_pte_at(kvm, gpa, (pte_t *)pud, pte);
645 		if (rmapp && n_rmap)
646 			kvmhv_insert_nest_rmap(kvm, rmapp, n_rmap);
647 		ret = 0;
648 		goto out_unlock;
649 	}
650 	if (pud_none(*pud)) {
651 		if (!new_pmd)
652 			goto out_unlock;
653 		pud_populate(kvm->mm, pud, new_pmd);
654 		new_pmd = NULL;
655 	}
656 	pmd = pmd_offset(pud, gpa);
657 	if (pmd_is_leaf(*pmd)) {
658 		unsigned long lgpa = gpa & PMD_MASK;
659 
660 		/* Check if we raced and someone else has set the same thing */
661 		if (level == 1) {
662 			if (pmd_raw(*pmd) == pte_raw(pte)) {
663 				ret = 0;
664 				goto out_unlock;
665 			}
666 			/* Valid 2MB page here already, add our extra bits */
667 			WARN_ON_ONCE((pmd_val(*pmd) ^ pte_val(pte)) &
668 							PTE_BITS_MUST_MATCH);
669 			kvmppc_radix_update_pte(kvm, pmdp_ptep(pmd),
670 					0, pte_val(pte), lgpa, PMD_SHIFT);
671 			ret = 0;
672 			goto out_unlock;
673 		}
674 
675 		/*
676 		 * If we raced with another CPU which has just put
677 		 * a 2MB pte in after we saw a pte page, try again.
678 		 */
679 		if (!new_ptep) {
680 			ret = -EAGAIN;
681 			goto out_unlock;
682 		}
683 		/* Valid 2MB page here already, remove it */
684 		kvmppc_unmap_pte(kvm, pmdp_ptep(pmd), lgpa, PMD_SHIFT, NULL,
685 				 lpid);
686 	}
687 	if (level == 1) {
688 		if (!pmd_none(*pmd)) {
689 			/*
690 			 * There's a page table page here, but we wanted to
691 			 * install a large page, so remove and free the page
692 			 * table page.
693 			 */
694 			kvmppc_unmap_free_pmd_entry_table(kvm, pmd, gpa, lpid);
695 		}
696 		kvmppc_radix_set_pte_at(kvm, gpa, pmdp_ptep(pmd), pte);
697 		if (rmapp && n_rmap)
698 			kvmhv_insert_nest_rmap(kvm, rmapp, n_rmap);
699 		ret = 0;
700 		goto out_unlock;
701 	}
702 	if (pmd_none(*pmd)) {
703 		if (!new_ptep)
704 			goto out_unlock;
705 		pmd_populate(kvm->mm, pmd, new_ptep);
706 		new_ptep = NULL;
707 	}
708 	ptep = pte_offset_kernel(pmd, gpa);
709 	if (pte_present(*ptep)) {
710 		/* Check if someone else set the same thing */
711 		if (pte_raw(*ptep) == pte_raw(pte)) {
712 			ret = 0;
713 			goto out_unlock;
714 		}
715 		/* Valid page here already, add our extra bits */
716 		WARN_ON_ONCE((pte_val(*ptep) ^ pte_val(pte)) &
717 							PTE_BITS_MUST_MATCH);
718 		kvmppc_radix_update_pte(kvm, ptep, 0, pte_val(pte), gpa, 0);
719 		ret = 0;
720 		goto out_unlock;
721 	}
722 	kvmppc_radix_set_pte_at(kvm, gpa, ptep, pte);
723 	if (rmapp && n_rmap)
724 		kvmhv_insert_nest_rmap(kvm, rmapp, n_rmap);
725 	ret = 0;
726 
727  out_unlock:
728 	spin_unlock(&kvm->mmu_lock);
729 	if (new_pud)
730 		pud_free(kvm->mm, new_pud);
731 	if (new_pmd)
732 		kvmppc_pmd_free(new_pmd);
733 	if (new_ptep)
734 		kvmppc_pte_free(new_ptep);
735 	return ret;
736 }
737 
738 bool kvmppc_hv_handle_set_rc(struct kvm *kvm, pgd_t *pgtable, bool writing,
739 			     unsigned long gpa, unsigned int lpid)
740 {
741 	unsigned long pgflags;
742 	unsigned int shift;
743 	pte_t *ptep;
744 
745 	/*
746 	 * Need to set an R or C bit in the 2nd-level tables;
747 	 * since we are just helping out the hardware here,
748 	 * it is sufficient to do what the hardware does.
749 	 */
750 	pgflags = _PAGE_ACCESSED;
751 	if (writing)
752 		pgflags |= _PAGE_DIRTY;
753 	/*
754 	 * We are walking the secondary (partition-scoped) page table here.
755 	 * We can do this without disabling irq because the Linux MM
756 	 * subsystem doesn't do THP splits and collapses on this tree.
757 	 */
758 	ptep = __find_linux_pte(pgtable, gpa, NULL, &shift);
759 	if (ptep && pte_present(*ptep) && (!writing || pte_write(*ptep))) {
760 		kvmppc_radix_update_pte(kvm, ptep, 0, pgflags, gpa, shift);
761 		return true;
762 	}
763 	return false;
764 }
765 
766 int kvmppc_book3s_instantiate_page(struct kvm_vcpu *vcpu,
767 				   unsigned long gpa,
768 				   struct kvm_memory_slot *memslot,
769 				   bool writing, bool kvm_ro,
770 				   pte_t *inserted_pte, unsigned int *levelp)
771 {
772 	struct kvm *kvm = vcpu->kvm;
773 	struct page *page = NULL;
774 	unsigned long mmu_seq;
775 	unsigned long hva, gfn = gpa >> PAGE_SHIFT;
776 	bool upgrade_write = false;
777 	bool *upgrade_p = &upgrade_write;
778 	pte_t pte, *ptep;
779 	unsigned int shift, level;
780 	int ret;
781 	bool large_enable;
782 
783 	/* used to check for invalidations in progress */
784 	mmu_seq = kvm->mmu_notifier_seq;
785 	smp_rmb();
786 
787 	/*
788 	 * Do a fast check first, since __gfn_to_pfn_memslot doesn't
789 	 * do it with !atomic && !async, which is how we call it.
790 	 * We always ask for write permission since the common case
791 	 * is that the page is writable.
792 	 */
793 	hva = gfn_to_hva_memslot(memslot, gfn);
794 	if (!kvm_ro && __get_user_pages_fast(hva, 1, 1, &page) == 1) {
795 		upgrade_write = true;
796 	} else {
797 		unsigned long pfn;
798 
799 		/* Call KVM generic code to do the slow-path check */
800 		pfn = __gfn_to_pfn_memslot(memslot, gfn, false, NULL,
801 					   writing, upgrade_p);
802 		if (is_error_noslot_pfn(pfn))
803 			return -EFAULT;
804 		page = NULL;
805 		if (pfn_valid(pfn)) {
806 			page = pfn_to_page(pfn);
807 			if (PageReserved(page))
808 				page = NULL;
809 		}
810 	}
811 
812 	/*
813 	 * Read the PTE from the process' radix tree and use that
814 	 * so we get the shift and attribute bits.
815 	 */
816 	local_irq_disable();
817 	ptep = __find_linux_pte(vcpu->arch.pgdir, hva, NULL, &shift);
818 	pte = __pte(0);
819 	if (ptep)
820 		pte = *ptep;
821 	local_irq_enable();
822 	/*
823 	 * If the PTE disappeared temporarily due to a THP
824 	 * collapse, just return and let the guest try again.
825 	 */
826 	if (!pte_present(pte)) {
827 		if (page)
828 			put_page(page);
829 		return RESUME_GUEST;
830 	}
831 
832 	/* If we're logging dirty pages, always map single pages */
833 	large_enable = !(memslot->flags & KVM_MEM_LOG_DIRTY_PAGES);
834 
835 	/* Get pte level from shift/size */
836 	if (large_enable && shift == PUD_SHIFT &&
837 	    (gpa & (PUD_SIZE - PAGE_SIZE)) ==
838 	    (hva & (PUD_SIZE - PAGE_SIZE))) {
839 		level = 2;
840 	} else if (large_enable && shift == PMD_SHIFT &&
841 		   (gpa & (PMD_SIZE - PAGE_SIZE)) ==
842 		   (hva & (PMD_SIZE - PAGE_SIZE))) {
843 		level = 1;
844 	} else {
845 		level = 0;
846 		if (shift > PAGE_SHIFT) {
847 			/*
848 			 * If the pte maps more than one page, bring over
849 			 * bits from the virtual address to get the real
850 			 * address of the specific single page we want.
851 			 */
852 			unsigned long rpnmask = (1ul << shift) - PAGE_SIZE;
853 			pte = __pte(pte_val(pte) | (hva & rpnmask));
854 		}
855 	}
856 
857 	pte = __pte(pte_val(pte) | _PAGE_EXEC | _PAGE_ACCESSED);
858 	if (writing || upgrade_write) {
859 		if (pte_val(pte) & _PAGE_WRITE)
860 			pte = __pte(pte_val(pte) | _PAGE_DIRTY);
861 	} else {
862 		pte = __pte(pte_val(pte) & ~(_PAGE_WRITE | _PAGE_DIRTY));
863 	}
864 
865 	/* Allocate space in the tree and write the PTE */
866 	ret = kvmppc_create_pte(kvm, kvm->arch.pgtable, pte, gpa, level,
867 				mmu_seq, kvm->arch.lpid, NULL, NULL);
868 	if (inserted_pte)
869 		*inserted_pte = pte;
870 	if (levelp)
871 		*levelp = level;
872 
873 	if (page) {
874 		if (!ret && (pte_val(pte) & _PAGE_WRITE))
875 			set_page_dirty_lock(page);
876 		put_page(page);
877 	}
878 
879 	/* Increment number of large pages if we (successfully) inserted one */
880 	if (!ret) {
881 		if (level == 1)
882 			kvm->stat.num_2M_pages++;
883 		else if (level == 2)
884 			kvm->stat.num_1G_pages++;
885 	}
886 
887 	return ret;
888 }
889 
890 int kvmppc_book3s_radix_page_fault(struct kvm_run *run, struct kvm_vcpu *vcpu,
891 				   unsigned long ea, unsigned long dsisr)
892 {
893 	struct kvm *kvm = vcpu->kvm;
894 	unsigned long gpa, gfn;
895 	struct kvm_memory_slot *memslot;
896 	long ret;
897 	bool writing = !!(dsisr & DSISR_ISSTORE);
898 	bool kvm_ro = false;
899 
900 	/* Check for unusual errors */
901 	if (dsisr & DSISR_UNSUPP_MMU) {
902 		pr_err("KVM: Got unsupported MMU fault\n");
903 		return -EFAULT;
904 	}
905 	if (dsisr & DSISR_BADACCESS) {
906 		/* Reflect to the guest as DSI */
907 		pr_err("KVM: Got radix HV page fault with DSISR=%lx\n", dsisr);
908 		kvmppc_core_queue_data_storage(vcpu, ea, dsisr);
909 		return RESUME_GUEST;
910 	}
911 
912 	/* Translate the logical address */
913 	gpa = vcpu->arch.fault_gpa & ~0xfffUL;
914 	gpa &= ~0xF000000000000000ul;
915 	gfn = gpa >> PAGE_SHIFT;
916 	if (!(dsisr & DSISR_PRTABLE_FAULT))
917 		gpa |= ea & 0xfff;
918 
919 	if (kvm->arch.secure_guest & KVMPPC_SECURE_INIT_DONE)
920 		return kvmppc_send_page_to_uv(kvm, gfn);
921 
922 	/* Get the corresponding memslot */
923 	memslot = gfn_to_memslot(kvm, gfn);
924 
925 	/* No memslot means it's an emulated MMIO region */
926 	if (!memslot || (memslot->flags & KVM_MEMSLOT_INVALID)) {
927 		if (dsisr & (DSISR_PRTABLE_FAULT | DSISR_BADACCESS |
928 			     DSISR_SET_RC)) {
929 			/*
930 			 * Bad address in guest page table tree, or other
931 			 * unusual error - reflect it to the guest as DSI.
932 			 */
933 			kvmppc_core_queue_data_storage(vcpu, ea, dsisr);
934 			return RESUME_GUEST;
935 		}
936 		return kvmppc_hv_emulate_mmio(run, vcpu, gpa, ea, writing);
937 	}
938 
939 	if (memslot->flags & KVM_MEM_READONLY) {
940 		if (writing) {
941 			/* give the guest a DSI */
942 			kvmppc_core_queue_data_storage(vcpu, ea, DSISR_ISSTORE |
943 						       DSISR_PROTFAULT);
944 			return RESUME_GUEST;
945 		}
946 		kvm_ro = true;
947 	}
948 
949 	/* Failed to set the reference/change bits */
950 	if (dsisr & DSISR_SET_RC) {
951 		spin_lock(&kvm->mmu_lock);
952 		if (kvmppc_hv_handle_set_rc(kvm, kvm->arch.pgtable,
953 					    writing, gpa, kvm->arch.lpid))
954 			dsisr &= ~DSISR_SET_RC;
955 		spin_unlock(&kvm->mmu_lock);
956 
957 		if (!(dsisr & (DSISR_BAD_FAULT_64S | DSISR_NOHPTE |
958 			       DSISR_PROTFAULT | DSISR_SET_RC)))
959 			return RESUME_GUEST;
960 	}
961 
962 	/* Try to insert a pte */
963 	ret = kvmppc_book3s_instantiate_page(vcpu, gpa, memslot, writing,
964 					     kvm_ro, NULL, NULL);
965 
966 	if (ret == 0 || ret == -EAGAIN)
967 		ret = RESUME_GUEST;
968 	return ret;
969 }
970 
971 /* Called with kvm->mmu_lock held */
972 int kvm_unmap_radix(struct kvm *kvm, struct kvm_memory_slot *memslot,
973 		    unsigned long gfn)
974 {
975 	pte_t *ptep;
976 	unsigned long gpa = gfn << PAGE_SHIFT;
977 	unsigned int shift;
978 
979 	if (kvm->arch.secure_guest & KVMPPC_SECURE_INIT_DONE) {
980 		uv_page_inval(kvm->arch.lpid, gpa, PAGE_SHIFT);
981 		return 0;
982 	}
983 
984 	ptep = __find_linux_pte(kvm->arch.pgtable, gpa, NULL, &shift);
985 	if (ptep && pte_present(*ptep))
986 		kvmppc_unmap_pte(kvm, ptep, gpa, shift, memslot,
987 				 kvm->arch.lpid);
988 	return 0;
989 }
990 
991 /* Called with kvm->mmu_lock held */
992 int kvm_age_radix(struct kvm *kvm, struct kvm_memory_slot *memslot,
993 		  unsigned long gfn)
994 {
995 	pte_t *ptep;
996 	unsigned long gpa = gfn << PAGE_SHIFT;
997 	unsigned int shift;
998 	int ref = 0;
999 	unsigned long old, *rmapp;
1000 
1001 	if (kvm->arch.secure_guest & KVMPPC_SECURE_INIT_DONE)
1002 		return ref;
1003 
1004 	ptep = __find_linux_pte(kvm->arch.pgtable, gpa, NULL, &shift);
1005 	if (ptep && pte_present(*ptep) && pte_young(*ptep)) {
1006 		old = kvmppc_radix_update_pte(kvm, ptep, _PAGE_ACCESSED, 0,
1007 					      gpa, shift);
1008 		/* XXX need to flush tlb here? */
1009 		/* Also clear bit in ptes in shadow pgtable for nested guests */
1010 		rmapp = &memslot->arch.rmap[gfn - memslot->base_gfn];
1011 		kvmhv_update_nest_rmap_rc_list(kvm, rmapp, _PAGE_ACCESSED, 0,
1012 					       old & PTE_RPN_MASK,
1013 					       1UL << shift);
1014 		ref = 1;
1015 	}
1016 	return ref;
1017 }
1018 
1019 /* Called with kvm->mmu_lock held */
1020 int kvm_test_age_radix(struct kvm *kvm, struct kvm_memory_slot *memslot,
1021 		       unsigned long gfn)
1022 {
1023 	pte_t *ptep;
1024 	unsigned long gpa = gfn << PAGE_SHIFT;
1025 	unsigned int shift;
1026 	int ref = 0;
1027 
1028 	if (kvm->arch.secure_guest & KVMPPC_SECURE_INIT_DONE)
1029 		return ref;
1030 
1031 	ptep = __find_linux_pte(kvm->arch.pgtable, gpa, NULL, &shift);
1032 	if (ptep && pte_present(*ptep) && pte_young(*ptep))
1033 		ref = 1;
1034 	return ref;
1035 }
1036 
1037 /* Returns the number of PAGE_SIZE pages that are dirty */
1038 static int kvm_radix_test_clear_dirty(struct kvm *kvm,
1039 				struct kvm_memory_slot *memslot, int pagenum)
1040 {
1041 	unsigned long gfn = memslot->base_gfn + pagenum;
1042 	unsigned long gpa = gfn << PAGE_SHIFT;
1043 	pte_t *ptep;
1044 	unsigned int shift;
1045 	int ret = 0;
1046 	unsigned long old, *rmapp;
1047 
1048 	if (kvm->arch.secure_guest & KVMPPC_SECURE_INIT_DONE)
1049 		return ret;
1050 
1051 	ptep = __find_linux_pte(kvm->arch.pgtable, gpa, NULL, &shift);
1052 	if (ptep && pte_present(*ptep) && pte_dirty(*ptep)) {
1053 		ret = 1;
1054 		if (shift)
1055 			ret = 1 << (shift - PAGE_SHIFT);
1056 		spin_lock(&kvm->mmu_lock);
1057 		old = kvmppc_radix_update_pte(kvm, ptep, _PAGE_DIRTY, 0,
1058 					      gpa, shift);
1059 		kvmppc_radix_tlbie_page(kvm, gpa, shift, kvm->arch.lpid);
1060 		/* Also clear bit in ptes in shadow pgtable for nested guests */
1061 		rmapp = &memslot->arch.rmap[gfn - memslot->base_gfn];
1062 		kvmhv_update_nest_rmap_rc_list(kvm, rmapp, _PAGE_DIRTY, 0,
1063 					       old & PTE_RPN_MASK,
1064 					       1UL << shift);
1065 		spin_unlock(&kvm->mmu_lock);
1066 	}
1067 	return ret;
1068 }
1069 
1070 long kvmppc_hv_get_dirty_log_radix(struct kvm *kvm,
1071 			struct kvm_memory_slot *memslot, unsigned long *map)
1072 {
1073 	unsigned long i, j;
1074 	int npages;
1075 
1076 	for (i = 0; i < memslot->npages; i = j) {
1077 		npages = kvm_radix_test_clear_dirty(kvm, memslot, i);
1078 
1079 		/*
1080 		 * Note that if npages > 0 then i must be a multiple of npages,
1081 		 * since huge pages are only used to back the guest at guest
1082 		 * real addresses that are a multiple of their size.
1083 		 * Since we have at most one PTE covering any given guest
1084 		 * real address, if npages > 1 we can skip to i + npages.
1085 		 */
1086 		j = i + 1;
1087 		if (npages) {
1088 			set_dirty_bits(map, i, npages);
1089 			j = i + npages;
1090 		}
1091 	}
1092 	return 0;
1093 }
1094 
1095 void kvmppc_radix_flush_memslot(struct kvm *kvm,
1096 				const struct kvm_memory_slot *memslot)
1097 {
1098 	unsigned long n;
1099 	pte_t *ptep;
1100 	unsigned long gpa;
1101 	unsigned int shift;
1102 
1103 	if (kvm->arch.secure_guest & KVMPPC_SECURE_INIT_START)
1104 		kvmppc_uvmem_drop_pages(memslot, kvm, true);
1105 
1106 	if (kvm->arch.secure_guest & KVMPPC_SECURE_INIT_DONE)
1107 		return;
1108 
1109 	gpa = memslot->base_gfn << PAGE_SHIFT;
1110 	spin_lock(&kvm->mmu_lock);
1111 	for (n = memslot->npages; n; --n) {
1112 		ptep = __find_linux_pte(kvm->arch.pgtable, gpa, NULL, &shift);
1113 		if (ptep && pte_present(*ptep))
1114 			kvmppc_unmap_pte(kvm, ptep, gpa, shift, memslot,
1115 					 kvm->arch.lpid);
1116 		gpa += PAGE_SIZE;
1117 	}
1118 	spin_unlock(&kvm->mmu_lock);
1119 }
1120 
1121 static void add_rmmu_ap_encoding(struct kvm_ppc_rmmu_info *info,
1122 				 int psize, int *indexp)
1123 {
1124 	if (!mmu_psize_defs[psize].shift)
1125 		return;
1126 	info->ap_encodings[*indexp] = mmu_psize_defs[psize].shift |
1127 		(mmu_psize_defs[psize].ap << 29);
1128 	++(*indexp);
1129 }
1130 
1131 int kvmhv_get_rmmu_info(struct kvm *kvm, struct kvm_ppc_rmmu_info *info)
1132 {
1133 	int i;
1134 
1135 	if (!radix_enabled())
1136 		return -EINVAL;
1137 	memset(info, 0, sizeof(*info));
1138 
1139 	/* 4k page size */
1140 	info->geometries[0].page_shift = 12;
1141 	info->geometries[0].level_bits[0] = 9;
1142 	for (i = 1; i < 4; ++i)
1143 		info->geometries[0].level_bits[i] = p9_supported_radix_bits[i];
1144 	/* 64k page size */
1145 	info->geometries[1].page_shift = 16;
1146 	for (i = 0; i < 4; ++i)
1147 		info->geometries[1].level_bits[i] = p9_supported_radix_bits[i];
1148 
1149 	i = 0;
1150 	add_rmmu_ap_encoding(info, MMU_PAGE_4K, &i);
1151 	add_rmmu_ap_encoding(info, MMU_PAGE_64K, &i);
1152 	add_rmmu_ap_encoding(info, MMU_PAGE_2M, &i);
1153 	add_rmmu_ap_encoding(info, MMU_PAGE_1G, &i);
1154 
1155 	return 0;
1156 }
1157 
1158 int kvmppc_init_vm_radix(struct kvm *kvm)
1159 {
1160 	kvm->arch.pgtable = pgd_alloc(kvm->mm);
1161 	if (!kvm->arch.pgtable)
1162 		return -ENOMEM;
1163 	return 0;
1164 }
1165 
1166 static void pte_ctor(void *addr)
1167 {
1168 	memset(addr, 0, RADIX_PTE_TABLE_SIZE);
1169 }
1170 
1171 static void pmd_ctor(void *addr)
1172 {
1173 	memset(addr, 0, RADIX_PMD_TABLE_SIZE);
1174 }
1175 
1176 struct debugfs_radix_state {
1177 	struct kvm	*kvm;
1178 	struct mutex	mutex;
1179 	unsigned long	gpa;
1180 	int		lpid;
1181 	int		chars_left;
1182 	int		buf_index;
1183 	char		buf[128];
1184 	u8		hdr;
1185 };
1186 
1187 static int debugfs_radix_open(struct inode *inode, struct file *file)
1188 {
1189 	struct kvm *kvm = inode->i_private;
1190 	struct debugfs_radix_state *p;
1191 
1192 	p = kzalloc(sizeof(*p), GFP_KERNEL);
1193 	if (!p)
1194 		return -ENOMEM;
1195 
1196 	kvm_get_kvm(kvm);
1197 	p->kvm = kvm;
1198 	mutex_init(&p->mutex);
1199 	file->private_data = p;
1200 
1201 	return nonseekable_open(inode, file);
1202 }
1203 
1204 static int debugfs_radix_release(struct inode *inode, struct file *file)
1205 {
1206 	struct debugfs_radix_state *p = file->private_data;
1207 
1208 	kvm_put_kvm(p->kvm);
1209 	kfree(p);
1210 	return 0;
1211 }
1212 
1213 static ssize_t debugfs_radix_read(struct file *file, char __user *buf,
1214 				 size_t len, loff_t *ppos)
1215 {
1216 	struct debugfs_radix_state *p = file->private_data;
1217 	ssize_t ret, r;
1218 	unsigned long n;
1219 	struct kvm *kvm;
1220 	unsigned long gpa;
1221 	pgd_t *pgt;
1222 	struct kvm_nested_guest *nested;
1223 	pgd_t pgd, *pgdp;
1224 	pud_t pud, *pudp;
1225 	pmd_t pmd, *pmdp;
1226 	pte_t *ptep;
1227 	int shift;
1228 	unsigned long pte;
1229 
1230 	kvm = p->kvm;
1231 	if (!kvm_is_radix(kvm))
1232 		return 0;
1233 
1234 	ret = mutex_lock_interruptible(&p->mutex);
1235 	if (ret)
1236 		return ret;
1237 
1238 	if (p->chars_left) {
1239 		n = p->chars_left;
1240 		if (n > len)
1241 			n = len;
1242 		r = copy_to_user(buf, p->buf + p->buf_index, n);
1243 		n -= r;
1244 		p->chars_left -= n;
1245 		p->buf_index += n;
1246 		buf += n;
1247 		len -= n;
1248 		ret = n;
1249 		if (r) {
1250 			if (!n)
1251 				ret = -EFAULT;
1252 			goto out;
1253 		}
1254 	}
1255 
1256 	gpa = p->gpa;
1257 	nested = NULL;
1258 	pgt = NULL;
1259 	while (len != 0 && p->lpid >= 0) {
1260 		if (gpa >= RADIX_PGTABLE_RANGE) {
1261 			gpa = 0;
1262 			pgt = NULL;
1263 			if (nested) {
1264 				kvmhv_put_nested(nested);
1265 				nested = NULL;
1266 			}
1267 			p->lpid = kvmhv_nested_next_lpid(kvm, p->lpid);
1268 			p->hdr = 0;
1269 			if (p->lpid < 0)
1270 				break;
1271 		}
1272 		if (!pgt) {
1273 			if (p->lpid == 0) {
1274 				pgt = kvm->arch.pgtable;
1275 			} else {
1276 				nested = kvmhv_get_nested(kvm, p->lpid, false);
1277 				if (!nested) {
1278 					gpa = RADIX_PGTABLE_RANGE;
1279 					continue;
1280 				}
1281 				pgt = nested->shadow_pgtable;
1282 			}
1283 		}
1284 		n = 0;
1285 		if (!p->hdr) {
1286 			if (p->lpid > 0)
1287 				n = scnprintf(p->buf, sizeof(p->buf),
1288 					      "\nNested LPID %d: ", p->lpid);
1289 			n += scnprintf(p->buf + n, sizeof(p->buf) - n,
1290 				      "pgdir: %lx\n", (unsigned long)pgt);
1291 			p->hdr = 1;
1292 			goto copy;
1293 		}
1294 
1295 		pgdp = pgt + pgd_index(gpa);
1296 		pgd = READ_ONCE(*pgdp);
1297 		if (!(pgd_val(pgd) & _PAGE_PRESENT)) {
1298 			gpa = (gpa & PGDIR_MASK) + PGDIR_SIZE;
1299 			continue;
1300 		}
1301 
1302 		pudp = pud_offset(&pgd, gpa);
1303 		pud = READ_ONCE(*pudp);
1304 		if (!(pud_val(pud) & _PAGE_PRESENT)) {
1305 			gpa = (gpa & PUD_MASK) + PUD_SIZE;
1306 			continue;
1307 		}
1308 		if (pud_val(pud) & _PAGE_PTE) {
1309 			pte = pud_val(pud);
1310 			shift = PUD_SHIFT;
1311 			goto leaf;
1312 		}
1313 
1314 		pmdp = pmd_offset(&pud, gpa);
1315 		pmd = READ_ONCE(*pmdp);
1316 		if (!(pmd_val(pmd) & _PAGE_PRESENT)) {
1317 			gpa = (gpa & PMD_MASK) + PMD_SIZE;
1318 			continue;
1319 		}
1320 		if (pmd_val(pmd) & _PAGE_PTE) {
1321 			pte = pmd_val(pmd);
1322 			shift = PMD_SHIFT;
1323 			goto leaf;
1324 		}
1325 
1326 		ptep = pte_offset_kernel(&pmd, gpa);
1327 		pte = pte_val(READ_ONCE(*ptep));
1328 		if (!(pte & _PAGE_PRESENT)) {
1329 			gpa += PAGE_SIZE;
1330 			continue;
1331 		}
1332 		shift = PAGE_SHIFT;
1333 	leaf:
1334 		n = scnprintf(p->buf, sizeof(p->buf),
1335 			      " %lx: %lx %d\n", gpa, pte, shift);
1336 		gpa += 1ul << shift;
1337 	copy:
1338 		p->chars_left = n;
1339 		if (n > len)
1340 			n = len;
1341 		r = copy_to_user(buf, p->buf, n);
1342 		n -= r;
1343 		p->chars_left -= n;
1344 		p->buf_index = n;
1345 		buf += n;
1346 		len -= n;
1347 		ret += n;
1348 		if (r) {
1349 			if (!ret)
1350 				ret = -EFAULT;
1351 			break;
1352 		}
1353 	}
1354 	p->gpa = gpa;
1355 	if (nested)
1356 		kvmhv_put_nested(nested);
1357 
1358  out:
1359 	mutex_unlock(&p->mutex);
1360 	return ret;
1361 }
1362 
1363 static ssize_t debugfs_radix_write(struct file *file, const char __user *buf,
1364 			   size_t len, loff_t *ppos)
1365 {
1366 	return -EACCES;
1367 }
1368 
1369 static const struct file_operations debugfs_radix_fops = {
1370 	.owner	 = THIS_MODULE,
1371 	.open	 = debugfs_radix_open,
1372 	.release = debugfs_radix_release,
1373 	.read	 = debugfs_radix_read,
1374 	.write	 = debugfs_radix_write,
1375 	.llseek	 = generic_file_llseek,
1376 };
1377 
1378 void kvmhv_radix_debugfs_init(struct kvm *kvm)
1379 {
1380 	debugfs_create_file("radix", 0400, kvm->arch.debugfs_dir, kvm,
1381 			    &debugfs_radix_fops);
1382 }
1383 
1384 int kvmppc_radix_init(void)
1385 {
1386 	unsigned long size = sizeof(void *) << RADIX_PTE_INDEX_SIZE;
1387 
1388 	kvm_pte_cache = kmem_cache_create("kvm-pte", size, size, 0, pte_ctor);
1389 	if (!kvm_pte_cache)
1390 		return -ENOMEM;
1391 
1392 	size = sizeof(void *) << RADIX_PMD_INDEX_SIZE;
1393 
1394 	kvm_pmd_cache = kmem_cache_create("kvm-pmd", size, size, 0, pmd_ctor);
1395 	if (!kvm_pmd_cache) {
1396 		kmem_cache_destroy(kvm_pte_cache);
1397 		return -ENOMEM;
1398 	}
1399 
1400 	return 0;
1401 }
1402 
1403 void kvmppc_radix_exit(void)
1404 {
1405 	kmem_cache_destroy(kvm_pte_cache);
1406 	kmem_cache_destroy(kvm_pmd_cache);
1407 }
1408