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 #include <linux/pgtable.h>
15 
16 #include <asm/kvm_ppc.h>
17 #include <asm/kvm_book3s.h>
18 #include <asm/page.h>
19 #include <asm/mmu.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 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 					  (to != NULL) ? __pa(to): 0,
44 					  (from != NULL) ? __pa(from): 0, n);
45 
46 	quadrant = 1;
47 	if (!pid)
48 		quadrant = 2;
49 	if (is_load)
50 		from = (void *) (eaddr | (quadrant << 62));
51 	else
52 		to = (void *) (eaddr | (quadrant << 62));
53 
54 	preempt_disable();
55 
56 	/* switch the lpid first to avoid running host with unallocated pid */
57 	old_lpid = mfspr(SPRN_LPID);
58 	if (old_lpid != lpid)
59 		mtspr(SPRN_LPID, lpid);
60 	if (quadrant == 1) {
61 		old_pid = mfspr(SPRN_PID);
62 		if (old_pid != pid)
63 			mtspr(SPRN_PID, pid);
64 	}
65 	isync();
66 
67 	if (is_load)
68 		ret = copy_from_user_nofault(to, (const void __user *)from, n);
69 	else
70 		ret = copy_to_user_nofault((void __user *)to, from, n);
71 
72 	/* switch the pid first to avoid running host with unallocated pid */
73 	if (quadrant == 1 && pid != old_pid)
74 		mtspr(SPRN_PID, old_pid);
75 	if (lpid != old_lpid)
76 		mtspr(SPRN_LPID, old_lpid);
77 	isync();
78 
79 	preempt_enable();
80 
81 	return ret;
82 }
83 EXPORT_SYMBOL_GPL(__kvmhv_copy_tofrom_guest_radix);
84 
85 static long kvmhv_copy_tofrom_guest_radix(struct kvm_vcpu *vcpu, gva_t eaddr,
86 					  void *to, void *from, unsigned long n)
87 {
88 	int lpid = vcpu->kvm->arch.lpid;
89 	int pid = vcpu->arch.pid;
90 
91 	/* This would cause a data segment intr so don't allow the access */
92 	if (eaddr & (0x3FFUL << 52))
93 		return -EINVAL;
94 
95 	/* Should we be using the nested lpid */
96 	if (vcpu->arch.nested)
97 		lpid = vcpu->arch.nested->shadow_lpid;
98 
99 	/* If accessing quadrant 3 then pid is expected to be 0 */
100 	if (((eaddr >> 62) & 0x3) == 0x3)
101 		pid = 0;
102 
103 	eaddr &= ~(0xFFFUL << 52);
104 
105 	return __kvmhv_copy_tofrom_guest_radix(lpid, pid, eaddr, to, from, n);
106 }
107 
108 long kvmhv_copy_from_guest_radix(struct kvm_vcpu *vcpu, gva_t eaddr, void *to,
109 				 unsigned long n)
110 {
111 	long ret;
112 
113 	ret = kvmhv_copy_tofrom_guest_radix(vcpu, eaddr, to, NULL, n);
114 	if (ret > 0)
115 		memset(to + (n - ret), 0, ret);
116 
117 	return ret;
118 }
119 EXPORT_SYMBOL_GPL(kvmhv_copy_from_guest_radix);
120 
121 long kvmhv_copy_to_guest_radix(struct kvm_vcpu *vcpu, gva_t eaddr, void *from,
122 			       unsigned long n)
123 {
124 	return kvmhv_copy_tofrom_guest_radix(vcpu, eaddr, NULL, from, n);
125 }
126 EXPORT_SYMBOL_GPL(kvmhv_copy_to_guest_radix);
127 
128 int kvmppc_mmu_walk_radix_tree(struct kvm_vcpu *vcpu, gva_t eaddr,
129 			       struct kvmppc_pte *gpte, u64 root,
130 			       u64 *pte_ret_p)
131 {
132 	struct kvm *kvm = vcpu->kvm;
133 	int ret, level, ps;
134 	unsigned long rts, bits, offset, index;
135 	u64 pte, base, gpa;
136 	__be64 rpte;
137 
138 	rts = ((root & RTS1_MASK) >> (RTS1_SHIFT - 3)) |
139 		((root & RTS2_MASK) >> RTS2_SHIFT);
140 	bits = root & RPDS_MASK;
141 	base = root & RPDB_MASK;
142 
143 	offset = rts + 31;
144 
145 	/* Current implementations only support 52-bit space */
146 	if (offset != 52)
147 		return -EINVAL;
148 
149 	/* Walk each level of the radix tree */
150 	for (level = 3; level >= 0; --level) {
151 		u64 addr;
152 		/* Check a valid size */
153 		if (level && bits != p9_supported_radix_bits[level])
154 			return -EINVAL;
155 		if (level == 0 && !(bits == 5 || bits == 9))
156 			return -EINVAL;
157 		offset -= bits;
158 		index = (eaddr >> offset) & ((1UL << bits) - 1);
159 		/* Check that low bits of page table base are zero */
160 		if (base & ((1UL << (bits + 3)) - 1))
161 			return -EINVAL;
162 		/* Read the entry from guest memory */
163 		addr = base + (index * sizeof(rpte));
164 		vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
165 		ret = kvm_read_guest(kvm, addr, &rpte, sizeof(rpte));
166 		srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
167 		if (ret) {
168 			if (pte_ret_p)
169 				*pte_ret_p = addr;
170 			return ret;
171 		}
172 		pte = __be64_to_cpu(rpte);
173 		if (!(pte & _PAGE_PRESENT))
174 			return -ENOENT;
175 		/* Check if a leaf entry */
176 		if (pte & _PAGE_PTE)
177 			break;
178 		/* Get ready to walk the next level */
179 		base = pte & RPDB_MASK;
180 		bits = pte & RPDS_MASK;
181 	}
182 
183 	/* Need a leaf at lowest level; 512GB pages not supported */
184 	if (level < 0 || level == 3)
185 		return -EINVAL;
186 
187 	/* We found a valid leaf PTE */
188 	/* Offset is now log base 2 of the page size */
189 	gpa = pte & 0x01fffffffffff000ul;
190 	if (gpa & ((1ul << offset) - 1))
191 		return -EINVAL;
192 	gpa |= eaddr & ((1ul << offset) - 1);
193 	for (ps = MMU_PAGE_4K; ps < MMU_PAGE_COUNT; ++ps)
194 		if (offset == mmu_psize_defs[ps].shift)
195 			break;
196 	gpte->page_size = ps;
197 	gpte->page_shift = offset;
198 
199 	gpte->eaddr = eaddr;
200 	gpte->raddr = gpa;
201 
202 	/* Work out permissions */
203 	gpte->may_read = !!(pte & _PAGE_READ);
204 	gpte->may_write = !!(pte & _PAGE_WRITE);
205 	gpte->may_execute = !!(pte & _PAGE_EXEC);
206 
207 	gpte->rc = pte & (_PAGE_ACCESSED | _PAGE_DIRTY);
208 
209 	if (pte_ret_p)
210 		*pte_ret_p = pte;
211 
212 	return 0;
213 }
214 
215 /*
216  * Used to walk a partition or process table radix tree in guest memory
217  * Note: We exploit the fact that a partition table and a process
218  * table have the same layout, a partition-scoped page table and a
219  * process-scoped page table have the same layout, and the 2nd
220  * doubleword of a partition table entry has the same layout as
221  * the PTCR register.
222  */
223 int kvmppc_mmu_radix_translate_table(struct kvm_vcpu *vcpu, gva_t eaddr,
224 				     struct kvmppc_pte *gpte, u64 table,
225 				     int table_index, u64 *pte_ret_p)
226 {
227 	struct kvm *kvm = vcpu->kvm;
228 	int ret;
229 	unsigned long size, ptbl, root;
230 	struct prtb_entry entry;
231 
232 	if ((table & PRTS_MASK) > 24)
233 		return -EINVAL;
234 	size = 1ul << ((table & PRTS_MASK) + 12);
235 
236 	/* Is the table big enough to contain this entry? */
237 	if ((table_index * sizeof(entry)) >= size)
238 		return -EINVAL;
239 
240 	/* Read the table to find the root of the radix tree */
241 	ptbl = (table & PRTB_MASK) + (table_index * sizeof(entry));
242 	vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
243 	ret = kvm_read_guest(kvm, ptbl, &entry, sizeof(entry));
244 	srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
245 	if (ret)
246 		return ret;
247 
248 	/* Root is stored in the first double word */
249 	root = be64_to_cpu(entry.prtb0);
250 
251 	return kvmppc_mmu_walk_radix_tree(vcpu, eaddr, gpte, root, pte_ret_p);
252 }
253 
254 int kvmppc_mmu_radix_xlate(struct kvm_vcpu *vcpu, gva_t eaddr,
255 			   struct kvmppc_pte *gpte, bool data, bool iswrite)
256 {
257 	u32 pid;
258 	u64 pte;
259 	int ret;
260 
261 	/* Work out effective PID */
262 	switch (eaddr >> 62) {
263 	case 0:
264 		pid = vcpu->arch.pid;
265 		break;
266 	case 3:
267 		pid = 0;
268 		break;
269 	default:
270 		return -EINVAL;
271 	}
272 
273 	ret = kvmppc_mmu_radix_translate_table(vcpu, eaddr, gpte,
274 				vcpu->kvm->arch.process_table, pid, &pte);
275 	if (ret)
276 		return ret;
277 
278 	/* Check privilege (applies only to process scoped translations) */
279 	if (kvmppc_get_msr(vcpu) & MSR_PR) {
280 		if (pte & _PAGE_PRIVILEGED) {
281 			gpte->may_read = 0;
282 			gpte->may_write = 0;
283 			gpte->may_execute = 0;
284 		}
285 	} else {
286 		if (!(pte & _PAGE_PRIVILEGED)) {
287 			/* Check AMR/IAMR to see if strict mode is in force */
288 			if (vcpu->arch.amr & (1ul << 62))
289 				gpte->may_read = 0;
290 			if (vcpu->arch.amr & (1ul << 63))
291 				gpte->may_write = 0;
292 			if (vcpu->arch.iamr & (1ul << 62))
293 				gpte->may_execute = 0;
294 		}
295 	}
296 
297 	return 0;
298 }
299 
300 void kvmppc_radix_tlbie_page(struct kvm *kvm, unsigned long addr,
301 			     unsigned int pshift, unsigned int lpid)
302 {
303 	unsigned long psize = PAGE_SIZE;
304 	int psi;
305 	long rc;
306 	unsigned long rb;
307 
308 	if (pshift)
309 		psize = 1UL << pshift;
310 	else
311 		pshift = PAGE_SHIFT;
312 
313 	addr &= ~(psize - 1);
314 
315 	if (!kvmhv_on_pseries()) {
316 		radix__flush_tlb_lpid_page(lpid, addr, psize);
317 		return;
318 	}
319 
320 	psi = shift_to_mmu_psize(pshift);
321 	rb = addr | (mmu_get_ap(psi) << PPC_BITLSHIFT(58));
322 	rc = plpar_hcall_norets(H_TLB_INVALIDATE, H_TLBIE_P1_ENC(0, 0, 1),
323 				lpid, rb);
324 	if (rc)
325 		pr_err("KVM: TLB page invalidation hcall failed, rc=%ld\n", rc);
326 }
327 
328 static void kvmppc_radix_flush_pwc(struct kvm *kvm, unsigned int lpid)
329 {
330 	long rc;
331 
332 	if (!kvmhv_on_pseries()) {
333 		radix__flush_pwc_lpid(lpid);
334 		return;
335 	}
336 
337 	rc = plpar_hcall_norets(H_TLB_INVALIDATE, H_TLBIE_P1_ENC(1, 0, 1),
338 				lpid, TLBIEL_INVAL_SET_LPID);
339 	if (rc)
340 		pr_err("KVM: TLB PWC invalidation hcall failed, rc=%ld\n", rc);
341 }
342 
343 static unsigned long kvmppc_radix_update_pte(struct kvm *kvm, pte_t *ptep,
344 				      unsigned long clr, unsigned long set,
345 				      unsigned long addr, unsigned int shift)
346 {
347 	return __radix_pte_update(ptep, clr, set);
348 }
349 
350 static void kvmppc_radix_set_pte_at(struct kvm *kvm, unsigned long addr,
351 			     pte_t *ptep, pte_t pte)
352 {
353 	radix__set_pte_at(kvm->mm, addr, ptep, pte, 0);
354 }
355 
356 static struct kmem_cache *kvm_pte_cache;
357 static struct kmem_cache *kvm_pmd_cache;
358 
359 static pte_t *kvmppc_pte_alloc(void)
360 {
361 	pte_t *pte;
362 
363 	pte = kmem_cache_alloc(kvm_pte_cache, GFP_KERNEL);
364 	/* pmd_populate() will only reference _pa(pte). */
365 	kmemleak_ignore(pte);
366 
367 	return pte;
368 }
369 
370 static void kvmppc_pte_free(pte_t *ptep)
371 {
372 	kmem_cache_free(kvm_pte_cache, ptep);
373 }
374 
375 static pmd_t *kvmppc_pmd_alloc(void)
376 {
377 	pmd_t *pmd;
378 
379 	pmd = kmem_cache_alloc(kvm_pmd_cache, GFP_KERNEL);
380 	/* pud_populate() will only reference _pa(pmd). */
381 	kmemleak_ignore(pmd);
382 
383 	return pmd;
384 }
385 
386 static void kvmppc_pmd_free(pmd_t *pmdp)
387 {
388 	kmem_cache_free(kvm_pmd_cache, pmdp);
389 }
390 
391 /* Called with kvm->mmu_lock held */
392 void kvmppc_unmap_pte(struct kvm *kvm, pte_t *pte, unsigned long gpa,
393 		      unsigned int shift,
394 		      const struct kvm_memory_slot *memslot,
395 		      unsigned int lpid)
396 
397 {
398 	unsigned long old;
399 	unsigned long gfn = gpa >> PAGE_SHIFT;
400 	unsigned long page_size = PAGE_SIZE;
401 	unsigned long hpa;
402 
403 	old = kvmppc_radix_update_pte(kvm, pte, ~0UL, 0, gpa, shift);
404 	kvmppc_radix_tlbie_page(kvm, gpa, shift, lpid);
405 
406 	/* The following only applies to L1 entries */
407 	if (lpid != kvm->arch.lpid)
408 		return;
409 
410 	if (!memslot) {
411 		memslot = gfn_to_memslot(kvm, gfn);
412 		if (!memslot)
413 			return;
414 	}
415 	if (shift) { /* 1GB or 2MB page */
416 		page_size = 1ul << shift;
417 		if (shift == PMD_SHIFT)
418 			kvm->stat.num_2M_pages--;
419 		else if (shift == PUD_SHIFT)
420 			kvm->stat.num_1G_pages--;
421 	}
422 
423 	gpa &= ~(page_size - 1);
424 	hpa = old & PTE_RPN_MASK;
425 	kvmhv_remove_nest_rmap_range(kvm, memslot, gpa, hpa, page_size);
426 
427 	if ((old & _PAGE_DIRTY) && memslot->dirty_bitmap)
428 		kvmppc_update_dirty_map(memslot, gfn, page_size);
429 }
430 
431 /*
432  * kvmppc_free_p?d are used to free existing page tables, and recursively
433  * descend and clear and free children.
434  * Callers are responsible for flushing the PWC.
435  *
436  * When page tables are being unmapped/freed as part of page fault path
437  * (full == false), valid ptes are generally not expected; however, there
438  * is one situation where they arise, which is when dirty page logging is
439  * turned off for a memslot while the VM is running.  The new memslot
440  * becomes visible to page faults before the memslot commit function
441  * gets to flush the memslot, which can lead to a 2MB page mapping being
442  * installed for a guest physical address where there are already 64kB
443  * (or 4kB) mappings (of sub-pages of the same 2MB page).
444  */
445 static void kvmppc_unmap_free_pte(struct kvm *kvm, pte_t *pte, bool full,
446 				  unsigned int lpid)
447 {
448 	if (full) {
449 		memset(pte, 0, sizeof(long) << RADIX_PTE_INDEX_SIZE);
450 	} else {
451 		pte_t *p = pte;
452 		unsigned long it;
453 
454 		for (it = 0; it < PTRS_PER_PTE; ++it, ++p) {
455 			if (pte_val(*p) == 0)
456 				continue;
457 			kvmppc_unmap_pte(kvm, p,
458 					 pte_pfn(*p) << PAGE_SHIFT,
459 					 PAGE_SHIFT, NULL, lpid);
460 		}
461 	}
462 
463 	kvmppc_pte_free(pte);
464 }
465 
466 static void kvmppc_unmap_free_pmd(struct kvm *kvm, pmd_t *pmd, bool full,
467 				  unsigned int lpid)
468 {
469 	unsigned long im;
470 	pmd_t *p = pmd;
471 
472 	for (im = 0; im < PTRS_PER_PMD; ++im, ++p) {
473 		if (!pmd_present(*p))
474 			continue;
475 		if (pmd_is_leaf(*p)) {
476 			if (full) {
477 				pmd_clear(p);
478 			} else {
479 				WARN_ON_ONCE(1);
480 				kvmppc_unmap_pte(kvm, (pte_t *)p,
481 					 pte_pfn(*(pte_t *)p) << PAGE_SHIFT,
482 					 PMD_SHIFT, NULL, lpid);
483 			}
484 		} else {
485 			pte_t *pte;
486 
487 			pte = pte_offset_map(p, 0);
488 			kvmppc_unmap_free_pte(kvm, pte, full, lpid);
489 			pmd_clear(p);
490 		}
491 	}
492 	kvmppc_pmd_free(pmd);
493 }
494 
495 static void kvmppc_unmap_free_pud(struct kvm *kvm, pud_t *pud,
496 				  unsigned int lpid)
497 {
498 	unsigned long iu;
499 	pud_t *p = pud;
500 
501 	for (iu = 0; iu < PTRS_PER_PUD; ++iu, ++p) {
502 		if (!pud_present(*p))
503 			continue;
504 		if (pud_is_leaf(*p)) {
505 			pud_clear(p);
506 		} else {
507 			pmd_t *pmd;
508 
509 			pmd = pmd_offset(p, 0);
510 			kvmppc_unmap_free_pmd(kvm, pmd, true, lpid);
511 			pud_clear(p);
512 		}
513 	}
514 	pud_free(kvm->mm, pud);
515 }
516 
517 void kvmppc_free_pgtable_radix(struct kvm *kvm, pgd_t *pgd, unsigned int lpid)
518 {
519 	unsigned long ig;
520 
521 	for (ig = 0; ig < PTRS_PER_PGD; ++ig, ++pgd) {
522 		p4d_t *p4d = p4d_offset(pgd, 0);
523 		pud_t *pud;
524 
525 		if (!p4d_present(*p4d))
526 			continue;
527 		pud = pud_offset(p4d, 0);
528 		kvmppc_unmap_free_pud(kvm, pud, lpid);
529 		p4d_clear(p4d);
530 	}
531 }
532 
533 void kvmppc_free_radix(struct kvm *kvm)
534 {
535 	if (kvm->arch.pgtable) {
536 		kvmppc_free_pgtable_radix(kvm, kvm->arch.pgtable,
537 					  kvm->arch.lpid);
538 		pgd_free(kvm->mm, kvm->arch.pgtable);
539 		kvm->arch.pgtable = NULL;
540 	}
541 }
542 
543 static void kvmppc_unmap_free_pmd_entry_table(struct kvm *kvm, pmd_t *pmd,
544 					unsigned long gpa, unsigned int lpid)
545 {
546 	pte_t *pte = pte_offset_kernel(pmd, 0);
547 
548 	/*
549 	 * Clearing the pmd entry then flushing the PWC ensures that the pte
550 	 * page no longer be cached by the MMU, so can be freed without
551 	 * flushing the PWC again.
552 	 */
553 	pmd_clear(pmd);
554 	kvmppc_radix_flush_pwc(kvm, lpid);
555 
556 	kvmppc_unmap_free_pte(kvm, pte, false, lpid);
557 }
558 
559 static void kvmppc_unmap_free_pud_entry_table(struct kvm *kvm, pud_t *pud,
560 					unsigned long gpa, unsigned int lpid)
561 {
562 	pmd_t *pmd = pmd_offset(pud, 0);
563 
564 	/*
565 	 * Clearing the pud entry then flushing the PWC ensures that the pmd
566 	 * page and any children pte pages will no longer be cached by the MMU,
567 	 * so can be freed without flushing the PWC again.
568 	 */
569 	pud_clear(pud);
570 	kvmppc_radix_flush_pwc(kvm, lpid);
571 
572 	kvmppc_unmap_free_pmd(kvm, pmd, false, lpid);
573 }
574 
575 /*
576  * There are a number of bits which may differ between different faults to
577  * the same partition scope entry. RC bits, in the course of cleaning and
578  * aging. And the write bit can change, either the access could have been
579  * upgraded, or a read fault could happen concurrently with a write fault
580  * that sets those bits first.
581  */
582 #define PTE_BITS_MUST_MATCH (~(_PAGE_WRITE | _PAGE_DIRTY | _PAGE_ACCESSED))
583 
584 int kvmppc_create_pte(struct kvm *kvm, pgd_t *pgtable, pte_t pte,
585 		      unsigned long gpa, unsigned int level,
586 		      unsigned long mmu_seq, unsigned int lpid,
587 		      unsigned long *rmapp, struct rmap_nested **n_rmap)
588 {
589 	pgd_t *pgd;
590 	p4d_t *p4d;
591 	pud_t *pud, *new_pud = NULL;
592 	pmd_t *pmd, *new_pmd = NULL;
593 	pte_t *ptep, *new_ptep = NULL;
594 	int ret;
595 
596 	/* Traverse the guest's 2nd-level tree, allocate new levels needed */
597 	pgd = pgtable + pgd_index(gpa);
598 	p4d = p4d_offset(pgd, gpa);
599 
600 	pud = NULL;
601 	if (p4d_present(*p4d))
602 		pud = pud_offset(p4d, gpa);
603 	else
604 		new_pud = pud_alloc_one(kvm->mm, gpa);
605 
606 	pmd = NULL;
607 	if (pud && pud_present(*pud) && !pud_is_leaf(*pud))
608 		pmd = pmd_offset(pud, gpa);
609 	else if (level <= 1)
610 		new_pmd = kvmppc_pmd_alloc();
611 
612 	if (level == 0 && !(pmd && pmd_present(*pmd) && !pmd_is_leaf(*pmd)))
613 		new_ptep = kvmppc_pte_alloc();
614 
615 	/* Check if we might have been invalidated; let the guest retry if so */
616 	spin_lock(&kvm->mmu_lock);
617 	ret = -EAGAIN;
618 	if (mmu_notifier_retry(kvm, mmu_seq))
619 		goto out_unlock;
620 
621 	/* Now traverse again under the lock and change the tree */
622 	ret = -ENOMEM;
623 	if (p4d_none(*p4d)) {
624 		if (!new_pud)
625 			goto out_unlock;
626 		p4d_populate(kvm->mm, p4d, new_pud);
627 		new_pud = NULL;
628 	}
629 	pud = pud_offset(p4d, gpa);
630 	if (pud_is_leaf(*pud)) {
631 		unsigned long hgpa = gpa & PUD_MASK;
632 
633 		/* Check if we raced and someone else has set the same thing */
634 		if (level == 2) {
635 			if (pud_raw(*pud) == pte_raw(pte)) {
636 				ret = 0;
637 				goto out_unlock;
638 			}
639 			/* Valid 1GB page here already, add our extra bits */
640 			WARN_ON_ONCE((pud_val(*pud) ^ pte_val(pte)) &
641 							PTE_BITS_MUST_MATCH);
642 			kvmppc_radix_update_pte(kvm, (pte_t *)pud,
643 					      0, pte_val(pte), hgpa, PUD_SHIFT);
644 			ret = 0;
645 			goto out_unlock;
646 		}
647 		/*
648 		 * If we raced with another CPU which has just put
649 		 * a 1GB pte in after we saw a pmd page, try again.
650 		 */
651 		if (!new_pmd) {
652 			ret = -EAGAIN;
653 			goto out_unlock;
654 		}
655 		/* Valid 1GB page here already, remove it */
656 		kvmppc_unmap_pte(kvm, (pte_t *)pud, hgpa, PUD_SHIFT, NULL,
657 				 lpid);
658 	}
659 	if (level == 2) {
660 		if (!pud_none(*pud)) {
661 			/*
662 			 * There's a page table page here, but we wanted to
663 			 * install a large page, so remove and free the page
664 			 * table page.
665 			 */
666 			kvmppc_unmap_free_pud_entry_table(kvm, pud, gpa, lpid);
667 		}
668 		kvmppc_radix_set_pte_at(kvm, gpa, (pte_t *)pud, pte);
669 		if (rmapp && n_rmap)
670 			kvmhv_insert_nest_rmap(kvm, rmapp, n_rmap);
671 		ret = 0;
672 		goto out_unlock;
673 	}
674 	if (pud_none(*pud)) {
675 		if (!new_pmd)
676 			goto out_unlock;
677 		pud_populate(kvm->mm, pud, new_pmd);
678 		new_pmd = NULL;
679 	}
680 	pmd = pmd_offset(pud, gpa);
681 	if (pmd_is_leaf(*pmd)) {
682 		unsigned long lgpa = gpa & PMD_MASK;
683 
684 		/* Check if we raced and someone else has set the same thing */
685 		if (level == 1) {
686 			if (pmd_raw(*pmd) == pte_raw(pte)) {
687 				ret = 0;
688 				goto out_unlock;
689 			}
690 			/* Valid 2MB page here already, add our extra bits */
691 			WARN_ON_ONCE((pmd_val(*pmd) ^ pte_val(pte)) &
692 							PTE_BITS_MUST_MATCH);
693 			kvmppc_radix_update_pte(kvm, pmdp_ptep(pmd),
694 					0, pte_val(pte), lgpa, PMD_SHIFT);
695 			ret = 0;
696 			goto out_unlock;
697 		}
698 
699 		/*
700 		 * If we raced with another CPU which has just put
701 		 * a 2MB pte in after we saw a pte page, try again.
702 		 */
703 		if (!new_ptep) {
704 			ret = -EAGAIN;
705 			goto out_unlock;
706 		}
707 		/* Valid 2MB page here already, remove it */
708 		kvmppc_unmap_pte(kvm, pmdp_ptep(pmd), lgpa, PMD_SHIFT, NULL,
709 				 lpid);
710 	}
711 	if (level == 1) {
712 		if (!pmd_none(*pmd)) {
713 			/*
714 			 * There's a page table page here, but we wanted to
715 			 * install a large page, so remove and free the page
716 			 * table page.
717 			 */
718 			kvmppc_unmap_free_pmd_entry_table(kvm, pmd, gpa, lpid);
719 		}
720 		kvmppc_radix_set_pte_at(kvm, gpa, pmdp_ptep(pmd), pte);
721 		if (rmapp && n_rmap)
722 			kvmhv_insert_nest_rmap(kvm, rmapp, n_rmap);
723 		ret = 0;
724 		goto out_unlock;
725 	}
726 	if (pmd_none(*pmd)) {
727 		if (!new_ptep)
728 			goto out_unlock;
729 		pmd_populate(kvm->mm, pmd, new_ptep);
730 		new_ptep = NULL;
731 	}
732 	ptep = pte_offset_kernel(pmd, gpa);
733 	if (pte_present(*ptep)) {
734 		/* Check if someone else set the same thing */
735 		if (pte_raw(*ptep) == pte_raw(pte)) {
736 			ret = 0;
737 			goto out_unlock;
738 		}
739 		/* Valid page here already, add our extra bits */
740 		WARN_ON_ONCE((pte_val(*ptep) ^ pte_val(pte)) &
741 							PTE_BITS_MUST_MATCH);
742 		kvmppc_radix_update_pte(kvm, ptep, 0, pte_val(pte), gpa, 0);
743 		ret = 0;
744 		goto out_unlock;
745 	}
746 	kvmppc_radix_set_pte_at(kvm, gpa, ptep, pte);
747 	if (rmapp && n_rmap)
748 		kvmhv_insert_nest_rmap(kvm, rmapp, n_rmap);
749 	ret = 0;
750 
751  out_unlock:
752 	spin_unlock(&kvm->mmu_lock);
753 	if (new_pud)
754 		pud_free(kvm->mm, new_pud);
755 	if (new_pmd)
756 		kvmppc_pmd_free(new_pmd);
757 	if (new_ptep)
758 		kvmppc_pte_free(new_ptep);
759 	return ret;
760 }
761 
762 bool kvmppc_hv_handle_set_rc(struct kvm *kvm, bool nested, bool writing,
763 			     unsigned long gpa, unsigned int lpid)
764 {
765 	unsigned long pgflags;
766 	unsigned int shift;
767 	pte_t *ptep;
768 
769 	/*
770 	 * Need to set an R or C bit in the 2nd-level tables;
771 	 * since we are just helping out the hardware here,
772 	 * it is sufficient to do what the hardware does.
773 	 */
774 	pgflags = _PAGE_ACCESSED;
775 	if (writing)
776 		pgflags |= _PAGE_DIRTY;
777 
778 	if (nested)
779 		ptep = find_kvm_nested_guest_pte(kvm, lpid, gpa, &shift);
780 	else
781 		ptep = find_kvm_secondary_pte(kvm, gpa, &shift);
782 
783 	if (ptep && pte_present(*ptep) && (!writing || pte_write(*ptep))) {
784 		kvmppc_radix_update_pte(kvm, ptep, 0, pgflags, gpa, shift);
785 		return true;
786 	}
787 	return false;
788 }
789 
790 int kvmppc_book3s_instantiate_page(struct kvm_vcpu *vcpu,
791 				   unsigned long gpa,
792 				   struct kvm_memory_slot *memslot,
793 				   bool writing, bool kvm_ro,
794 				   pte_t *inserted_pte, unsigned int *levelp)
795 {
796 	struct kvm *kvm = vcpu->kvm;
797 	struct page *page = NULL;
798 	unsigned long mmu_seq;
799 	unsigned long hva, gfn = gpa >> PAGE_SHIFT;
800 	bool upgrade_write = false;
801 	bool *upgrade_p = &upgrade_write;
802 	pte_t pte, *ptep;
803 	unsigned int shift, level;
804 	int ret;
805 	bool large_enable;
806 
807 	/* used to check for invalidations in progress */
808 	mmu_seq = kvm->mmu_notifier_seq;
809 	smp_rmb();
810 
811 	/*
812 	 * Do a fast check first, since __gfn_to_pfn_memslot doesn't
813 	 * do it with !atomic && !async, which is how we call it.
814 	 * We always ask for write permission since the common case
815 	 * is that the page is writable.
816 	 */
817 	hva = gfn_to_hva_memslot(memslot, gfn);
818 	if (!kvm_ro && get_user_page_fast_only(hva, FOLL_WRITE, &page)) {
819 		upgrade_write = true;
820 	} else {
821 		unsigned long pfn;
822 
823 		/* Call KVM generic code to do the slow-path check */
824 		pfn = __gfn_to_pfn_memslot(memslot, gfn, false, NULL,
825 					   writing, upgrade_p);
826 		if (is_error_noslot_pfn(pfn))
827 			return -EFAULT;
828 		page = NULL;
829 		if (pfn_valid(pfn)) {
830 			page = pfn_to_page(pfn);
831 			if (PageReserved(page))
832 				page = NULL;
833 		}
834 	}
835 
836 	/*
837 	 * Read the PTE from the process' radix tree and use that
838 	 * so we get the shift and attribute bits.
839 	 */
840 	spin_lock(&kvm->mmu_lock);
841 	ptep = find_kvm_host_pte(kvm, mmu_seq, hva, &shift);
842 	pte = __pte(0);
843 	if (ptep)
844 		pte = READ_ONCE(*ptep);
845 	spin_unlock(&kvm->mmu_lock);
846 	/*
847 	 * If the PTE disappeared temporarily due to a THP
848 	 * collapse, just return and let the guest try again.
849 	 */
850 	if (!pte_present(pte)) {
851 		if (page)
852 			put_page(page);
853 		return RESUME_GUEST;
854 	}
855 
856 	/* If we're logging dirty pages, always map single pages */
857 	large_enable = !(memslot->flags & KVM_MEM_LOG_DIRTY_PAGES);
858 
859 	/* Get pte level from shift/size */
860 	if (large_enable && shift == PUD_SHIFT &&
861 	    (gpa & (PUD_SIZE - PAGE_SIZE)) ==
862 	    (hva & (PUD_SIZE - PAGE_SIZE))) {
863 		level = 2;
864 	} else if (large_enable && shift == PMD_SHIFT &&
865 		   (gpa & (PMD_SIZE - PAGE_SIZE)) ==
866 		   (hva & (PMD_SIZE - PAGE_SIZE))) {
867 		level = 1;
868 	} else {
869 		level = 0;
870 		if (shift > PAGE_SHIFT) {
871 			/*
872 			 * If the pte maps more than one page, bring over
873 			 * bits from the virtual address to get the real
874 			 * address of the specific single page we want.
875 			 */
876 			unsigned long rpnmask = (1ul << shift) - PAGE_SIZE;
877 			pte = __pte(pte_val(pte) | (hva & rpnmask));
878 		}
879 	}
880 
881 	pte = __pte(pte_val(pte) | _PAGE_EXEC | _PAGE_ACCESSED);
882 	if (writing || upgrade_write) {
883 		if (pte_val(pte) & _PAGE_WRITE)
884 			pte = __pte(pte_val(pte) | _PAGE_DIRTY);
885 	} else {
886 		pte = __pte(pte_val(pte) & ~(_PAGE_WRITE | _PAGE_DIRTY));
887 	}
888 
889 	/* Allocate space in the tree and write the PTE */
890 	ret = kvmppc_create_pte(kvm, kvm->arch.pgtable, pte, gpa, level,
891 				mmu_seq, kvm->arch.lpid, NULL, NULL);
892 	if (inserted_pte)
893 		*inserted_pte = pte;
894 	if (levelp)
895 		*levelp = level;
896 
897 	if (page) {
898 		if (!ret && (pte_val(pte) & _PAGE_WRITE))
899 			set_page_dirty_lock(page);
900 		put_page(page);
901 	}
902 
903 	/* Increment number of large pages if we (successfully) inserted one */
904 	if (!ret) {
905 		if (level == 1)
906 			kvm->stat.num_2M_pages++;
907 		else if (level == 2)
908 			kvm->stat.num_1G_pages++;
909 	}
910 
911 	return ret;
912 }
913 
914 int kvmppc_book3s_radix_page_fault(struct kvm_vcpu *vcpu,
915 				   unsigned long ea, unsigned long dsisr)
916 {
917 	struct kvm *kvm = vcpu->kvm;
918 	unsigned long gpa, gfn;
919 	struct kvm_memory_slot *memslot;
920 	long ret;
921 	bool writing = !!(dsisr & DSISR_ISSTORE);
922 	bool kvm_ro = false;
923 
924 	/* Check for unusual errors */
925 	if (dsisr & DSISR_UNSUPP_MMU) {
926 		pr_err("KVM: Got unsupported MMU fault\n");
927 		return -EFAULT;
928 	}
929 	if (dsisr & DSISR_BADACCESS) {
930 		/* Reflect to the guest as DSI */
931 		pr_err("KVM: Got radix HV page fault with DSISR=%lx\n", dsisr);
932 		kvmppc_core_queue_data_storage(vcpu, ea, dsisr);
933 		return RESUME_GUEST;
934 	}
935 
936 	/* Translate the logical address */
937 	gpa = vcpu->arch.fault_gpa & ~0xfffUL;
938 	gpa &= ~0xF000000000000000ul;
939 	gfn = gpa >> PAGE_SHIFT;
940 	if (!(dsisr & DSISR_PRTABLE_FAULT))
941 		gpa |= ea & 0xfff;
942 
943 	if (kvm->arch.secure_guest & KVMPPC_SECURE_INIT_DONE)
944 		return kvmppc_send_page_to_uv(kvm, gfn);
945 
946 	/* Get the corresponding memslot */
947 	memslot = gfn_to_memslot(kvm, gfn);
948 
949 	/* No memslot means it's an emulated MMIO region */
950 	if (!memslot || (memslot->flags & KVM_MEMSLOT_INVALID)) {
951 		if (dsisr & (DSISR_PRTABLE_FAULT | DSISR_BADACCESS |
952 			     DSISR_SET_RC)) {
953 			/*
954 			 * Bad address in guest page table tree, or other
955 			 * unusual error - reflect it to the guest as DSI.
956 			 */
957 			kvmppc_core_queue_data_storage(vcpu, ea, dsisr);
958 			return RESUME_GUEST;
959 		}
960 		return kvmppc_hv_emulate_mmio(vcpu, gpa, ea, writing);
961 	}
962 
963 	if (memslot->flags & KVM_MEM_READONLY) {
964 		if (writing) {
965 			/* give the guest a DSI */
966 			kvmppc_core_queue_data_storage(vcpu, ea, DSISR_ISSTORE |
967 						       DSISR_PROTFAULT);
968 			return RESUME_GUEST;
969 		}
970 		kvm_ro = true;
971 	}
972 
973 	/* Failed to set the reference/change bits */
974 	if (dsisr & DSISR_SET_RC) {
975 		spin_lock(&kvm->mmu_lock);
976 		if (kvmppc_hv_handle_set_rc(kvm, false, writing,
977 					    gpa, kvm->arch.lpid))
978 			dsisr &= ~DSISR_SET_RC;
979 		spin_unlock(&kvm->mmu_lock);
980 
981 		if (!(dsisr & (DSISR_BAD_FAULT_64S | DSISR_NOHPTE |
982 			       DSISR_PROTFAULT | DSISR_SET_RC)))
983 			return RESUME_GUEST;
984 	}
985 
986 	/* Try to insert a pte */
987 	ret = kvmppc_book3s_instantiate_page(vcpu, gpa, memslot, writing,
988 					     kvm_ro, NULL, NULL);
989 
990 	if (ret == 0 || ret == -EAGAIN)
991 		ret = RESUME_GUEST;
992 	return ret;
993 }
994 
995 /* Called with kvm->mmu_lock held */
996 int kvm_unmap_radix(struct kvm *kvm, struct kvm_memory_slot *memslot,
997 		    unsigned long gfn)
998 {
999 	pte_t *ptep;
1000 	unsigned long gpa = gfn << PAGE_SHIFT;
1001 	unsigned int shift;
1002 
1003 	if (kvm->arch.secure_guest & KVMPPC_SECURE_INIT_DONE) {
1004 		uv_page_inval(kvm->arch.lpid, gpa, PAGE_SHIFT);
1005 		return 0;
1006 	}
1007 
1008 	ptep = find_kvm_secondary_pte(kvm, gpa, &shift);
1009 	if (ptep && pte_present(*ptep))
1010 		kvmppc_unmap_pte(kvm, ptep, gpa, shift, memslot,
1011 				 kvm->arch.lpid);
1012 	return 0;
1013 }
1014 
1015 /* Called with kvm->mmu_lock held */
1016 int kvm_age_radix(struct kvm *kvm, struct kvm_memory_slot *memslot,
1017 		  unsigned long gfn)
1018 {
1019 	pte_t *ptep;
1020 	unsigned long gpa = gfn << PAGE_SHIFT;
1021 	unsigned int shift;
1022 	int ref = 0;
1023 	unsigned long old, *rmapp;
1024 
1025 	if (kvm->arch.secure_guest & KVMPPC_SECURE_INIT_DONE)
1026 		return ref;
1027 
1028 	ptep = find_kvm_secondary_pte(kvm, gpa, &shift);
1029 	if (ptep && pte_present(*ptep) && pte_young(*ptep)) {
1030 		old = kvmppc_radix_update_pte(kvm, ptep, _PAGE_ACCESSED, 0,
1031 					      gpa, shift);
1032 		/* XXX need to flush tlb here? */
1033 		/* Also clear bit in ptes in shadow pgtable for nested guests */
1034 		rmapp = &memslot->arch.rmap[gfn - memslot->base_gfn];
1035 		kvmhv_update_nest_rmap_rc_list(kvm, rmapp, _PAGE_ACCESSED, 0,
1036 					       old & PTE_RPN_MASK,
1037 					       1UL << shift);
1038 		ref = 1;
1039 	}
1040 	return ref;
1041 }
1042 
1043 /* Called with kvm->mmu_lock held */
1044 int kvm_test_age_radix(struct kvm *kvm, struct kvm_memory_slot *memslot,
1045 		       unsigned long gfn)
1046 {
1047 	pte_t *ptep;
1048 	unsigned long gpa = gfn << PAGE_SHIFT;
1049 	unsigned int shift;
1050 	int ref = 0;
1051 
1052 	if (kvm->arch.secure_guest & KVMPPC_SECURE_INIT_DONE)
1053 		return ref;
1054 
1055 	ptep = find_kvm_secondary_pte(kvm, gpa, &shift);
1056 	if (ptep && pte_present(*ptep) && pte_young(*ptep))
1057 		ref = 1;
1058 	return ref;
1059 }
1060 
1061 /* Returns the number of PAGE_SIZE pages that are dirty */
1062 static int kvm_radix_test_clear_dirty(struct kvm *kvm,
1063 				struct kvm_memory_slot *memslot, int pagenum)
1064 {
1065 	unsigned long gfn = memslot->base_gfn + pagenum;
1066 	unsigned long gpa = gfn << PAGE_SHIFT;
1067 	pte_t *ptep, pte;
1068 	unsigned int shift;
1069 	int ret = 0;
1070 	unsigned long old, *rmapp;
1071 
1072 	if (kvm->arch.secure_guest & KVMPPC_SECURE_INIT_DONE)
1073 		return ret;
1074 
1075 	/*
1076 	 * For performance reasons we don't hold kvm->mmu_lock while walking the
1077 	 * partition scoped table.
1078 	 */
1079 	ptep = find_kvm_secondary_pte_unlocked(kvm, gpa, &shift);
1080 	if (!ptep)
1081 		return 0;
1082 
1083 	pte = READ_ONCE(*ptep);
1084 	if (pte_present(pte) && pte_dirty(pte)) {
1085 		spin_lock(&kvm->mmu_lock);
1086 		/*
1087 		 * Recheck the pte again
1088 		 */
1089 		if (pte_val(pte) != pte_val(*ptep)) {
1090 			/*
1091 			 * We have KVM_MEM_LOG_DIRTY_PAGES enabled. Hence we can
1092 			 * only find PAGE_SIZE pte entries here. We can continue
1093 			 * to use the pte addr returned by above page table
1094 			 * walk.
1095 			 */
1096 			if (!pte_present(*ptep) || !pte_dirty(*ptep)) {
1097 				spin_unlock(&kvm->mmu_lock);
1098 				return 0;
1099 			}
1100 		}
1101 
1102 		ret = 1;
1103 		VM_BUG_ON(shift);
1104 		old = kvmppc_radix_update_pte(kvm, ptep, _PAGE_DIRTY, 0,
1105 					      gpa, shift);
1106 		kvmppc_radix_tlbie_page(kvm, gpa, shift, kvm->arch.lpid);
1107 		/* Also clear bit in ptes in shadow pgtable for nested guests */
1108 		rmapp = &memslot->arch.rmap[gfn - memslot->base_gfn];
1109 		kvmhv_update_nest_rmap_rc_list(kvm, rmapp, _PAGE_DIRTY, 0,
1110 					       old & PTE_RPN_MASK,
1111 					       1UL << shift);
1112 		spin_unlock(&kvm->mmu_lock);
1113 	}
1114 	return ret;
1115 }
1116 
1117 long kvmppc_hv_get_dirty_log_radix(struct kvm *kvm,
1118 			struct kvm_memory_slot *memslot, unsigned long *map)
1119 {
1120 	unsigned long i, j;
1121 	int npages;
1122 
1123 	for (i = 0; i < memslot->npages; i = j) {
1124 		npages = kvm_radix_test_clear_dirty(kvm, memslot, i);
1125 
1126 		/*
1127 		 * Note that if npages > 0 then i must be a multiple of npages,
1128 		 * since huge pages are only used to back the guest at guest
1129 		 * real addresses that are a multiple of their size.
1130 		 * Since we have at most one PTE covering any given guest
1131 		 * real address, if npages > 1 we can skip to i + npages.
1132 		 */
1133 		j = i + 1;
1134 		if (npages) {
1135 			set_dirty_bits(map, i, npages);
1136 			j = i + npages;
1137 		}
1138 	}
1139 	return 0;
1140 }
1141 
1142 void kvmppc_radix_flush_memslot(struct kvm *kvm,
1143 				const struct kvm_memory_slot *memslot)
1144 {
1145 	unsigned long n;
1146 	pte_t *ptep;
1147 	unsigned long gpa;
1148 	unsigned int shift;
1149 
1150 	if (kvm->arch.secure_guest & KVMPPC_SECURE_INIT_START)
1151 		kvmppc_uvmem_drop_pages(memslot, kvm, true);
1152 
1153 	if (kvm->arch.secure_guest & KVMPPC_SECURE_INIT_DONE)
1154 		return;
1155 
1156 	gpa = memslot->base_gfn << PAGE_SHIFT;
1157 	spin_lock(&kvm->mmu_lock);
1158 	for (n = memslot->npages; n; --n) {
1159 		ptep = find_kvm_secondary_pte(kvm, gpa, &shift);
1160 		if (ptep && pte_present(*ptep))
1161 			kvmppc_unmap_pte(kvm, ptep, gpa, shift, memslot,
1162 					 kvm->arch.lpid);
1163 		gpa += PAGE_SIZE;
1164 	}
1165 	/*
1166 	 * Increase the mmu notifier sequence number to prevent any page
1167 	 * fault that read the memslot earlier from writing a PTE.
1168 	 */
1169 	kvm->mmu_notifier_seq++;
1170 	spin_unlock(&kvm->mmu_lock);
1171 }
1172 
1173 static void add_rmmu_ap_encoding(struct kvm_ppc_rmmu_info *info,
1174 				 int psize, int *indexp)
1175 {
1176 	if (!mmu_psize_defs[psize].shift)
1177 		return;
1178 	info->ap_encodings[*indexp] = mmu_psize_defs[psize].shift |
1179 		(mmu_psize_defs[psize].ap << 29);
1180 	++(*indexp);
1181 }
1182 
1183 int kvmhv_get_rmmu_info(struct kvm *kvm, struct kvm_ppc_rmmu_info *info)
1184 {
1185 	int i;
1186 
1187 	if (!radix_enabled())
1188 		return -EINVAL;
1189 	memset(info, 0, sizeof(*info));
1190 
1191 	/* 4k page size */
1192 	info->geometries[0].page_shift = 12;
1193 	info->geometries[0].level_bits[0] = 9;
1194 	for (i = 1; i < 4; ++i)
1195 		info->geometries[0].level_bits[i] = p9_supported_radix_bits[i];
1196 	/* 64k page size */
1197 	info->geometries[1].page_shift = 16;
1198 	for (i = 0; i < 4; ++i)
1199 		info->geometries[1].level_bits[i] = p9_supported_radix_bits[i];
1200 
1201 	i = 0;
1202 	add_rmmu_ap_encoding(info, MMU_PAGE_4K, &i);
1203 	add_rmmu_ap_encoding(info, MMU_PAGE_64K, &i);
1204 	add_rmmu_ap_encoding(info, MMU_PAGE_2M, &i);
1205 	add_rmmu_ap_encoding(info, MMU_PAGE_1G, &i);
1206 
1207 	return 0;
1208 }
1209 
1210 int kvmppc_init_vm_radix(struct kvm *kvm)
1211 {
1212 	kvm->arch.pgtable = pgd_alloc(kvm->mm);
1213 	if (!kvm->arch.pgtable)
1214 		return -ENOMEM;
1215 	return 0;
1216 }
1217 
1218 static void pte_ctor(void *addr)
1219 {
1220 	memset(addr, 0, RADIX_PTE_TABLE_SIZE);
1221 }
1222 
1223 static void pmd_ctor(void *addr)
1224 {
1225 	memset(addr, 0, RADIX_PMD_TABLE_SIZE);
1226 }
1227 
1228 struct debugfs_radix_state {
1229 	struct kvm	*kvm;
1230 	struct mutex	mutex;
1231 	unsigned long	gpa;
1232 	int		lpid;
1233 	int		chars_left;
1234 	int		buf_index;
1235 	char		buf[128];
1236 	u8		hdr;
1237 };
1238 
1239 static int debugfs_radix_open(struct inode *inode, struct file *file)
1240 {
1241 	struct kvm *kvm = inode->i_private;
1242 	struct debugfs_radix_state *p;
1243 
1244 	p = kzalloc(sizeof(*p), GFP_KERNEL);
1245 	if (!p)
1246 		return -ENOMEM;
1247 
1248 	kvm_get_kvm(kvm);
1249 	p->kvm = kvm;
1250 	mutex_init(&p->mutex);
1251 	file->private_data = p;
1252 
1253 	return nonseekable_open(inode, file);
1254 }
1255 
1256 static int debugfs_radix_release(struct inode *inode, struct file *file)
1257 {
1258 	struct debugfs_radix_state *p = file->private_data;
1259 
1260 	kvm_put_kvm(p->kvm);
1261 	kfree(p);
1262 	return 0;
1263 }
1264 
1265 static ssize_t debugfs_radix_read(struct file *file, char __user *buf,
1266 				 size_t len, loff_t *ppos)
1267 {
1268 	struct debugfs_radix_state *p = file->private_data;
1269 	ssize_t ret, r;
1270 	unsigned long n;
1271 	struct kvm *kvm;
1272 	unsigned long gpa;
1273 	pgd_t *pgt;
1274 	struct kvm_nested_guest *nested;
1275 	pgd_t *pgdp;
1276 	p4d_t p4d, *p4dp;
1277 	pud_t pud, *pudp;
1278 	pmd_t pmd, *pmdp;
1279 	pte_t *ptep;
1280 	int shift;
1281 	unsigned long pte;
1282 
1283 	kvm = p->kvm;
1284 	if (!kvm_is_radix(kvm))
1285 		return 0;
1286 
1287 	ret = mutex_lock_interruptible(&p->mutex);
1288 	if (ret)
1289 		return ret;
1290 
1291 	if (p->chars_left) {
1292 		n = p->chars_left;
1293 		if (n > len)
1294 			n = len;
1295 		r = copy_to_user(buf, p->buf + p->buf_index, n);
1296 		n -= r;
1297 		p->chars_left -= n;
1298 		p->buf_index += n;
1299 		buf += n;
1300 		len -= n;
1301 		ret = n;
1302 		if (r) {
1303 			if (!n)
1304 				ret = -EFAULT;
1305 			goto out;
1306 		}
1307 	}
1308 
1309 	gpa = p->gpa;
1310 	nested = NULL;
1311 	pgt = NULL;
1312 	while (len != 0 && p->lpid >= 0) {
1313 		if (gpa >= RADIX_PGTABLE_RANGE) {
1314 			gpa = 0;
1315 			pgt = NULL;
1316 			if (nested) {
1317 				kvmhv_put_nested(nested);
1318 				nested = NULL;
1319 			}
1320 			p->lpid = kvmhv_nested_next_lpid(kvm, p->lpid);
1321 			p->hdr = 0;
1322 			if (p->lpid < 0)
1323 				break;
1324 		}
1325 		if (!pgt) {
1326 			if (p->lpid == 0) {
1327 				pgt = kvm->arch.pgtable;
1328 			} else {
1329 				nested = kvmhv_get_nested(kvm, p->lpid, false);
1330 				if (!nested) {
1331 					gpa = RADIX_PGTABLE_RANGE;
1332 					continue;
1333 				}
1334 				pgt = nested->shadow_pgtable;
1335 			}
1336 		}
1337 		n = 0;
1338 		if (!p->hdr) {
1339 			if (p->lpid > 0)
1340 				n = scnprintf(p->buf, sizeof(p->buf),
1341 					      "\nNested LPID %d: ", p->lpid);
1342 			n += scnprintf(p->buf + n, sizeof(p->buf) - n,
1343 				      "pgdir: %lx\n", (unsigned long)pgt);
1344 			p->hdr = 1;
1345 			goto copy;
1346 		}
1347 
1348 		pgdp = pgt + pgd_index(gpa);
1349 		p4dp = p4d_offset(pgdp, gpa);
1350 		p4d = READ_ONCE(*p4dp);
1351 		if (!(p4d_val(p4d) & _PAGE_PRESENT)) {
1352 			gpa = (gpa & P4D_MASK) + P4D_SIZE;
1353 			continue;
1354 		}
1355 
1356 		pudp = pud_offset(&p4d, gpa);
1357 		pud = READ_ONCE(*pudp);
1358 		if (!(pud_val(pud) & _PAGE_PRESENT)) {
1359 			gpa = (gpa & PUD_MASK) + PUD_SIZE;
1360 			continue;
1361 		}
1362 		if (pud_val(pud) & _PAGE_PTE) {
1363 			pte = pud_val(pud);
1364 			shift = PUD_SHIFT;
1365 			goto leaf;
1366 		}
1367 
1368 		pmdp = pmd_offset(&pud, gpa);
1369 		pmd = READ_ONCE(*pmdp);
1370 		if (!(pmd_val(pmd) & _PAGE_PRESENT)) {
1371 			gpa = (gpa & PMD_MASK) + PMD_SIZE;
1372 			continue;
1373 		}
1374 		if (pmd_val(pmd) & _PAGE_PTE) {
1375 			pte = pmd_val(pmd);
1376 			shift = PMD_SHIFT;
1377 			goto leaf;
1378 		}
1379 
1380 		ptep = pte_offset_kernel(&pmd, gpa);
1381 		pte = pte_val(READ_ONCE(*ptep));
1382 		if (!(pte & _PAGE_PRESENT)) {
1383 			gpa += PAGE_SIZE;
1384 			continue;
1385 		}
1386 		shift = PAGE_SHIFT;
1387 	leaf:
1388 		n = scnprintf(p->buf, sizeof(p->buf),
1389 			      " %lx: %lx %d\n", gpa, pte, shift);
1390 		gpa += 1ul << shift;
1391 	copy:
1392 		p->chars_left = n;
1393 		if (n > len)
1394 			n = len;
1395 		r = copy_to_user(buf, p->buf, n);
1396 		n -= r;
1397 		p->chars_left -= n;
1398 		p->buf_index = n;
1399 		buf += n;
1400 		len -= n;
1401 		ret += n;
1402 		if (r) {
1403 			if (!ret)
1404 				ret = -EFAULT;
1405 			break;
1406 		}
1407 	}
1408 	p->gpa = gpa;
1409 	if (nested)
1410 		kvmhv_put_nested(nested);
1411 
1412  out:
1413 	mutex_unlock(&p->mutex);
1414 	return ret;
1415 }
1416 
1417 static ssize_t debugfs_radix_write(struct file *file, const char __user *buf,
1418 			   size_t len, loff_t *ppos)
1419 {
1420 	return -EACCES;
1421 }
1422 
1423 static const struct file_operations debugfs_radix_fops = {
1424 	.owner	 = THIS_MODULE,
1425 	.open	 = debugfs_radix_open,
1426 	.release = debugfs_radix_release,
1427 	.read	 = debugfs_radix_read,
1428 	.write	 = debugfs_radix_write,
1429 	.llseek	 = generic_file_llseek,
1430 };
1431 
1432 void kvmhv_radix_debugfs_init(struct kvm *kvm)
1433 {
1434 	debugfs_create_file("radix", 0400, kvm->arch.debugfs_dir, kvm,
1435 			    &debugfs_radix_fops);
1436 }
1437 
1438 int kvmppc_radix_init(void)
1439 {
1440 	unsigned long size = sizeof(void *) << RADIX_PTE_INDEX_SIZE;
1441 
1442 	kvm_pte_cache = kmem_cache_create("kvm-pte", size, size, 0, pte_ctor);
1443 	if (!kvm_pte_cache)
1444 		return -ENOMEM;
1445 
1446 	size = sizeof(void *) << RADIX_PMD_INDEX_SIZE;
1447 
1448 	kvm_pmd_cache = kmem_cache_create("kvm-pmd", size, size, 0, pmd_ctor);
1449 	if (!kvm_pmd_cache) {
1450 		kmem_cache_destroy(kvm_pte_cache);
1451 		return -ENOMEM;
1452 	}
1453 
1454 	return 0;
1455 }
1456 
1457 void kvmppc_radix_exit(void)
1458 {
1459 	kmem_cache_destroy(kvm_pte_cache);
1460 	kmem_cache_destroy(kvm_pmd_cache);
1461 }
1462