xref: /openbmc/linux/arch/s390/mm/gmap.c (revision 8730046c)
1 /*
2  *  KVM guest address space mapping code
3  *
4  *    Copyright IBM Corp. 2007, 2016
5  *    Author(s): Martin Schwidefsky <schwidefsky@de.ibm.com>
6  */
7 
8 #include <linux/kernel.h>
9 #include <linux/mm.h>
10 #include <linux/swap.h>
11 #include <linux/smp.h>
12 #include <linux/spinlock.h>
13 #include <linux/slab.h>
14 #include <linux/swapops.h>
15 #include <linux/ksm.h>
16 #include <linux/mman.h>
17 
18 #include <asm/pgtable.h>
19 #include <asm/pgalloc.h>
20 #include <asm/gmap.h>
21 #include <asm/tlb.h>
22 
23 #define GMAP_SHADOW_FAKE_TABLE 1ULL
24 
25 /**
26  * gmap_alloc - allocate and initialize a guest address space
27  * @mm: pointer to the parent mm_struct
28  * @limit: maximum address of the gmap address space
29  *
30  * Returns a guest address space structure.
31  */
32 static struct gmap *gmap_alloc(unsigned long limit)
33 {
34 	struct gmap *gmap;
35 	struct page *page;
36 	unsigned long *table;
37 	unsigned long etype, atype;
38 
39 	if (limit < (1UL << 31)) {
40 		limit = (1UL << 31) - 1;
41 		atype = _ASCE_TYPE_SEGMENT;
42 		etype = _SEGMENT_ENTRY_EMPTY;
43 	} else if (limit < (1UL << 42)) {
44 		limit = (1UL << 42) - 1;
45 		atype = _ASCE_TYPE_REGION3;
46 		etype = _REGION3_ENTRY_EMPTY;
47 	} else if (limit < (1UL << 53)) {
48 		limit = (1UL << 53) - 1;
49 		atype = _ASCE_TYPE_REGION2;
50 		etype = _REGION2_ENTRY_EMPTY;
51 	} else {
52 		limit = -1UL;
53 		atype = _ASCE_TYPE_REGION1;
54 		etype = _REGION1_ENTRY_EMPTY;
55 	}
56 	gmap = kzalloc(sizeof(struct gmap), GFP_KERNEL);
57 	if (!gmap)
58 		goto out;
59 	INIT_LIST_HEAD(&gmap->crst_list);
60 	INIT_LIST_HEAD(&gmap->children);
61 	INIT_LIST_HEAD(&gmap->pt_list);
62 	INIT_RADIX_TREE(&gmap->guest_to_host, GFP_KERNEL);
63 	INIT_RADIX_TREE(&gmap->host_to_guest, GFP_ATOMIC);
64 	INIT_RADIX_TREE(&gmap->host_to_rmap, GFP_ATOMIC);
65 	spin_lock_init(&gmap->guest_table_lock);
66 	spin_lock_init(&gmap->shadow_lock);
67 	atomic_set(&gmap->ref_count, 1);
68 	page = alloc_pages(GFP_KERNEL, 2);
69 	if (!page)
70 		goto out_free;
71 	page->index = 0;
72 	list_add(&page->lru, &gmap->crst_list);
73 	table = (unsigned long *) page_to_phys(page);
74 	crst_table_init(table, etype);
75 	gmap->table = table;
76 	gmap->asce = atype | _ASCE_TABLE_LENGTH |
77 		_ASCE_USER_BITS | __pa(table);
78 	gmap->asce_end = limit;
79 	return gmap;
80 
81 out_free:
82 	kfree(gmap);
83 out:
84 	return NULL;
85 }
86 
87 /**
88  * gmap_create - create a guest address space
89  * @mm: pointer to the parent mm_struct
90  * @limit: maximum size of the gmap address space
91  *
92  * Returns a guest address space structure.
93  */
94 struct gmap *gmap_create(struct mm_struct *mm, unsigned long limit)
95 {
96 	struct gmap *gmap;
97 	unsigned long gmap_asce;
98 
99 	gmap = gmap_alloc(limit);
100 	if (!gmap)
101 		return NULL;
102 	gmap->mm = mm;
103 	spin_lock(&mm->context.gmap_lock);
104 	list_add_rcu(&gmap->list, &mm->context.gmap_list);
105 	if (list_is_singular(&mm->context.gmap_list))
106 		gmap_asce = gmap->asce;
107 	else
108 		gmap_asce = -1UL;
109 	WRITE_ONCE(mm->context.gmap_asce, gmap_asce);
110 	spin_unlock(&mm->context.gmap_lock);
111 	return gmap;
112 }
113 EXPORT_SYMBOL_GPL(gmap_create);
114 
115 static void gmap_flush_tlb(struct gmap *gmap)
116 {
117 	if (MACHINE_HAS_IDTE)
118 		__tlb_flush_idte(gmap->asce);
119 	else
120 		__tlb_flush_global();
121 }
122 
123 static void gmap_radix_tree_free(struct radix_tree_root *root)
124 {
125 	struct radix_tree_iter iter;
126 	unsigned long indices[16];
127 	unsigned long index;
128 	void **slot;
129 	int i, nr;
130 
131 	/* A radix tree is freed by deleting all of its entries */
132 	index = 0;
133 	do {
134 		nr = 0;
135 		radix_tree_for_each_slot(slot, root, &iter, index) {
136 			indices[nr] = iter.index;
137 			if (++nr == 16)
138 				break;
139 		}
140 		for (i = 0; i < nr; i++) {
141 			index = indices[i];
142 			radix_tree_delete(root, index);
143 		}
144 	} while (nr > 0);
145 }
146 
147 static void gmap_rmap_radix_tree_free(struct radix_tree_root *root)
148 {
149 	struct gmap_rmap *rmap, *rnext, *head;
150 	struct radix_tree_iter iter;
151 	unsigned long indices[16];
152 	unsigned long index;
153 	void **slot;
154 	int i, nr;
155 
156 	/* A radix tree is freed by deleting all of its entries */
157 	index = 0;
158 	do {
159 		nr = 0;
160 		radix_tree_for_each_slot(slot, root, &iter, index) {
161 			indices[nr] = iter.index;
162 			if (++nr == 16)
163 				break;
164 		}
165 		for (i = 0; i < nr; i++) {
166 			index = indices[i];
167 			head = radix_tree_delete(root, index);
168 			gmap_for_each_rmap_safe(rmap, rnext, head)
169 				kfree(rmap);
170 		}
171 	} while (nr > 0);
172 }
173 
174 /**
175  * gmap_free - free a guest address space
176  * @gmap: pointer to the guest address space structure
177  *
178  * No locks required. There are no references to this gmap anymore.
179  */
180 static void gmap_free(struct gmap *gmap)
181 {
182 	struct page *page, *next;
183 
184 	/* Flush tlb of all gmaps (if not already done for shadows) */
185 	if (!(gmap_is_shadow(gmap) && gmap->removed))
186 		gmap_flush_tlb(gmap);
187 	/* Free all segment & region tables. */
188 	list_for_each_entry_safe(page, next, &gmap->crst_list, lru)
189 		__free_pages(page, 2);
190 	gmap_radix_tree_free(&gmap->guest_to_host);
191 	gmap_radix_tree_free(&gmap->host_to_guest);
192 
193 	/* Free additional data for a shadow gmap */
194 	if (gmap_is_shadow(gmap)) {
195 		/* Free all page tables. */
196 		list_for_each_entry_safe(page, next, &gmap->pt_list, lru)
197 			page_table_free_pgste(page);
198 		gmap_rmap_radix_tree_free(&gmap->host_to_rmap);
199 		/* Release reference to the parent */
200 		gmap_put(gmap->parent);
201 	}
202 
203 	kfree(gmap);
204 }
205 
206 /**
207  * gmap_get - increase reference counter for guest address space
208  * @gmap: pointer to the guest address space structure
209  *
210  * Returns the gmap pointer
211  */
212 struct gmap *gmap_get(struct gmap *gmap)
213 {
214 	atomic_inc(&gmap->ref_count);
215 	return gmap;
216 }
217 EXPORT_SYMBOL_GPL(gmap_get);
218 
219 /**
220  * gmap_put - decrease reference counter for guest address space
221  * @gmap: pointer to the guest address space structure
222  *
223  * If the reference counter reaches zero the guest address space is freed.
224  */
225 void gmap_put(struct gmap *gmap)
226 {
227 	if (atomic_dec_return(&gmap->ref_count) == 0)
228 		gmap_free(gmap);
229 }
230 EXPORT_SYMBOL_GPL(gmap_put);
231 
232 /**
233  * gmap_remove - remove a guest address space but do not free it yet
234  * @gmap: pointer to the guest address space structure
235  */
236 void gmap_remove(struct gmap *gmap)
237 {
238 	struct gmap *sg, *next;
239 	unsigned long gmap_asce;
240 
241 	/* Remove all shadow gmaps linked to this gmap */
242 	if (!list_empty(&gmap->children)) {
243 		spin_lock(&gmap->shadow_lock);
244 		list_for_each_entry_safe(sg, next, &gmap->children, list) {
245 			list_del(&sg->list);
246 			gmap_put(sg);
247 		}
248 		spin_unlock(&gmap->shadow_lock);
249 	}
250 	/* Remove gmap from the pre-mm list */
251 	spin_lock(&gmap->mm->context.gmap_lock);
252 	list_del_rcu(&gmap->list);
253 	if (list_empty(&gmap->mm->context.gmap_list))
254 		gmap_asce = 0;
255 	else if (list_is_singular(&gmap->mm->context.gmap_list))
256 		gmap_asce = list_first_entry(&gmap->mm->context.gmap_list,
257 					     struct gmap, list)->asce;
258 	else
259 		gmap_asce = -1UL;
260 	WRITE_ONCE(gmap->mm->context.gmap_asce, gmap_asce);
261 	spin_unlock(&gmap->mm->context.gmap_lock);
262 	synchronize_rcu();
263 	/* Put reference */
264 	gmap_put(gmap);
265 }
266 EXPORT_SYMBOL_GPL(gmap_remove);
267 
268 /**
269  * gmap_enable - switch primary space to the guest address space
270  * @gmap: pointer to the guest address space structure
271  */
272 void gmap_enable(struct gmap *gmap)
273 {
274 	S390_lowcore.gmap = (unsigned long) gmap;
275 }
276 EXPORT_SYMBOL_GPL(gmap_enable);
277 
278 /**
279  * gmap_disable - switch back to the standard primary address space
280  * @gmap: pointer to the guest address space structure
281  */
282 void gmap_disable(struct gmap *gmap)
283 {
284 	S390_lowcore.gmap = 0UL;
285 }
286 EXPORT_SYMBOL_GPL(gmap_disable);
287 
288 /**
289  * gmap_get_enabled - get a pointer to the currently enabled gmap
290  *
291  * Returns a pointer to the currently enabled gmap. 0 if none is enabled.
292  */
293 struct gmap *gmap_get_enabled(void)
294 {
295 	return (struct gmap *) S390_lowcore.gmap;
296 }
297 EXPORT_SYMBOL_GPL(gmap_get_enabled);
298 
299 /*
300  * gmap_alloc_table is assumed to be called with mmap_sem held
301  */
302 static int gmap_alloc_table(struct gmap *gmap, unsigned long *table,
303 			    unsigned long init, unsigned long gaddr)
304 {
305 	struct page *page;
306 	unsigned long *new;
307 
308 	/* since we dont free the gmap table until gmap_free we can unlock */
309 	page = alloc_pages(GFP_KERNEL, 2);
310 	if (!page)
311 		return -ENOMEM;
312 	new = (unsigned long *) page_to_phys(page);
313 	crst_table_init(new, init);
314 	spin_lock(&gmap->guest_table_lock);
315 	if (*table & _REGION_ENTRY_INVALID) {
316 		list_add(&page->lru, &gmap->crst_list);
317 		*table = (unsigned long) new | _REGION_ENTRY_LENGTH |
318 			(*table & _REGION_ENTRY_TYPE_MASK);
319 		page->index = gaddr;
320 		page = NULL;
321 	}
322 	spin_unlock(&gmap->guest_table_lock);
323 	if (page)
324 		__free_pages(page, 2);
325 	return 0;
326 }
327 
328 /**
329  * __gmap_segment_gaddr - find virtual address from segment pointer
330  * @entry: pointer to a segment table entry in the guest address space
331  *
332  * Returns the virtual address in the guest address space for the segment
333  */
334 static unsigned long __gmap_segment_gaddr(unsigned long *entry)
335 {
336 	struct page *page;
337 	unsigned long offset, mask;
338 
339 	offset = (unsigned long) entry / sizeof(unsigned long);
340 	offset = (offset & (PTRS_PER_PMD - 1)) * PMD_SIZE;
341 	mask = ~(PTRS_PER_PMD * sizeof(pmd_t) - 1);
342 	page = virt_to_page((void *)((unsigned long) entry & mask));
343 	return page->index + offset;
344 }
345 
346 /**
347  * __gmap_unlink_by_vmaddr - unlink a single segment via a host address
348  * @gmap: pointer to the guest address space structure
349  * @vmaddr: address in the host process address space
350  *
351  * Returns 1 if a TLB flush is required
352  */
353 static int __gmap_unlink_by_vmaddr(struct gmap *gmap, unsigned long vmaddr)
354 {
355 	unsigned long *entry;
356 	int flush = 0;
357 
358 	BUG_ON(gmap_is_shadow(gmap));
359 	spin_lock(&gmap->guest_table_lock);
360 	entry = radix_tree_delete(&gmap->host_to_guest, vmaddr >> PMD_SHIFT);
361 	if (entry) {
362 		flush = (*entry != _SEGMENT_ENTRY_INVALID);
363 		*entry = _SEGMENT_ENTRY_INVALID;
364 	}
365 	spin_unlock(&gmap->guest_table_lock);
366 	return flush;
367 }
368 
369 /**
370  * __gmap_unmap_by_gaddr - unmap a single segment via a guest address
371  * @gmap: pointer to the guest address space structure
372  * @gaddr: address in the guest address space
373  *
374  * Returns 1 if a TLB flush is required
375  */
376 static int __gmap_unmap_by_gaddr(struct gmap *gmap, unsigned long gaddr)
377 {
378 	unsigned long vmaddr;
379 
380 	vmaddr = (unsigned long) radix_tree_delete(&gmap->guest_to_host,
381 						   gaddr >> PMD_SHIFT);
382 	return vmaddr ? __gmap_unlink_by_vmaddr(gmap, vmaddr) : 0;
383 }
384 
385 /**
386  * gmap_unmap_segment - unmap segment from the guest address space
387  * @gmap: pointer to the guest address space structure
388  * @to: address in the guest address space
389  * @len: length of the memory area to unmap
390  *
391  * Returns 0 if the unmap succeeded, -EINVAL if not.
392  */
393 int gmap_unmap_segment(struct gmap *gmap, unsigned long to, unsigned long len)
394 {
395 	unsigned long off;
396 	int flush;
397 
398 	BUG_ON(gmap_is_shadow(gmap));
399 	if ((to | len) & (PMD_SIZE - 1))
400 		return -EINVAL;
401 	if (len == 0 || to + len < to)
402 		return -EINVAL;
403 
404 	flush = 0;
405 	down_write(&gmap->mm->mmap_sem);
406 	for (off = 0; off < len; off += PMD_SIZE)
407 		flush |= __gmap_unmap_by_gaddr(gmap, to + off);
408 	up_write(&gmap->mm->mmap_sem);
409 	if (flush)
410 		gmap_flush_tlb(gmap);
411 	return 0;
412 }
413 EXPORT_SYMBOL_GPL(gmap_unmap_segment);
414 
415 /**
416  * gmap_map_segment - map a segment to the guest address space
417  * @gmap: pointer to the guest address space structure
418  * @from: source address in the parent address space
419  * @to: target address in the guest address space
420  * @len: length of the memory area to map
421  *
422  * Returns 0 if the mmap succeeded, -EINVAL or -ENOMEM if not.
423  */
424 int gmap_map_segment(struct gmap *gmap, unsigned long from,
425 		     unsigned long to, unsigned long len)
426 {
427 	unsigned long off;
428 	int flush;
429 
430 	BUG_ON(gmap_is_shadow(gmap));
431 	if ((from | to | len) & (PMD_SIZE - 1))
432 		return -EINVAL;
433 	if (len == 0 || from + len < from || to + len < to ||
434 	    from + len - 1 > TASK_MAX_SIZE || to + len - 1 > gmap->asce_end)
435 		return -EINVAL;
436 
437 	flush = 0;
438 	down_write(&gmap->mm->mmap_sem);
439 	for (off = 0; off < len; off += PMD_SIZE) {
440 		/* Remove old translation */
441 		flush |= __gmap_unmap_by_gaddr(gmap, to + off);
442 		/* Store new translation */
443 		if (radix_tree_insert(&gmap->guest_to_host,
444 				      (to + off) >> PMD_SHIFT,
445 				      (void *) from + off))
446 			break;
447 	}
448 	up_write(&gmap->mm->mmap_sem);
449 	if (flush)
450 		gmap_flush_tlb(gmap);
451 	if (off >= len)
452 		return 0;
453 	gmap_unmap_segment(gmap, to, len);
454 	return -ENOMEM;
455 }
456 EXPORT_SYMBOL_GPL(gmap_map_segment);
457 
458 /**
459  * __gmap_translate - translate a guest address to a user space address
460  * @gmap: pointer to guest mapping meta data structure
461  * @gaddr: guest address
462  *
463  * Returns user space address which corresponds to the guest address or
464  * -EFAULT if no such mapping exists.
465  * This function does not establish potentially missing page table entries.
466  * The mmap_sem of the mm that belongs to the address space must be held
467  * when this function gets called.
468  *
469  * Note: Can also be called for shadow gmaps.
470  */
471 unsigned long __gmap_translate(struct gmap *gmap, unsigned long gaddr)
472 {
473 	unsigned long vmaddr;
474 
475 	vmaddr = (unsigned long)
476 		radix_tree_lookup(&gmap->guest_to_host, gaddr >> PMD_SHIFT);
477 	/* Note: guest_to_host is empty for a shadow gmap */
478 	return vmaddr ? (vmaddr | (gaddr & ~PMD_MASK)) : -EFAULT;
479 }
480 EXPORT_SYMBOL_GPL(__gmap_translate);
481 
482 /**
483  * gmap_translate - translate a guest address to a user space address
484  * @gmap: pointer to guest mapping meta data structure
485  * @gaddr: guest address
486  *
487  * Returns user space address which corresponds to the guest address or
488  * -EFAULT if no such mapping exists.
489  * This function does not establish potentially missing page table entries.
490  */
491 unsigned long gmap_translate(struct gmap *gmap, unsigned long gaddr)
492 {
493 	unsigned long rc;
494 
495 	down_read(&gmap->mm->mmap_sem);
496 	rc = __gmap_translate(gmap, gaddr);
497 	up_read(&gmap->mm->mmap_sem);
498 	return rc;
499 }
500 EXPORT_SYMBOL_GPL(gmap_translate);
501 
502 /**
503  * gmap_unlink - disconnect a page table from the gmap shadow tables
504  * @gmap: pointer to guest mapping meta data structure
505  * @table: pointer to the host page table
506  * @vmaddr: vm address associated with the host page table
507  */
508 void gmap_unlink(struct mm_struct *mm, unsigned long *table,
509 		 unsigned long vmaddr)
510 {
511 	struct gmap *gmap;
512 	int flush;
513 
514 	rcu_read_lock();
515 	list_for_each_entry_rcu(gmap, &mm->context.gmap_list, list) {
516 		flush = __gmap_unlink_by_vmaddr(gmap, vmaddr);
517 		if (flush)
518 			gmap_flush_tlb(gmap);
519 	}
520 	rcu_read_unlock();
521 }
522 
523 /**
524  * gmap_link - set up shadow page tables to connect a host to a guest address
525  * @gmap: pointer to guest mapping meta data structure
526  * @gaddr: guest address
527  * @vmaddr: vm address
528  *
529  * Returns 0 on success, -ENOMEM for out of memory conditions, and -EFAULT
530  * if the vm address is already mapped to a different guest segment.
531  * The mmap_sem of the mm that belongs to the address space must be held
532  * when this function gets called.
533  */
534 int __gmap_link(struct gmap *gmap, unsigned long gaddr, unsigned long vmaddr)
535 {
536 	struct mm_struct *mm;
537 	unsigned long *table;
538 	spinlock_t *ptl;
539 	pgd_t *pgd;
540 	pud_t *pud;
541 	pmd_t *pmd;
542 	int rc;
543 
544 	BUG_ON(gmap_is_shadow(gmap));
545 	/* Create higher level tables in the gmap page table */
546 	table = gmap->table;
547 	if ((gmap->asce & _ASCE_TYPE_MASK) >= _ASCE_TYPE_REGION1) {
548 		table += (gaddr >> 53) & 0x7ff;
549 		if ((*table & _REGION_ENTRY_INVALID) &&
550 		    gmap_alloc_table(gmap, table, _REGION2_ENTRY_EMPTY,
551 				     gaddr & 0xffe0000000000000UL))
552 			return -ENOMEM;
553 		table = (unsigned long *)(*table & _REGION_ENTRY_ORIGIN);
554 	}
555 	if ((gmap->asce & _ASCE_TYPE_MASK) >= _ASCE_TYPE_REGION2) {
556 		table += (gaddr >> 42) & 0x7ff;
557 		if ((*table & _REGION_ENTRY_INVALID) &&
558 		    gmap_alloc_table(gmap, table, _REGION3_ENTRY_EMPTY,
559 				     gaddr & 0xfffffc0000000000UL))
560 			return -ENOMEM;
561 		table = (unsigned long *)(*table & _REGION_ENTRY_ORIGIN);
562 	}
563 	if ((gmap->asce & _ASCE_TYPE_MASK) >= _ASCE_TYPE_REGION3) {
564 		table += (gaddr >> 31) & 0x7ff;
565 		if ((*table & _REGION_ENTRY_INVALID) &&
566 		    gmap_alloc_table(gmap, table, _SEGMENT_ENTRY_EMPTY,
567 				     gaddr & 0xffffffff80000000UL))
568 			return -ENOMEM;
569 		table = (unsigned long *)(*table & _REGION_ENTRY_ORIGIN);
570 	}
571 	table += (gaddr >> 20) & 0x7ff;
572 	/* Walk the parent mm page table */
573 	mm = gmap->mm;
574 	pgd = pgd_offset(mm, vmaddr);
575 	VM_BUG_ON(pgd_none(*pgd));
576 	pud = pud_offset(pgd, vmaddr);
577 	VM_BUG_ON(pud_none(*pud));
578 	/* large puds cannot yet be handled */
579 	if (pud_large(*pud))
580 		return -EFAULT;
581 	pmd = pmd_offset(pud, vmaddr);
582 	VM_BUG_ON(pmd_none(*pmd));
583 	/* large pmds cannot yet be handled */
584 	if (pmd_large(*pmd))
585 		return -EFAULT;
586 	/* Link gmap segment table entry location to page table. */
587 	rc = radix_tree_preload(GFP_KERNEL);
588 	if (rc)
589 		return rc;
590 	ptl = pmd_lock(mm, pmd);
591 	spin_lock(&gmap->guest_table_lock);
592 	if (*table == _SEGMENT_ENTRY_INVALID) {
593 		rc = radix_tree_insert(&gmap->host_to_guest,
594 				       vmaddr >> PMD_SHIFT, table);
595 		if (!rc)
596 			*table = pmd_val(*pmd);
597 	} else
598 		rc = 0;
599 	spin_unlock(&gmap->guest_table_lock);
600 	spin_unlock(ptl);
601 	radix_tree_preload_end();
602 	return rc;
603 }
604 
605 /**
606  * gmap_fault - resolve a fault on a guest address
607  * @gmap: pointer to guest mapping meta data structure
608  * @gaddr: guest address
609  * @fault_flags: flags to pass down to handle_mm_fault()
610  *
611  * Returns 0 on success, -ENOMEM for out of memory conditions, and -EFAULT
612  * if the vm address is already mapped to a different guest segment.
613  */
614 int gmap_fault(struct gmap *gmap, unsigned long gaddr,
615 	       unsigned int fault_flags)
616 {
617 	unsigned long vmaddr;
618 	int rc;
619 	bool unlocked;
620 
621 	down_read(&gmap->mm->mmap_sem);
622 
623 retry:
624 	unlocked = false;
625 	vmaddr = __gmap_translate(gmap, gaddr);
626 	if (IS_ERR_VALUE(vmaddr)) {
627 		rc = vmaddr;
628 		goto out_up;
629 	}
630 	if (fixup_user_fault(current, gmap->mm, vmaddr, fault_flags,
631 			     &unlocked)) {
632 		rc = -EFAULT;
633 		goto out_up;
634 	}
635 	/*
636 	 * In the case that fixup_user_fault unlocked the mmap_sem during
637 	 * faultin redo __gmap_translate to not race with a map/unmap_segment.
638 	 */
639 	if (unlocked)
640 		goto retry;
641 
642 	rc = __gmap_link(gmap, gaddr, vmaddr);
643 out_up:
644 	up_read(&gmap->mm->mmap_sem);
645 	return rc;
646 }
647 EXPORT_SYMBOL_GPL(gmap_fault);
648 
649 /*
650  * this function is assumed to be called with mmap_sem held
651  */
652 void __gmap_zap(struct gmap *gmap, unsigned long gaddr)
653 {
654 	unsigned long vmaddr;
655 	spinlock_t *ptl;
656 	pte_t *ptep;
657 
658 	/* Find the vm address for the guest address */
659 	vmaddr = (unsigned long) radix_tree_lookup(&gmap->guest_to_host,
660 						   gaddr >> PMD_SHIFT);
661 	if (vmaddr) {
662 		vmaddr |= gaddr & ~PMD_MASK;
663 		/* Get pointer to the page table entry */
664 		ptep = get_locked_pte(gmap->mm, vmaddr, &ptl);
665 		if (likely(ptep))
666 			ptep_zap_unused(gmap->mm, vmaddr, ptep, 0);
667 		pte_unmap_unlock(ptep, ptl);
668 	}
669 }
670 EXPORT_SYMBOL_GPL(__gmap_zap);
671 
672 void gmap_discard(struct gmap *gmap, unsigned long from, unsigned long to)
673 {
674 	unsigned long gaddr, vmaddr, size;
675 	struct vm_area_struct *vma;
676 
677 	down_read(&gmap->mm->mmap_sem);
678 	for (gaddr = from; gaddr < to;
679 	     gaddr = (gaddr + PMD_SIZE) & PMD_MASK) {
680 		/* Find the vm address for the guest address */
681 		vmaddr = (unsigned long)
682 			radix_tree_lookup(&gmap->guest_to_host,
683 					  gaddr >> PMD_SHIFT);
684 		if (!vmaddr)
685 			continue;
686 		vmaddr |= gaddr & ~PMD_MASK;
687 		/* Find vma in the parent mm */
688 		vma = find_vma(gmap->mm, vmaddr);
689 		size = min(to - gaddr, PMD_SIZE - (gaddr & ~PMD_MASK));
690 		zap_page_range(vma, vmaddr, size, NULL);
691 	}
692 	up_read(&gmap->mm->mmap_sem);
693 }
694 EXPORT_SYMBOL_GPL(gmap_discard);
695 
696 static LIST_HEAD(gmap_notifier_list);
697 static DEFINE_SPINLOCK(gmap_notifier_lock);
698 
699 /**
700  * gmap_register_pte_notifier - register a pte invalidation callback
701  * @nb: pointer to the gmap notifier block
702  */
703 void gmap_register_pte_notifier(struct gmap_notifier *nb)
704 {
705 	spin_lock(&gmap_notifier_lock);
706 	list_add_rcu(&nb->list, &gmap_notifier_list);
707 	spin_unlock(&gmap_notifier_lock);
708 }
709 EXPORT_SYMBOL_GPL(gmap_register_pte_notifier);
710 
711 /**
712  * gmap_unregister_pte_notifier - remove a pte invalidation callback
713  * @nb: pointer to the gmap notifier block
714  */
715 void gmap_unregister_pte_notifier(struct gmap_notifier *nb)
716 {
717 	spin_lock(&gmap_notifier_lock);
718 	list_del_rcu(&nb->list);
719 	spin_unlock(&gmap_notifier_lock);
720 	synchronize_rcu();
721 }
722 EXPORT_SYMBOL_GPL(gmap_unregister_pte_notifier);
723 
724 /**
725  * gmap_call_notifier - call all registered invalidation callbacks
726  * @gmap: pointer to guest mapping meta data structure
727  * @start: start virtual address in the guest address space
728  * @end: end virtual address in the guest address space
729  */
730 static void gmap_call_notifier(struct gmap *gmap, unsigned long start,
731 			       unsigned long end)
732 {
733 	struct gmap_notifier *nb;
734 
735 	list_for_each_entry(nb, &gmap_notifier_list, list)
736 		nb->notifier_call(gmap, start, end);
737 }
738 
739 /**
740  * gmap_table_walk - walk the gmap page tables
741  * @gmap: pointer to guest mapping meta data structure
742  * @gaddr: virtual address in the guest address space
743  * @level: page table level to stop at
744  *
745  * Returns a table entry pointer for the given guest address and @level
746  * @level=0 : returns a pointer to a page table table entry (or NULL)
747  * @level=1 : returns a pointer to a segment table entry (or NULL)
748  * @level=2 : returns a pointer to a region-3 table entry (or NULL)
749  * @level=3 : returns a pointer to a region-2 table entry (or NULL)
750  * @level=4 : returns a pointer to a region-1 table entry (or NULL)
751  *
752  * Returns NULL if the gmap page tables could not be walked to the
753  * requested level.
754  *
755  * Note: Can also be called for shadow gmaps.
756  */
757 static inline unsigned long *gmap_table_walk(struct gmap *gmap,
758 					     unsigned long gaddr, int level)
759 {
760 	unsigned long *table;
761 
762 	if ((gmap->asce & _ASCE_TYPE_MASK) + 4 < (level * 4))
763 		return NULL;
764 	if (gmap_is_shadow(gmap) && gmap->removed)
765 		return NULL;
766 	if (gaddr & (-1UL << (31 + ((gmap->asce & _ASCE_TYPE_MASK) >> 2)*11)))
767 		return NULL;
768 	table = gmap->table;
769 	switch (gmap->asce & _ASCE_TYPE_MASK) {
770 	case _ASCE_TYPE_REGION1:
771 		table += (gaddr >> 53) & 0x7ff;
772 		if (level == 4)
773 			break;
774 		if (*table & _REGION_ENTRY_INVALID)
775 			return NULL;
776 		table = (unsigned long *)(*table & _REGION_ENTRY_ORIGIN);
777 		/* Fallthrough */
778 	case _ASCE_TYPE_REGION2:
779 		table += (gaddr >> 42) & 0x7ff;
780 		if (level == 3)
781 			break;
782 		if (*table & _REGION_ENTRY_INVALID)
783 			return NULL;
784 		table = (unsigned long *)(*table & _REGION_ENTRY_ORIGIN);
785 		/* Fallthrough */
786 	case _ASCE_TYPE_REGION3:
787 		table += (gaddr >> 31) & 0x7ff;
788 		if (level == 2)
789 			break;
790 		if (*table & _REGION_ENTRY_INVALID)
791 			return NULL;
792 		table = (unsigned long *)(*table & _REGION_ENTRY_ORIGIN);
793 		/* Fallthrough */
794 	case _ASCE_TYPE_SEGMENT:
795 		table += (gaddr >> 20) & 0x7ff;
796 		if (level == 1)
797 			break;
798 		if (*table & _REGION_ENTRY_INVALID)
799 			return NULL;
800 		table = (unsigned long *)(*table & _SEGMENT_ENTRY_ORIGIN);
801 		table += (gaddr >> 12) & 0xff;
802 	}
803 	return table;
804 }
805 
806 /**
807  * gmap_pte_op_walk - walk the gmap page table, get the page table lock
808  *		      and return the pte pointer
809  * @gmap: pointer to guest mapping meta data structure
810  * @gaddr: virtual address in the guest address space
811  * @ptl: pointer to the spinlock pointer
812  *
813  * Returns a pointer to the locked pte for a guest address, or NULL
814  *
815  * Note: Can also be called for shadow gmaps.
816  */
817 static pte_t *gmap_pte_op_walk(struct gmap *gmap, unsigned long gaddr,
818 			       spinlock_t **ptl)
819 {
820 	unsigned long *table;
821 
822 	if (gmap_is_shadow(gmap))
823 		spin_lock(&gmap->guest_table_lock);
824 	/* Walk the gmap page table, lock and get pte pointer */
825 	table = gmap_table_walk(gmap, gaddr, 1); /* get segment pointer */
826 	if (!table || *table & _SEGMENT_ENTRY_INVALID) {
827 		if (gmap_is_shadow(gmap))
828 			spin_unlock(&gmap->guest_table_lock);
829 		return NULL;
830 	}
831 	if (gmap_is_shadow(gmap)) {
832 		*ptl = &gmap->guest_table_lock;
833 		return pte_offset_map((pmd_t *) table, gaddr);
834 	}
835 	return pte_alloc_map_lock(gmap->mm, (pmd_t *) table, gaddr, ptl);
836 }
837 
838 /**
839  * gmap_pte_op_fixup - force a page in and connect the gmap page table
840  * @gmap: pointer to guest mapping meta data structure
841  * @gaddr: virtual address in the guest address space
842  * @vmaddr: address in the host process address space
843  * @prot: indicates access rights: PROT_NONE, PROT_READ or PROT_WRITE
844  *
845  * Returns 0 if the caller can retry __gmap_translate (might fail again),
846  * -ENOMEM if out of memory and -EFAULT if anything goes wrong while fixing
847  * up or connecting the gmap page table.
848  */
849 static int gmap_pte_op_fixup(struct gmap *gmap, unsigned long gaddr,
850 			     unsigned long vmaddr, int prot)
851 {
852 	struct mm_struct *mm = gmap->mm;
853 	unsigned int fault_flags;
854 	bool unlocked = false;
855 
856 	BUG_ON(gmap_is_shadow(gmap));
857 	fault_flags = (prot == PROT_WRITE) ? FAULT_FLAG_WRITE : 0;
858 	if (fixup_user_fault(current, mm, vmaddr, fault_flags, &unlocked))
859 		return -EFAULT;
860 	if (unlocked)
861 		/* lost mmap_sem, caller has to retry __gmap_translate */
862 		return 0;
863 	/* Connect the page tables */
864 	return __gmap_link(gmap, gaddr, vmaddr);
865 }
866 
867 /**
868  * gmap_pte_op_end - release the page table lock
869  * @ptl: pointer to the spinlock pointer
870  */
871 static void gmap_pte_op_end(spinlock_t *ptl)
872 {
873 	spin_unlock(ptl);
874 }
875 
876 /*
877  * gmap_protect_range - remove access rights to memory and set pgste bits
878  * @gmap: pointer to guest mapping meta data structure
879  * @gaddr: virtual address in the guest address space
880  * @len: size of area
881  * @prot: indicates access rights: PROT_NONE, PROT_READ or PROT_WRITE
882  * @bits: pgste notification bits to set
883  *
884  * Returns 0 if successfully protected, -ENOMEM if out of memory and
885  * -EFAULT if gaddr is invalid (or mapping for shadows is missing).
886  *
887  * Called with sg->mm->mmap_sem in read.
888  *
889  * Note: Can also be called for shadow gmaps.
890  */
891 static int gmap_protect_range(struct gmap *gmap, unsigned long gaddr,
892 			      unsigned long len, int prot, unsigned long bits)
893 {
894 	unsigned long vmaddr;
895 	spinlock_t *ptl;
896 	pte_t *ptep;
897 	int rc;
898 
899 	while (len) {
900 		rc = -EAGAIN;
901 		ptep = gmap_pte_op_walk(gmap, gaddr, &ptl);
902 		if (ptep) {
903 			rc = ptep_force_prot(gmap->mm, gaddr, ptep, prot, bits);
904 			gmap_pte_op_end(ptl);
905 		}
906 		if (rc) {
907 			vmaddr = __gmap_translate(gmap, gaddr);
908 			if (IS_ERR_VALUE(vmaddr))
909 				return vmaddr;
910 			rc = gmap_pte_op_fixup(gmap, gaddr, vmaddr, prot);
911 			if (rc)
912 				return rc;
913 			continue;
914 		}
915 		gaddr += PAGE_SIZE;
916 		len -= PAGE_SIZE;
917 	}
918 	return 0;
919 }
920 
921 /**
922  * gmap_mprotect_notify - change access rights for a range of ptes and
923  *                        call the notifier if any pte changes again
924  * @gmap: pointer to guest mapping meta data structure
925  * @gaddr: virtual address in the guest address space
926  * @len: size of area
927  * @prot: indicates access rights: PROT_NONE, PROT_READ or PROT_WRITE
928  *
929  * Returns 0 if for each page in the given range a gmap mapping exists,
930  * the new access rights could be set and the notifier could be armed.
931  * If the gmap mapping is missing for one or more pages -EFAULT is
932  * returned. If no memory could be allocated -ENOMEM is returned.
933  * This function establishes missing page table entries.
934  */
935 int gmap_mprotect_notify(struct gmap *gmap, unsigned long gaddr,
936 			 unsigned long len, int prot)
937 {
938 	int rc;
939 
940 	if ((gaddr & ~PAGE_MASK) || (len & ~PAGE_MASK) || gmap_is_shadow(gmap))
941 		return -EINVAL;
942 	if (!MACHINE_HAS_ESOP && prot == PROT_READ)
943 		return -EINVAL;
944 	down_read(&gmap->mm->mmap_sem);
945 	rc = gmap_protect_range(gmap, gaddr, len, prot, PGSTE_IN_BIT);
946 	up_read(&gmap->mm->mmap_sem);
947 	return rc;
948 }
949 EXPORT_SYMBOL_GPL(gmap_mprotect_notify);
950 
951 /**
952  * gmap_read_table - get an unsigned long value from a guest page table using
953  *                   absolute addressing, without marking the page referenced.
954  * @gmap: pointer to guest mapping meta data structure
955  * @gaddr: virtual address in the guest address space
956  * @val: pointer to the unsigned long value to return
957  *
958  * Returns 0 if the value was read, -ENOMEM if out of memory and -EFAULT
959  * if reading using the virtual address failed.
960  *
961  * Called with gmap->mm->mmap_sem in read.
962  */
963 int gmap_read_table(struct gmap *gmap, unsigned long gaddr, unsigned long *val)
964 {
965 	unsigned long address, vmaddr;
966 	spinlock_t *ptl;
967 	pte_t *ptep, pte;
968 	int rc;
969 
970 	while (1) {
971 		rc = -EAGAIN;
972 		ptep = gmap_pte_op_walk(gmap, gaddr, &ptl);
973 		if (ptep) {
974 			pte = *ptep;
975 			if (pte_present(pte) && (pte_val(pte) & _PAGE_READ)) {
976 				address = pte_val(pte) & PAGE_MASK;
977 				address += gaddr & ~PAGE_MASK;
978 				*val = *(unsigned long *) address;
979 				pte_val(*ptep) |= _PAGE_YOUNG;
980 				/* Do *NOT* clear the _PAGE_INVALID bit! */
981 				rc = 0;
982 			}
983 			gmap_pte_op_end(ptl);
984 		}
985 		if (!rc)
986 			break;
987 		vmaddr = __gmap_translate(gmap, gaddr);
988 		if (IS_ERR_VALUE(vmaddr)) {
989 			rc = vmaddr;
990 			break;
991 		}
992 		rc = gmap_pte_op_fixup(gmap, gaddr, vmaddr, PROT_READ);
993 		if (rc)
994 			break;
995 	}
996 	return rc;
997 }
998 EXPORT_SYMBOL_GPL(gmap_read_table);
999 
1000 /**
1001  * gmap_insert_rmap - add a rmap to the host_to_rmap radix tree
1002  * @sg: pointer to the shadow guest address space structure
1003  * @vmaddr: vm address associated with the rmap
1004  * @rmap: pointer to the rmap structure
1005  *
1006  * Called with the sg->guest_table_lock
1007  */
1008 static inline void gmap_insert_rmap(struct gmap *sg, unsigned long vmaddr,
1009 				    struct gmap_rmap *rmap)
1010 {
1011 	void **slot;
1012 
1013 	BUG_ON(!gmap_is_shadow(sg));
1014 	slot = radix_tree_lookup_slot(&sg->host_to_rmap, vmaddr >> PAGE_SHIFT);
1015 	if (slot) {
1016 		rmap->next = radix_tree_deref_slot_protected(slot,
1017 							&sg->guest_table_lock);
1018 		radix_tree_replace_slot(&sg->host_to_rmap, slot, rmap);
1019 	} else {
1020 		rmap->next = NULL;
1021 		radix_tree_insert(&sg->host_to_rmap, vmaddr >> PAGE_SHIFT,
1022 				  rmap);
1023 	}
1024 }
1025 
1026 /**
1027  * gmap_protect_rmap - modify access rights to memory and create an rmap
1028  * @sg: pointer to the shadow guest address space structure
1029  * @raddr: rmap address in the shadow gmap
1030  * @paddr: address in the parent guest address space
1031  * @len: length of the memory area to protect
1032  * @prot: indicates access rights: none, read-only or read-write
1033  *
1034  * Returns 0 if successfully protected and the rmap was created, -ENOMEM
1035  * if out of memory and -EFAULT if paddr is invalid.
1036  */
1037 static int gmap_protect_rmap(struct gmap *sg, unsigned long raddr,
1038 			     unsigned long paddr, unsigned long len, int prot)
1039 {
1040 	struct gmap *parent;
1041 	struct gmap_rmap *rmap;
1042 	unsigned long vmaddr;
1043 	spinlock_t *ptl;
1044 	pte_t *ptep;
1045 	int rc;
1046 
1047 	BUG_ON(!gmap_is_shadow(sg));
1048 	parent = sg->parent;
1049 	while (len) {
1050 		vmaddr = __gmap_translate(parent, paddr);
1051 		if (IS_ERR_VALUE(vmaddr))
1052 			return vmaddr;
1053 		rmap = kzalloc(sizeof(*rmap), GFP_KERNEL);
1054 		if (!rmap)
1055 			return -ENOMEM;
1056 		rmap->raddr = raddr;
1057 		rc = radix_tree_preload(GFP_KERNEL);
1058 		if (rc) {
1059 			kfree(rmap);
1060 			return rc;
1061 		}
1062 		rc = -EAGAIN;
1063 		ptep = gmap_pte_op_walk(parent, paddr, &ptl);
1064 		if (ptep) {
1065 			spin_lock(&sg->guest_table_lock);
1066 			rc = ptep_force_prot(parent->mm, paddr, ptep, prot,
1067 					     PGSTE_VSIE_BIT);
1068 			if (!rc)
1069 				gmap_insert_rmap(sg, vmaddr, rmap);
1070 			spin_unlock(&sg->guest_table_lock);
1071 			gmap_pte_op_end(ptl);
1072 		}
1073 		radix_tree_preload_end();
1074 		if (rc) {
1075 			kfree(rmap);
1076 			rc = gmap_pte_op_fixup(parent, paddr, vmaddr, prot);
1077 			if (rc)
1078 				return rc;
1079 			continue;
1080 		}
1081 		paddr += PAGE_SIZE;
1082 		len -= PAGE_SIZE;
1083 	}
1084 	return 0;
1085 }
1086 
1087 #define _SHADOW_RMAP_MASK	0x7
1088 #define _SHADOW_RMAP_REGION1	0x5
1089 #define _SHADOW_RMAP_REGION2	0x4
1090 #define _SHADOW_RMAP_REGION3	0x3
1091 #define _SHADOW_RMAP_SEGMENT	0x2
1092 #define _SHADOW_RMAP_PGTABLE	0x1
1093 
1094 /**
1095  * gmap_idte_one - invalidate a single region or segment table entry
1096  * @asce: region or segment table *origin* + table-type bits
1097  * @vaddr: virtual address to identify the table entry to flush
1098  *
1099  * The invalid bit of a single region or segment table entry is set
1100  * and the associated TLB entries depending on the entry are flushed.
1101  * The table-type of the @asce identifies the portion of the @vaddr
1102  * that is used as the invalidation index.
1103  */
1104 static inline void gmap_idte_one(unsigned long asce, unsigned long vaddr)
1105 {
1106 	asm volatile(
1107 		"	.insn	rrf,0xb98e0000,%0,%1,0,0"
1108 		: : "a" (asce), "a" (vaddr) : "cc", "memory");
1109 }
1110 
1111 /**
1112  * gmap_unshadow_page - remove a page from a shadow page table
1113  * @sg: pointer to the shadow guest address space structure
1114  * @raddr: rmap address in the shadow guest address space
1115  *
1116  * Called with the sg->guest_table_lock
1117  */
1118 static void gmap_unshadow_page(struct gmap *sg, unsigned long raddr)
1119 {
1120 	unsigned long *table;
1121 
1122 	BUG_ON(!gmap_is_shadow(sg));
1123 	table = gmap_table_walk(sg, raddr, 0); /* get page table pointer */
1124 	if (!table || *table & _PAGE_INVALID)
1125 		return;
1126 	gmap_call_notifier(sg, raddr, raddr + (1UL << 12) - 1);
1127 	ptep_unshadow_pte(sg->mm, raddr, (pte_t *) table);
1128 }
1129 
1130 /**
1131  * __gmap_unshadow_pgt - remove all entries from a shadow page table
1132  * @sg: pointer to the shadow guest address space structure
1133  * @raddr: rmap address in the shadow guest address space
1134  * @pgt: pointer to the start of a shadow page table
1135  *
1136  * Called with the sg->guest_table_lock
1137  */
1138 static void __gmap_unshadow_pgt(struct gmap *sg, unsigned long raddr,
1139 				unsigned long *pgt)
1140 {
1141 	int i;
1142 
1143 	BUG_ON(!gmap_is_shadow(sg));
1144 	for (i = 0; i < 256; i++, raddr += 1UL << 12)
1145 		pgt[i] = _PAGE_INVALID;
1146 }
1147 
1148 /**
1149  * gmap_unshadow_pgt - remove a shadow page table from a segment entry
1150  * @sg: pointer to the shadow guest address space structure
1151  * @raddr: address in the shadow guest address space
1152  *
1153  * Called with the sg->guest_table_lock
1154  */
1155 static void gmap_unshadow_pgt(struct gmap *sg, unsigned long raddr)
1156 {
1157 	unsigned long sto, *ste, *pgt;
1158 	struct page *page;
1159 
1160 	BUG_ON(!gmap_is_shadow(sg));
1161 	ste = gmap_table_walk(sg, raddr, 1); /* get segment pointer */
1162 	if (!ste || !(*ste & _SEGMENT_ENTRY_ORIGIN))
1163 		return;
1164 	gmap_call_notifier(sg, raddr, raddr + (1UL << 20) - 1);
1165 	sto = (unsigned long) (ste - ((raddr >> 20) & 0x7ff));
1166 	gmap_idte_one(sto | _ASCE_TYPE_SEGMENT, raddr);
1167 	pgt = (unsigned long *)(*ste & _SEGMENT_ENTRY_ORIGIN);
1168 	*ste = _SEGMENT_ENTRY_EMPTY;
1169 	__gmap_unshadow_pgt(sg, raddr, pgt);
1170 	/* Free page table */
1171 	page = pfn_to_page(__pa(pgt) >> PAGE_SHIFT);
1172 	list_del(&page->lru);
1173 	page_table_free_pgste(page);
1174 }
1175 
1176 /**
1177  * __gmap_unshadow_sgt - remove all entries from a shadow segment table
1178  * @sg: pointer to the shadow guest address space structure
1179  * @raddr: rmap address in the shadow guest address space
1180  * @sgt: pointer to the start of a shadow segment table
1181  *
1182  * Called with the sg->guest_table_lock
1183  */
1184 static void __gmap_unshadow_sgt(struct gmap *sg, unsigned long raddr,
1185 				unsigned long *sgt)
1186 {
1187 	unsigned long asce, *pgt;
1188 	struct page *page;
1189 	int i;
1190 
1191 	BUG_ON(!gmap_is_shadow(sg));
1192 	asce = (unsigned long) sgt | _ASCE_TYPE_SEGMENT;
1193 	for (i = 0; i < 2048; i++, raddr += 1UL << 20) {
1194 		if (!(sgt[i] & _SEGMENT_ENTRY_ORIGIN))
1195 			continue;
1196 		pgt = (unsigned long *)(sgt[i] & _REGION_ENTRY_ORIGIN);
1197 		sgt[i] = _SEGMENT_ENTRY_EMPTY;
1198 		__gmap_unshadow_pgt(sg, raddr, pgt);
1199 		/* Free page table */
1200 		page = pfn_to_page(__pa(pgt) >> PAGE_SHIFT);
1201 		list_del(&page->lru);
1202 		page_table_free_pgste(page);
1203 	}
1204 }
1205 
1206 /**
1207  * gmap_unshadow_sgt - remove a shadow segment table from a region-3 entry
1208  * @sg: pointer to the shadow guest address space structure
1209  * @raddr: rmap address in the shadow guest address space
1210  *
1211  * Called with the shadow->guest_table_lock
1212  */
1213 static void gmap_unshadow_sgt(struct gmap *sg, unsigned long raddr)
1214 {
1215 	unsigned long r3o, *r3e, *sgt;
1216 	struct page *page;
1217 
1218 	BUG_ON(!gmap_is_shadow(sg));
1219 	r3e = gmap_table_walk(sg, raddr, 2); /* get region-3 pointer */
1220 	if (!r3e || !(*r3e & _REGION_ENTRY_ORIGIN))
1221 		return;
1222 	gmap_call_notifier(sg, raddr, raddr + (1UL << 31) - 1);
1223 	r3o = (unsigned long) (r3e - ((raddr >> 31) & 0x7ff));
1224 	gmap_idte_one(r3o | _ASCE_TYPE_REGION3, raddr);
1225 	sgt = (unsigned long *)(*r3e & _REGION_ENTRY_ORIGIN);
1226 	*r3e = _REGION3_ENTRY_EMPTY;
1227 	__gmap_unshadow_sgt(sg, raddr, sgt);
1228 	/* Free segment table */
1229 	page = pfn_to_page(__pa(sgt) >> PAGE_SHIFT);
1230 	list_del(&page->lru);
1231 	__free_pages(page, 2);
1232 }
1233 
1234 /**
1235  * __gmap_unshadow_r3t - remove all entries from a shadow region-3 table
1236  * @sg: pointer to the shadow guest address space structure
1237  * @raddr: address in the shadow guest address space
1238  * @r3t: pointer to the start of a shadow region-3 table
1239  *
1240  * Called with the sg->guest_table_lock
1241  */
1242 static void __gmap_unshadow_r3t(struct gmap *sg, unsigned long raddr,
1243 				unsigned long *r3t)
1244 {
1245 	unsigned long asce, *sgt;
1246 	struct page *page;
1247 	int i;
1248 
1249 	BUG_ON(!gmap_is_shadow(sg));
1250 	asce = (unsigned long) r3t | _ASCE_TYPE_REGION3;
1251 	for (i = 0; i < 2048; i++, raddr += 1UL << 31) {
1252 		if (!(r3t[i] & _REGION_ENTRY_ORIGIN))
1253 			continue;
1254 		sgt = (unsigned long *)(r3t[i] & _REGION_ENTRY_ORIGIN);
1255 		r3t[i] = _REGION3_ENTRY_EMPTY;
1256 		__gmap_unshadow_sgt(sg, raddr, sgt);
1257 		/* Free segment table */
1258 		page = pfn_to_page(__pa(sgt) >> PAGE_SHIFT);
1259 		list_del(&page->lru);
1260 		__free_pages(page, 2);
1261 	}
1262 }
1263 
1264 /**
1265  * gmap_unshadow_r3t - remove a shadow region-3 table from a region-2 entry
1266  * @sg: pointer to the shadow guest address space structure
1267  * @raddr: rmap address in the shadow guest address space
1268  *
1269  * Called with the sg->guest_table_lock
1270  */
1271 static void gmap_unshadow_r3t(struct gmap *sg, unsigned long raddr)
1272 {
1273 	unsigned long r2o, *r2e, *r3t;
1274 	struct page *page;
1275 
1276 	BUG_ON(!gmap_is_shadow(sg));
1277 	r2e = gmap_table_walk(sg, raddr, 3); /* get region-2 pointer */
1278 	if (!r2e || !(*r2e & _REGION_ENTRY_ORIGIN))
1279 		return;
1280 	gmap_call_notifier(sg, raddr, raddr + (1UL << 42) - 1);
1281 	r2o = (unsigned long) (r2e - ((raddr >> 42) & 0x7ff));
1282 	gmap_idte_one(r2o | _ASCE_TYPE_REGION2, raddr);
1283 	r3t = (unsigned long *)(*r2e & _REGION_ENTRY_ORIGIN);
1284 	*r2e = _REGION2_ENTRY_EMPTY;
1285 	__gmap_unshadow_r3t(sg, raddr, r3t);
1286 	/* Free region 3 table */
1287 	page = pfn_to_page(__pa(r3t) >> PAGE_SHIFT);
1288 	list_del(&page->lru);
1289 	__free_pages(page, 2);
1290 }
1291 
1292 /**
1293  * __gmap_unshadow_r2t - remove all entries from a shadow region-2 table
1294  * @sg: pointer to the shadow guest address space structure
1295  * @raddr: rmap address in the shadow guest address space
1296  * @r2t: pointer to the start of a shadow region-2 table
1297  *
1298  * Called with the sg->guest_table_lock
1299  */
1300 static void __gmap_unshadow_r2t(struct gmap *sg, unsigned long raddr,
1301 				unsigned long *r2t)
1302 {
1303 	unsigned long asce, *r3t;
1304 	struct page *page;
1305 	int i;
1306 
1307 	BUG_ON(!gmap_is_shadow(sg));
1308 	asce = (unsigned long) r2t | _ASCE_TYPE_REGION2;
1309 	for (i = 0; i < 2048; i++, raddr += 1UL << 42) {
1310 		if (!(r2t[i] & _REGION_ENTRY_ORIGIN))
1311 			continue;
1312 		r3t = (unsigned long *)(r2t[i] & _REGION_ENTRY_ORIGIN);
1313 		r2t[i] = _REGION2_ENTRY_EMPTY;
1314 		__gmap_unshadow_r3t(sg, raddr, r3t);
1315 		/* Free region 3 table */
1316 		page = pfn_to_page(__pa(r3t) >> PAGE_SHIFT);
1317 		list_del(&page->lru);
1318 		__free_pages(page, 2);
1319 	}
1320 }
1321 
1322 /**
1323  * gmap_unshadow_r2t - remove a shadow region-2 table from a region-1 entry
1324  * @sg: pointer to the shadow guest address space structure
1325  * @raddr: rmap address in the shadow guest address space
1326  *
1327  * Called with the sg->guest_table_lock
1328  */
1329 static void gmap_unshadow_r2t(struct gmap *sg, unsigned long raddr)
1330 {
1331 	unsigned long r1o, *r1e, *r2t;
1332 	struct page *page;
1333 
1334 	BUG_ON(!gmap_is_shadow(sg));
1335 	r1e = gmap_table_walk(sg, raddr, 4); /* get region-1 pointer */
1336 	if (!r1e || !(*r1e & _REGION_ENTRY_ORIGIN))
1337 		return;
1338 	gmap_call_notifier(sg, raddr, raddr + (1UL << 53) - 1);
1339 	r1o = (unsigned long) (r1e - ((raddr >> 53) & 0x7ff));
1340 	gmap_idte_one(r1o | _ASCE_TYPE_REGION1, raddr);
1341 	r2t = (unsigned long *)(*r1e & _REGION_ENTRY_ORIGIN);
1342 	*r1e = _REGION1_ENTRY_EMPTY;
1343 	__gmap_unshadow_r2t(sg, raddr, r2t);
1344 	/* Free region 2 table */
1345 	page = pfn_to_page(__pa(r2t) >> PAGE_SHIFT);
1346 	list_del(&page->lru);
1347 	__free_pages(page, 2);
1348 }
1349 
1350 /**
1351  * __gmap_unshadow_r1t - remove all entries from a shadow region-1 table
1352  * @sg: pointer to the shadow guest address space structure
1353  * @raddr: rmap address in the shadow guest address space
1354  * @r1t: pointer to the start of a shadow region-1 table
1355  *
1356  * Called with the shadow->guest_table_lock
1357  */
1358 static void __gmap_unshadow_r1t(struct gmap *sg, unsigned long raddr,
1359 				unsigned long *r1t)
1360 {
1361 	unsigned long asce, *r2t;
1362 	struct page *page;
1363 	int i;
1364 
1365 	BUG_ON(!gmap_is_shadow(sg));
1366 	asce = (unsigned long) r1t | _ASCE_TYPE_REGION1;
1367 	for (i = 0; i < 2048; i++, raddr += 1UL << 53) {
1368 		if (!(r1t[i] & _REGION_ENTRY_ORIGIN))
1369 			continue;
1370 		r2t = (unsigned long *)(r1t[i] & _REGION_ENTRY_ORIGIN);
1371 		__gmap_unshadow_r2t(sg, raddr, r2t);
1372 		/* Clear entry and flush translation r1t -> r2t */
1373 		gmap_idte_one(asce, raddr);
1374 		r1t[i] = _REGION1_ENTRY_EMPTY;
1375 		/* Free region 2 table */
1376 		page = pfn_to_page(__pa(r2t) >> PAGE_SHIFT);
1377 		list_del(&page->lru);
1378 		__free_pages(page, 2);
1379 	}
1380 }
1381 
1382 /**
1383  * gmap_unshadow - remove a shadow page table completely
1384  * @sg: pointer to the shadow guest address space structure
1385  *
1386  * Called with sg->guest_table_lock
1387  */
1388 static void gmap_unshadow(struct gmap *sg)
1389 {
1390 	unsigned long *table;
1391 
1392 	BUG_ON(!gmap_is_shadow(sg));
1393 	if (sg->removed)
1394 		return;
1395 	sg->removed = 1;
1396 	gmap_call_notifier(sg, 0, -1UL);
1397 	gmap_flush_tlb(sg);
1398 	table = (unsigned long *)(sg->asce & _ASCE_ORIGIN);
1399 	switch (sg->asce & _ASCE_TYPE_MASK) {
1400 	case _ASCE_TYPE_REGION1:
1401 		__gmap_unshadow_r1t(sg, 0, table);
1402 		break;
1403 	case _ASCE_TYPE_REGION2:
1404 		__gmap_unshadow_r2t(sg, 0, table);
1405 		break;
1406 	case _ASCE_TYPE_REGION3:
1407 		__gmap_unshadow_r3t(sg, 0, table);
1408 		break;
1409 	case _ASCE_TYPE_SEGMENT:
1410 		__gmap_unshadow_sgt(sg, 0, table);
1411 		break;
1412 	}
1413 }
1414 
1415 /**
1416  * gmap_find_shadow - find a specific asce in the list of shadow tables
1417  * @parent: pointer to the parent gmap
1418  * @asce: ASCE for which the shadow table is created
1419  * @edat_level: edat level to be used for the shadow translation
1420  *
1421  * Returns the pointer to a gmap if a shadow table with the given asce is
1422  * already available, ERR_PTR(-EAGAIN) if another one is just being created,
1423  * otherwise NULL
1424  */
1425 static struct gmap *gmap_find_shadow(struct gmap *parent, unsigned long asce,
1426 				     int edat_level)
1427 {
1428 	struct gmap *sg;
1429 
1430 	list_for_each_entry(sg, &parent->children, list) {
1431 		if (sg->orig_asce != asce || sg->edat_level != edat_level ||
1432 		    sg->removed)
1433 			continue;
1434 		if (!sg->initialized)
1435 			return ERR_PTR(-EAGAIN);
1436 		atomic_inc(&sg->ref_count);
1437 		return sg;
1438 	}
1439 	return NULL;
1440 }
1441 
1442 /**
1443  * gmap_shadow_valid - check if a shadow guest address space matches the
1444  *                     given properties and is still valid
1445  * @sg: pointer to the shadow guest address space structure
1446  * @asce: ASCE for which the shadow table is requested
1447  * @edat_level: edat level to be used for the shadow translation
1448  *
1449  * Returns 1 if the gmap shadow is still valid and matches the given
1450  * properties, the caller can continue using it. Returns 0 otherwise, the
1451  * caller has to request a new shadow gmap in this case.
1452  *
1453  */
1454 int gmap_shadow_valid(struct gmap *sg, unsigned long asce, int edat_level)
1455 {
1456 	if (sg->removed)
1457 		return 0;
1458 	return sg->orig_asce == asce && sg->edat_level == edat_level;
1459 }
1460 EXPORT_SYMBOL_GPL(gmap_shadow_valid);
1461 
1462 /**
1463  * gmap_shadow - create/find a shadow guest address space
1464  * @parent: pointer to the parent gmap
1465  * @asce: ASCE for which the shadow table is created
1466  * @edat_level: edat level to be used for the shadow translation
1467  *
1468  * The pages of the top level page table referred by the asce parameter
1469  * will be set to read-only and marked in the PGSTEs of the kvm process.
1470  * The shadow table will be removed automatically on any change to the
1471  * PTE mapping for the source table.
1472  *
1473  * Returns a guest address space structure, ERR_PTR(-ENOMEM) if out of memory,
1474  * ERR_PTR(-EAGAIN) if the caller has to retry and ERR_PTR(-EFAULT) if the
1475  * parent gmap table could not be protected.
1476  */
1477 struct gmap *gmap_shadow(struct gmap *parent, unsigned long asce,
1478 			 int edat_level)
1479 {
1480 	struct gmap *sg, *new;
1481 	unsigned long limit;
1482 	int rc;
1483 
1484 	BUG_ON(gmap_is_shadow(parent));
1485 	spin_lock(&parent->shadow_lock);
1486 	sg = gmap_find_shadow(parent, asce, edat_level);
1487 	spin_unlock(&parent->shadow_lock);
1488 	if (sg)
1489 		return sg;
1490 	/* Create a new shadow gmap */
1491 	limit = -1UL >> (33 - (((asce & _ASCE_TYPE_MASK) >> 2) * 11));
1492 	if (asce & _ASCE_REAL_SPACE)
1493 		limit = -1UL;
1494 	new = gmap_alloc(limit);
1495 	if (!new)
1496 		return ERR_PTR(-ENOMEM);
1497 	new->mm = parent->mm;
1498 	new->parent = gmap_get(parent);
1499 	new->orig_asce = asce;
1500 	new->edat_level = edat_level;
1501 	new->initialized = false;
1502 	spin_lock(&parent->shadow_lock);
1503 	/* Recheck if another CPU created the same shadow */
1504 	sg = gmap_find_shadow(parent, asce, edat_level);
1505 	if (sg) {
1506 		spin_unlock(&parent->shadow_lock);
1507 		gmap_free(new);
1508 		return sg;
1509 	}
1510 	if (asce & _ASCE_REAL_SPACE) {
1511 		/* only allow one real-space gmap shadow */
1512 		list_for_each_entry(sg, &parent->children, list) {
1513 			if (sg->orig_asce & _ASCE_REAL_SPACE) {
1514 				spin_lock(&sg->guest_table_lock);
1515 				gmap_unshadow(sg);
1516 				spin_unlock(&sg->guest_table_lock);
1517 				list_del(&sg->list);
1518 				gmap_put(sg);
1519 				break;
1520 			}
1521 		}
1522 	}
1523 	atomic_set(&new->ref_count, 2);
1524 	list_add(&new->list, &parent->children);
1525 	if (asce & _ASCE_REAL_SPACE) {
1526 		/* nothing to protect, return right away */
1527 		new->initialized = true;
1528 		spin_unlock(&parent->shadow_lock);
1529 		return new;
1530 	}
1531 	spin_unlock(&parent->shadow_lock);
1532 	/* protect after insertion, so it will get properly invalidated */
1533 	down_read(&parent->mm->mmap_sem);
1534 	rc = gmap_protect_range(parent, asce & _ASCE_ORIGIN,
1535 				((asce & _ASCE_TABLE_LENGTH) + 1) * 4096,
1536 				PROT_READ, PGSTE_VSIE_BIT);
1537 	up_read(&parent->mm->mmap_sem);
1538 	spin_lock(&parent->shadow_lock);
1539 	new->initialized = true;
1540 	if (rc) {
1541 		list_del(&new->list);
1542 		gmap_free(new);
1543 		new = ERR_PTR(rc);
1544 	}
1545 	spin_unlock(&parent->shadow_lock);
1546 	return new;
1547 }
1548 EXPORT_SYMBOL_GPL(gmap_shadow);
1549 
1550 /**
1551  * gmap_shadow_r2t - create an empty shadow region 2 table
1552  * @sg: pointer to the shadow guest address space structure
1553  * @saddr: faulting address in the shadow gmap
1554  * @r2t: parent gmap address of the region 2 table to get shadowed
1555  * @fake: r2t references contiguous guest memory block, not a r2t
1556  *
1557  * The r2t parameter specifies the address of the source table. The
1558  * four pages of the source table are made read-only in the parent gmap
1559  * address space. A write to the source table area @r2t will automatically
1560  * remove the shadow r2 table and all of its decendents.
1561  *
1562  * Returns 0 if successfully shadowed or already shadowed, -EAGAIN if the
1563  * shadow table structure is incomplete, -ENOMEM if out of memory and
1564  * -EFAULT if an address in the parent gmap could not be resolved.
1565  *
1566  * Called with sg->mm->mmap_sem in read.
1567  */
1568 int gmap_shadow_r2t(struct gmap *sg, unsigned long saddr, unsigned long r2t,
1569 		    int fake)
1570 {
1571 	unsigned long raddr, origin, offset, len;
1572 	unsigned long *s_r2t, *table;
1573 	struct page *page;
1574 	int rc;
1575 
1576 	BUG_ON(!gmap_is_shadow(sg));
1577 	/* Allocate a shadow region second table */
1578 	page = alloc_pages(GFP_KERNEL, 2);
1579 	if (!page)
1580 		return -ENOMEM;
1581 	page->index = r2t & _REGION_ENTRY_ORIGIN;
1582 	if (fake)
1583 		page->index |= GMAP_SHADOW_FAKE_TABLE;
1584 	s_r2t = (unsigned long *) page_to_phys(page);
1585 	/* Install shadow region second table */
1586 	spin_lock(&sg->guest_table_lock);
1587 	table = gmap_table_walk(sg, saddr, 4); /* get region-1 pointer */
1588 	if (!table) {
1589 		rc = -EAGAIN;		/* Race with unshadow */
1590 		goto out_free;
1591 	}
1592 	if (!(*table & _REGION_ENTRY_INVALID)) {
1593 		rc = 0;			/* Already established */
1594 		goto out_free;
1595 	} else if (*table & _REGION_ENTRY_ORIGIN) {
1596 		rc = -EAGAIN;		/* Race with shadow */
1597 		goto out_free;
1598 	}
1599 	crst_table_init(s_r2t, _REGION2_ENTRY_EMPTY);
1600 	/* mark as invalid as long as the parent table is not protected */
1601 	*table = (unsigned long) s_r2t | _REGION_ENTRY_LENGTH |
1602 		 _REGION_ENTRY_TYPE_R1 | _REGION_ENTRY_INVALID;
1603 	if (sg->edat_level >= 1)
1604 		*table |= (r2t & _REGION_ENTRY_PROTECT);
1605 	list_add(&page->lru, &sg->crst_list);
1606 	if (fake) {
1607 		/* nothing to protect for fake tables */
1608 		*table &= ~_REGION_ENTRY_INVALID;
1609 		spin_unlock(&sg->guest_table_lock);
1610 		return 0;
1611 	}
1612 	spin_unlock(&sg->guest_table_lock);
1613 	/* Make r2t read-only in parent gmap page table */
1614 	raddr = (saddr & 0xffe0000000000000UL) | _SHADOW_RMAP_REGION1;
1615 	origin = r2t & _REGION_ENTRY_ORIGIN;
1616 	offset = ((r2t & _REGION_ENTRY_OFFSET) >> 6) * 4096;
1617 	len = ((r2t & _REGION_ENTRY_LENGTH) + 1) * 4096 - offset;
1618 	rc = gmap_protect_rmap(sg, raddr, origin + offset, len, PROT_READ);
1619 	spin_lock(&sg->guest_table_lock);
1620 	if (!rc) {
1621 		table = gmap_table_walk(sg, saddr, 4);
1622 		if (!table || (*table & _REGION_ENTRY_ORIGIN) !=
1623 			      (unsigned long) s_r2t)
1624 			rc = -EAGAIN;		/* Race with unshadow */
1625 		else
1626 			*table &= ~_REGION_ENTRY_INVALID;
1627 	} else {
1628 		gmap_unshadow_r2t(sg, raddr);
1629 	}
1630 	spin_unlock(&sg->guest_table_lock);
1631 	return rc;
1632 out_free:
1633 	spin_unlock(&sg->guest_table_lock);
1634 	__free_pages(page, 2);
1635 	return rc;
1636 }
1637 EXPORT_SYMBOL_GPL(gmap_shadow_r2t);
1638 
1639 /**
1640  * gmap_shadow_r3t - create a shadow region 3 table
1641  * @sg: pointer to the shadow guest address space structure
1642  * @saddr: faulting address in the shadow gmap
1643  * @r3t: parent gmap address of the region 3 table to get shadowed
1644  * @fake: r3t references contiguous guest memory block, not a r3t
1645  *
1646  * Returns 0 if successfully shadowed or already shadowed, -EAGAIN if the
1647  * shadow table structure is incomplete, -ENOMEM if out of memory and
1648  * -EFAULT if an address in the parent gmap could not be resolved.
1649  *
1650  * Called with sg->mm->mmap_sem in read.
1651  */
1652 int gmap_shadow_r3t(struct gmap *sg, unsigned long saddr, unsigned long r3t,
1653 		    int fake)
1654 {
1655 	unsigned long raddr, origin, offset, len;
1656 	unsigned long *s_r3t, *table;
1657 	struct page *page;
1658 	int rc;
1659 
1660 	BUG_ON(!gmap_is_shadow(sg));
1661 	/* Allocate a shadow region second table */
1662 	page = alloc_pages(GFP_KERNEL, 2);
1663 	if (!page)
1664 		return -ENOMEM;
1665 	page->index = r3t & _REGION_ENTRY_ORIGIN;
1666 	if (fake)
1667 		page->index |= GMAP_SHADOW_FAKE_TABLE;
1668 	s_r3t = (unsigned long *) page_to_phys(page);
1669 	/* Install shadow region second table */
1670 	spin_lock(&sg->guest_table_lock);
1671 	table = gmap_table_walk(sg, saddr, 3); /* get region-2 pointer */
1672 	if (!table) {
1673 		rc = -EAGAIN;		/* Race with unshadow */
1674 		goto out_free;
1675 	}
1676 	if (!(*table & _REGION_ENTRY_INVALID)) {
1677 		rc = 0;			/* Already established */
1678 		goto out_free;
1679 	} else if (*table & _REGION_ENTRY_ORIGIN) {
1680 		rc = -EAGAIN;		/* Race with shadow */
1681 	}
1682 	crst_table_init(s_r3t, _REGION3_ENTRY_EMPTY);
1683 	/* mark as invalid as long as the parent table is not protected */
1684 	*table = (unsigned long) s_r3t | _REGION_ENTRY_LENGTH |
1685 		 _REGION_ENTRY_TYPE_R2 | _REGION_ENTRY_INVALID;
1686 	if (sg->edat_level >= 1)
1687 		*table |= (r3t & _REGION_ENTRY_PROTECT);
1688 	list_add(&page->lru, &sg->crst_list);
1689 	if (fake) {
1690 		/* nothing to protect for fake tables */
1691 		*table &= ~_REGION_ENTRY_INVALID;
1692 		spin_unlock(&sg->guest_table_lock);
1693 		return 0;
1694 	}
1695 	spin_unlock(&sg->guest_table_lock);
1696 	/* Make r3t read-only in parent gmap page table */
1697 	raddr = (saddr & 0xfffffc0000000000UL) | _SHADOW_RMAP_REGION2;
1698 	origin = r3t & _REGION_ENTRY_ORIGIN;
1699 	offset = ((r3t & _REGION_ENTRY_OFFSET) >> 6) * 4096;
1700 	len = ((r3t & _REGION_ENTRY_LENGTH) + 1) * 4096 - offset;
1701 	rc = gmap_protect_rmap(sg, raddr, origin + offset, len, PROT_READ);
1702 	spin_lock(&sg->guest_table_lock);
1703 	if (!rc) {
1704 		table = gmap_table_walk(sg, saddr, 3);
1705 		if (!table || (*table & _REGION_ENTRY_ORIGIN) !=
1706 			      (unsigned long) s_r3t)
1707 			rc = -EAGAIN;		/* Race with unshadow */
1708 		else
1709 			*table &= ~_REGION_ENTRY_INVALID;
1710 	} else {
1711 		gmap_unshadow_r3t(sg, raddr);
1712 	}
1713 	spin_unlock(&sg->guest_table_lock);
1714 	return rc;
1715 out_free:
1716 	spin_unlock(&sg->guest_table_lock);
1717 	__free_pages(page, 2);
1718 	return rc;
1719 }
1720 EXPORT_SYMBOL_GPL(gmap_shadow_r3t);
1721 
1722 /**
1723  * gmap_shadow_sgt - create a shadow segment table
1724  * @sg: pointer to the shadow guest address space structure
1725  * @saddr: faulting address in the shadow gmap
1726  * @sgt: parent gmap address of the segment table to get shadowed
1727  * @fake: sgt references contiguous guest memory block, not a sgt
1728  *
1729  * Returns: 0 if successfully shadowed or already shadowed, -EAGAIN if the
1730  * shadow table structure is incomplete, -ENOMEM if out of memory and
1731  * -EFAULT if an address in the parent gmap could not be resolved.
1732  *
1733  * Called with sg->mm->mmap_sem in read.
1734  */
1735 int gmap_shadow_sgt(struct gmap *sg, unsigned long saddr, unsigned long sgt,
1736 		    int fake)
1737 {
1738 	unsigned long raddr, origin, offset, len;
1739 	unsigned long *s_sgt, *table;
1740 	struct page *page;
1741 	int rc;
1742 
1743 	BUG_ON(!gmap_is_shadow(sg) || (sgt & _REGION3_ENTRY_LARGE));
1744 	/* Allocate a shadow segment table */
1745 	page = alloc_pages(GFP_KERNEL, 2);
1746 	if (!page)
1747 		return -ENOMEM;
1748 	page->index = sgt & _REGION_ENTRY_ORIGIN;
1749 	if (fake)
1750 		page->index |= GMAP_SHADOW_FAKE_TABLE;
1751 	s_sgt = (unsigned long *) page_to_phys(page);
1752 	/* Install shadow region second table */
1753 	spin_lock(&sg->guest_table_lock);
1754 	table = gmap_table_walk(sg, saddr, 2); /* get region-3 pointer */
1755 	if (!table) {
1756 		rc = -EAGAIN;		/* Race with unshadow */
1757 		goto out_free;
1758 	}
1759 	if (!(*table & _REGION_ENTRY_INVALID)) {
1760 		rc = 0;			/* Already established */
1761 		goto out_free;
1762 	} else if (*table & _REGION_ENTRY_ORIGIN) {
1763 		rc = -EAGAIN;		/* Race with shadow */
1764 		goto out_free;
1765 	}
1766 	crst_table_init(s_sgt, _SEGMENT_ENTRY_EMPTY);
1767 	/* mark as invalid as long as the parent table is not protected */
1768 	*table = (unsigned long) s_sgt | _REGION_ENTRY_LENGTH |
1769 		 _REGION_ENTRY_TYPE_R3 | _REGION_ENTRY_INVALID;
1770 	if (sg->edat_level >= 1)
1771 		*table |= sgt & _REGION_ENTRY_PROTECT;
1772 	list_add(&page->lru, &sg->crst_list);
1773 	if (fake) {
1774 		/* nothing to protect for fake tables */
1775 		*table &= ~_REGION_ENTRY_INVALID;
1776 		spin_unlock(&sg->guest_table_lock);
1777 		return 0;
1778 	}
1779 	spin_unlock(&sg->guest_table_lock);
1780 	/* Make sgt read-only in parent gmap page table */
1781 	raddr = (saddr & 0xffffffff80000000UL) | _SHADOW_RMAP_REGION3;
1782 	origin = sgt & _REGION_ENTRY_ORIGIN;
1783 	offset = ((sgt & _REGION_ENTRY_OFFSET) >> 6) * 4096;
1784 	len = ((sgt & _REGION_ENTRY_LENGTH) + 1) * 4096 - offset;
1785 	rc = gmap_protect_rmap(sg, raddr, origin + offset, len, PROT_READ);
1786 	spin_lock(&sg->guest_table_lock);
1787 	if (!rc) {
1788 		table = gmap_table_walk(sg, saddr, 2);
1789 		if (!table || (*table & _REGION_ENTRY_ORIGIN) !=
1790 			      (unsigned long) s_sgt)
1791 			rc = -EAGAIN;		/* Race with unshadow */
1792 		else
1793 			*table &= ~_REGION_ENTRY_INVALID;
1794 	} else {
1795 		gmap_unshadow_sgt(sg, raddr);
1796 	}
1797 	spin_unlock(&sg->guest_table_lock);
1798 	return rc;
1799 out_free:
1800 	spin_unlock(&sg->guest_table_lock);
1801 	__free_pages(page, 2);
1802 	return rc;
1803 }
1804 EXPORT_SYMBOL_GPL(gmap_shadow_sgt);
1805 
1806 /**
1807  * gmap_shadow_lookup_pgtable - find a shadow page table
1808  * @sg: pointer to the shadow guest address space structure
1809  * @saddr: the address in the shadow aguest address space
1810  * @pgt: parent gmap address of the page table to get shadowed
1811  * @dat_protection: if the pgtable is marked as protected by dat
1812  * @fake: pgt references contiguous guest memory block, not a pgtable
1813  *
1814  * Returns 0 if the shadow page table was found and -EAGAIN if the page
1815  * table was not found.
1816  *
1817  * Called with sg->mm->mmap_sem in read.
1818  */
1819 int gmap_shadow_pgt_lookup(struct gmap *sg, unsigned long saddr,
1820 			   unsigned long *pgt, int *dat_protection,
1821 			   int *fake)
1822 {
1823 	unsigned long *table;
1824 	struct page *page;
1825 	int rc;
1826 
1827 	BUG_ON(!gmap_is_shadow(sg));
1828 	spin_lock(&sg->guest_table_lock);
1829 	table = gmap_table_walk(sg, saddr, 1); /* get segment pointer */
1830 	if (table && !(*table & _SEGMENT_ENTRY_INVALID)) {
1831 		/* Shadow page tables are full pages (pte+pgste) */
1832 		page = pfn_to_page(*table >> PAGE_SHIFT);
1833 		*pgt = page->index & ~GMAP_SHADOW_FAKE_TABLE;
1834 		*dat_protection = !!(*table & _SEGMENT_ENTRY_PROTECT);
1835 		*fake = !!(page->index & GMAP_SHADOW_FAKE_TABLE);
1836 		rc = 0;
1837 	} else  {
1838 		rc = -EAGAIN;
1839 	}
1840 	spin_unlock(&sg->guest_table_lock);
1841 	return rc;
1842 
1843 }
1844 EXPORT_SYMBOL_GPL(gmap_shadow_pgt_lookup);
1845 
1846 /**
1847  * gmap_shadow_pgt - instantiate a shadow page table
1848  * @sg: pointer to the shadow guest address space structure
1849  * @saddr: faulting address in the shadow gmap
1850  * @pgt: parent gmap address of the page table to get shadowed
1851  * @fake: pgt references contiguous guest memory block, not a pgtable
1852  *
1853  * Returns 0 if successfully shadowed or already shadowed, -EAGAIN if the
1854  * shadow table structure is incomplete, -ENOMEM if out of memory,
1855  * -EFAULT if an address in the parent gmap could not be resolved and
1856  *
1857  * Called with gmap->mm->mmap_sem in read
1858  */
1859 int gmap_shadow_pgt(struct gmap *sg, unsigned long saddr, unsigned long pgt,
1860 		    int fake)
1861 {
1862 	unsigned long raddr, origin;
1863 	unsigned long *s_pgt, *table;
1864 	struct page *page;
1865 	int rc;
1866 
1867 	BUG_ON(!gmap_is_shadow(sg) || (pgt & _SEGMENT_ENTRY_LARGE));
1868 	/* Allocate a shadow page table */
1869 	page = page_table_alloc_pgste(sg->mm);
1870 	if (!page)
1871 		return -ENOMEM;
1872 	page->index = pgt & _SEGMENT_ENTRY_ORIGIN;
1873 	if (fake)
1874 		page->index |= GMAP_SHADOW_FAKE_TABLE;
1875 	s_pgt = (unsigned long *) page_to_phys(page);
1876 	/* Install shadow page table */
1877 	spin_lock(&sg->guest_table_lock);
1878 	table = gmap_table_walk(sg, saddr, 1); /* get segment pointer */
1879 	if (!table) {
1880 		rc = -EAGAIN;		/* Race with unshadow */
1881 		goto out_free;
1882 	}
1883 	if (!(*table & _SEGMENT_ENTRY_INVALID)) {
1884 		rc = 0;			/* Already established */
1885 		goto out_free;
1886 	} else if (*table & _SEGMENT_ENTRY_ORIGIN) {
1887 		rc = -EAGAIN;		/* Race with shadow */
1888 		goto out_free;
1889 	}
1890 	/* mark as invalid as long as the parent table is not protected */
1891 	*table = (unsigned long) s_pgt | _SEGMENT_ENTRY |
1892 		 (pgt & _SEGMENT_ENTRY_PROTECT) | _SEGMENT_ENTRY_INVALID;
1893 	list_add(&page->lru, &sg->pt_list);
1894 	if (fake) {
1895 		/* nothing to protect for fake tables */
1896 		*table &= ~_SEGMENT_ENTRY_INVALID;
1897 		spin_unlock(&sg->guest_table_lock);
1898 		return 0;
1899 	}
1900 	spin_unlock(&sg->guest_table_lock);
1901 	/* Make pgt read-only in parent gmap page table (not the pgste) */
1902 	raddr = (saddr & 0xfffffffffff00000UL) | _SHADOW_RMAP_SEGMENT;
1903 	origin = pgt & _SEGMENT_ENTRY_ORIGIN & PAGE_MASK;
1904 	rc = gmap_protect_rmap(sg, raddr, origin, PAGE_SIZE, PROT_READ);
1905 	spin_lock(&sg->guest_table_lock);
1906 	if (!rc) {
1907 		table = gmap_table_walk(sg, saddr, 1);
1908 		if (!table || (*table & _SEGMENT_ENTRY_ORIGIN) !=
1909 			      (unsigned long) s_pgt)
1910 			rc = -EAGAIN;		/* Race with unshadow */
1911 		else
1912 			*table &= ~_SEGMENT_ENTRY_INVALID;
1913 	} else {
1914 		gmap_unshadow_pgt(sg, raddr);
1915 	}
1916 	spin_unlock(&sg->guest_table_lock);
1917 	return rc;
1918 out_free:
1919 	spin_unlock(&sg->guest_table_lock);
1920 	page_table_free_pgste(page);
1921 	return rc;
1922 
1923 }
1924 EXPORT_SYMBOL_GPL(gmap_shadow_pgt);
1925 
1926 /**
1927  * gmap_shadow_page - create a shadow page mapping
1928  * @sg: pointer to the shadow guest address space structure
1929  * @saddr: faulting address in the shadow gmap
1930  * @pte: pte in parent gmap address space to get shadowed
1931  *
1932  * Returns 0 if successfully shadowed or already shadowed, -EAGAIN if the
1933  * shadow table structure is incomplete, -ENOMEM if out of memory and
1934  * -EFAULT if an address in the parent gmap could not be resolved.
1935  *
1936  * Called with sg->mm->mmap_sem in read.
1937  */
1938 int gmap_shadow_page(struct gmap *sg, unsigned long saddr, pte_t pte)
1939 {
1940 	struct gmap *parent;
1941 	struct gmap_rmap *rmap;
1942 	unsigned long vmaddr, paddr;
1943 	spinlock_t *ptl;
1944 	pte_t *sptep, *tptep;
1945 	int prot;
1946 	int rc;
1947 
1948 	BUG_ON(!gmap_is_shadow(sg));
1949 	parent = sg->parent;
1950 	prot = (pte_val(pte) & _PAGE_PROTECT) ? PROT_READ : PROT_WRITE;
1951 
1952 	rmap = kzalloc(sizeof(*rmap), GFP_KERNEL);
1953 	if (!rmap)
1954 		return -ENOMEM;
1955 	rmap->raddr = (saddr & PAGE_MASK) | _SHADOW_RMAP_PGTABLE;
1956 
1957 	while (1) {
1958 		paddr = pte_val(pte) & PAGE_MASK;
1959 		vmaddr = __gmap_translate(parent, paddr);
1960 		if (IS_ERR_VALUE(vmaddr)) {
1961 			rc = vmaddr;
1962 			break;
1963 		}
1964 		rc = radix_tree_preload(GFP_KERNEL);
1965 		if (rc)
1966 			break;
1967 		rc = -EAGAIN;
1968 		sptep = gmap_pte_op_walk(parent, paddr, &ptl);
1969 		if (sptep) {
1970 			spin_lock(&sg->guest_table_lock);
1971 			/* Get page table pointer */
1972 			tptep = (pte_t *) gmap_table_walk(sg, saddr, 0);
1973 			if (!tptep) {
1974 				spin_unlock(&sg->guest_table_lock);
1975 				gmap_pte_op_end(ptl);
1976 				radix_tree_preload_end();
1977 				break;
1978 			}
1979 			rc = ptep_shadow_pte(sg->mm, saddr, sptep, tptep, pte);
1980 			if (rc > 0) {
1981 				/* Success and a new mapping */
1982 				gmap_insert_rmap(sg, vmaddr, rmap);
1983 				rmap = NULL;
1984 				rc = 0;
1985 			}
1986 			gmap_pte_op_end(ptl);
1987 			spin_unlock(&sg->guest_table_lock);
1988 		}
1989 		radix_tree_preload_end();
1990 		if (!rc)
1991 			break;
1992 		rc = gmap_pte_op_fixup(parent, paddr, vmaddr, prot);
1993 		if (rc)
1994 			break;
1995 	}
1996 	kfree(rmap);
1997 	return rc;
1998 }
1999 EXPORT_SYMBOL_GPL(gmap_shadow_page);
2000 
2001 /**
2002  * gmap_shadow_notify - handle notifications for shadow gmap
2003  *
2004  * Called with sg->parent->shadow_lock.
2005  */
2006 static void gmap_shadow_notify(struct gmap *sg, unsigned long vmaddr,
2007 			       unsigned long offset, pte_t *pte)
2008 {
2009 	struct gmap_rmap *rmap, *rnext, *head;
2010 	unsigned long gaddr, start, end, bits, raddr;
2011 	unsigned long *table;
2012 
2013 	BUG_ON(!gmap_is_shadow(sg));
2014 	spin_lock(&sg->parent->guest_table_lock);
2015 	table = radix_tree_lookup(&sg->parent->host_to_guest,
2016 				  vmaddr >> PMD_SHIFT);
2017 	gaddr = table ? __gmap_segment_gaddr(table) + offset : 0;
2018 	spin_unlock(&sg->parent->guest_table_lock);
2019 	if (!table)
2020 		return;
2021 
2022 	spin_lock(&sg->guest_table_lock);
2023 	if (sg->removed) {
2024 		spin_unlock(&sg->guest_table_lock);
2025 		return;
2026 	}
2027 	/* Check for top level table */
2028 	start = sg->orig_asce & _ASCE_ORIGIN;
2029 	end = start + ((sg->orig_asce & _ASCE_TABLE_LENGTH) + 1) * 4096;
2030 	if (!(sg->orig_asce & _ASCE_REAL_SPACE) && gaddr >= start &&
2031 	    gaddr < end) {
2032 		/* The complete shadow table has to go */
2033 		gmap_unshadow(sg);
2034 		spin_unlock(&sg->guest_table_lock);
2035 		list_del(&sg->list);
2036 		gmap_put(sg);
2037 		return;
2038 	}
2039 	/* Remove the page table tree from on specific entry */
2040 	head = radix_tree_delete(&sg->host_to_rmap, vmaddr >> 12);
2041 	gmap_for_each_rmap_safe(rmap, rnext, head) {
2042 		bits = rmap->raddr & _SHADOW_RMAP_MASK;
2043 		raddr = rmap->raddr ^ bits;
2044 		switch (bits) {
2045 		case _SHADOW_RMAP_REGION1:
2046 			gmap_unshadow_r2t(sg, raddr);
2047 			break;
2048 		case _SHADOW_RMAP_REGION2:
2049 			gmap_unshadow_r3t(sg, raddr);
2050 			break;
2051 		case _SHADOW_RMAP_REGION3:
2052 			gmap_unshadow_sgt(sg, raddr);
2053 			break;
2054 		case _SHADOW_RMAP_SEGMENT:
2055 			gmap_unshadow_pgt(sg, raddr);
2056 			break;
2057 		case _SHADOW_RMAP_PGTABLE:
2058 			gmap_unshadow_page(sg, raddr);
2059 			break;
2060 		}
2061 		kfree(rmap);
2062 	}
2063 	spin_unlock(&sg->guest_table_lock);
2064 }
2065 
2066 /**
2067  * ptep_notify - call all invalidation callbacks for a specific pte.
2068  * @mm: pointer to the process mm_struct
2069  * @addr: virtual address in the process address space
2070  * @pte: pointer to the page table entry
2071  * @bits: bits from the pgste that caused the notify call
2072  *
2073  * This function is assumed to be called with the page table lock held
2074  * for the pte to notify.
2075  */
2076 void ptep_notify(struct mm_struct *mm, unsigned long vmaddr,
2077 		 pte_t *pte, unsigned long bits)
2078 {
2079 	unsigned long offset, gaddr;
2080 	unsigned long *table;
2081 	struct gmap *gmap, *sg, *next;
2082 
2083 	offset = ((unsigned long) pte) & (255 * sizeof(pte_t));
2084 	offset = offset * (4096 / sizeof(pte_t));
2085 	rcu_read_lock();
2086 	list_for_each_entry_rcu(gmap, &mm->context.gmap_list, list) {
2087 		if (!list_empty(&gmap->children) && (bits & PGSTE_VSIE_BIT)) {
2088 			spin_lock(&gmap->shadow_lock);
2089 			list_for_each_entry_safe(sg, next,
2090 						 &gmap->children, list)
2091 				gmap_shadow_notify(sg, vmaddr, offset, pte);
2092 			spin_unlock(&gmap->shadow_lock);
2093 		}
2094 		if (!(bits & PGSTE_IN_BIT))
2095 			continue;
2096 		spin_lock(&gmap->guest_table_lock);
2097 		table = radix_tree_lookup(&gmap->host_to_guest,
2098 					  vmaddr >> PMD_SHIFT);
2099 		if (table)
2100 			gaddr = __gmap_segment_gaddr(table) + offset;
2101 		spin_unlock(&gmap->guest_table_lock);
2102 		if (table)
2103 			gmap_call_notifier(gmap, gaddr, gaddr + PAGE_SIZE - 1);
2104 	}
2105 	rcu_read_unlock();
2106 }
2107 EXPORT_SYMBOL_GPL(ptep_notify);
2108 
2109 static inline void thp_split_mm(struct mm_struct *mm)
2110 {
2111 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
2112 	struct vm_area_struct *vma;
2113 	unsigned long addr;
2114 
2115 	for (vma = mm->mmap; vma != NULL; vma = vma->vm_next) {
2116 		for (addr = vma->vm_start;
2117 		     addr < vma->vm_end;
2118 		     addr += PAGE_SIZE)
2119 			follow_page(vma, addr, FOLL_SPLIT);
2120 		vma->vm_flags &= ~VM_HUGEPAGE;
2121 		vma->vm_flags |= VM_NOHUGEPAGE;
2122 	}
2123 	mm->def_flags |= VM_NOHUGEPAGE;
2124 #endif
2125 }
2126 
2127 /*
2128  * switch on pgstes for its userspace process (for kvm)
2129  */
2130 int s390_enable_sie(void)
2131 {
2132 	struct mm_struct *mm = current->mm;
2133 
2134 	/* Do we have pgstes? if yes, we are done */
2135 	if (mm_has_pgste(mm))
2136 		return 0;
2137 	/* Fail if the page tables are 2K */
2138 	if (!mm_alloc_pgste(mm))
2139 		return -EINVAL;
2140 	down_write(&mm->mmap_sem);
2141 	mm->context.has_pgste = 1;
2142 	/* split thp mappings and disable thp for future mappings */
2143 	thp_split_mm(mm);
2144 	up_write(&mm->mmap_sem);
2145 	return 0;
2146 }
2147 EXPORT_SYMBOL_GPL(s390_enable_sie);
2148 
2149 /*
2150  * Enable storage key handling from now on and initialize the storage
2151  * keys with the default key.
2152  */
2153 static int __s390_enable_skey(pte_t *pte, unsigned long addr,
2154 			      unsigned long next, struct mm_walk *walk)
2155 {
2156 	/*
2157 	 * Remove all zero page mappings,
2158 	 * after establishing a policy to forbid zero page mappings
2159 	 * following faults for that page will get fresh anonymous pages
2160 	 */
2161 	if (is_zero_pfn(pte_pfn(*pte)))
2162 		ptep_xchg_direct(walk->mm, addr, pte, __pte(_PAGE_INVALID));
2163 	/* Clear storage key */
2164 	ptep_zap_key(walk->mm, addr, pte);
2165 	return 0;
2166 }
2167 
2168 int s390_enable_skey(void)
2169 {
2170 	struct mm_walk walk = { .pte_entry = __s390_enable_skey };
2171 	struct mm_struct *mm = current->mm;
2172 	struct vm_area_struct *vma;
2173 	int rc = 0;
2174 
2175 	down_write(&mm->mmap_sem);
2176 	if (mm_use_skey(mm))
2177 		goto out_up;
2178 
2179 	mm->context.use_skey = 1;
2180 	for (vma = mm->mmap; vma; vma = vma->vm_next) {
2181 		if (ksm_madvise(vma, vma->vm_start, vma->vm_end,
2182 				MADV_UNMERGEABLE, &vma->vm_flags)) {
2183 			mm->context.use_skey = 0;
2184 			rc = -ENOMEM;
2185 			goto out_up;
2186 		}
2187 	}
2188 	mm->def_flags &= ~VM_MERGEABLE;
2189 
2190 	walk.mm = mm;
2191 	walk_page_range(0, TASK_SIZE, &walk);
2192 
2193 out_up:
2194 	up_write(&mm->mmap_sem);
2195 	return rc;
2196 }
2197 EXPORT_SYMBOL_GPL(s390_enable_skey);
2198 
2199 /*
2200  * Reset CMMA state, make all pages stable again.
2201  */
2202 static int __s390_reset_cmma(pte_t *pte, unsigned long addr,
2203 			     unsigned long next, struct mm_walk *walk)
2204 {
2205 	ptep_zap_unused(walk->mm, addr, pte, 1);
2206 	return 0;
2207 }
2208 
2209 void s390_reset_cmma(struct mm_struct *mm)
2210 {
2211 	struct mm_walk walk = { .pte_entry = __s390_reset_cmma };
2212 
2213 	down_write(&mm->mmap_sem);
2214 	walk.mm = mm;
2215 	walk_page_range(0, TASK_SIZE, &walk);
2216 	up_write(&mm->mmap_sem);
2217 }
2218 EXPORT_SYMBOL_GPL(s390_reset_cmma);
2219