xref: /openbmc/linux/arch/s390/mm/gmap.c (revision 80483c3a)
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 
98 	gmap = gmap_alloc(limit);
99 	if (!gmap)
100 		return NULL;
101 	gmap->mm = mm;
102 	spin_lock(&mm->context.gmap_lock);
103 	list_add_rcu(&gmap->list, &mm->context.gmap_list);
104 	spin_unlock(&mm->context.gmap_lock);
105 	return gmap;
106 }
107 EXPORT_SYMBOL_GPL(gmap_create);
108 
109 static void gmap_flush_tlb(struct gmap *gmap)
110 {
111 	if (MACHINE_HAS_IDTE)
112 		__tlb_flush_idte(gmap->asce);
113 	else
114 		__tlb_flush_global();
115 }
116 
117 static void gmap_radix_tree_free(struct radix_tree_root *root)
118 {
119 	struct radix_tree_iter iter;
120 	unsigned long indices[16];
121 	unsigned long index;
122 	void **slot;
123 	int i, nr;
124 
125 	/* A radix tree is freed by deleting all of its entries */
126 	index = 0;
127 	do {
128 		nr = 0;
129 		radix_tree_for_each_slot(slot, root, &iter, index) {
130 			indices[nr] = iter.index;
131 			if (++nr == 16)
132 				break;
133 		}
134 		for (i = 0; i < nr; i++) {
135 			index = indices[i];
136 			radix_tree_delete(root, index);
137 		}
138 	} while (nr > 0);
139 }
140 
141 static void gmap_rmap_radix_tree_free(struct radix_tree_root *root)
142 {
143 	struct gmap_rmap *rmap, *rnext, *head;
144 	struct radix_tree_iter iter;
145 	unsigned long indices[16];
146 	unsigned long index;
147 	void **slot;
148 	int i, nr;
149 
150 	/* A radix tree is freed by deleting all of its entries */
151 	index = 0;
152 	do {
153 		nr = 0;
154 		radix_tree_for_each_slot(slot, root, &iter, index) {
155 			indices[nr] = iter.index;
156 			if (++nr == 16)
157 				break;
158 		}
159 		for (i = 0; i < nr; i++) {
160 			index = indices[i];
161 			head = radix_tree_delete(root, index);
162 			gmap_for_each_rmap_safe(rmap, rnext, head)
163 				kfree(rmap);
164 		}
165 	} while (nr > 0);
166 }
167 
168 /**
169  * gmap_free - free a guest address space
170  * @gmap: pointer to the guest address space structure
171  *
172  * No locks required. There are no references to this gmap anymore.
173  */
174 static void gmap_free(struct gmap *gmap)
175 {
176 	struct page *page, *next;
177 
178 	/* Flush tlb of all gmaps (if not already done for shadows) */
179 	if (!(gmap_is_shadow(gmap) && gmap->removed))
180 		gmap_flush_tlb(gmap);
181 	/* Free all segment & region tables. */
182 	list_for_each_entry_safe(page, next, &gmap->crst_list, lru)
183 		__free_pages(page, 2);
184 	gmap_radix_tree_free(&gmap->guest_to_host);
185 	gmap_radix_tree_free(&gmap->host_to_guest);
186 
187 	/* Free additional data for a shadow gmap */
188 	if (gmap_is_shadow(gmap)) {
189 		/* Free all page tables. */
190 		list_for_each_entry_safe(page, next, &gmap->pt_list, lru)
191 			page_table_free_pgste(page);
192 		gmap_rmap_radix_tree_free(&gmap->host_to_rmap);
193 		/* Release reference to the parent */
194 		gmap_put(gmap->parent);
195 	}
196 
197 	kfree(gmap);
198 }
199 
200 /**
201  * gmap_get - increase reference counter for guest address space
202  * @gmap: pointer to the guest address space structure
203  *
204  * Returns the gmap pointer
205  */
206 struct gmap *gmap_get(struct gmap *gmap)
207 {
208 	atomic_inc(&gmap->ref_count);
209 	return gmap;
210 }
211 EXPORT_SYMBOL_GPL(gmap_get);
212 
213 /**
214  * gmap_put - decrease reference counter for guest address space
215  * @gmap: pointer to the guest address space structure
216  *
217  * If the reference counter reaches zero the guest address space is freed.
218  */
219 void gmap_put(struct gmap *gmap)
220 {
221 	if (atomic_dec_return(&gmap->ref_count) == 0)
222 		gmap_free(gmap);
223 }
224 EXPORT_SYMBOL_GPL(gmap_put);
225 
226 /**
227  * gmap_remove - remove a guest address space but do not free it yet
228  * @gmap: pointer to the guest address space structure
229  */
230 void gmap_remove(struct gmap *gmap)
231 {
232 	struct gmap *sg, *next;
233 
234 	/* Remove all shadow gmaps linked to this gmap */
235 	if (!list_empty(&gmap->children)) {
236 		spin_lock(&gmap->shadow_lock);
237 		list_for_each_entry_safe(sg, next, &gmap->children, list) {
238 			list_del(&sg->list);
239 			gmap_put(sg);
240 		}
241 		spin_unlock(&gmap->shadow_lock);
242 	}
243 	/* Remove gmap from the pre-mm list */
244 	spin_lock(&gmap->mm->context.gmap_lock);
245 	list_del_rcu(&gmap->list);
246 	spin_unlock(&gmap->mm->context.gmap_lock);
247 	synchronize_rcu();
248 	/* Put reference */
249 	gmap_put(gmap);
250 }
251 EXPORT_SYMBOL_GPL(gmap_remove);
252 
253 /**
254  * gmap_enable - switch primary space to the guest address space
255  * @gmap: pointer to the guest address space structure
256  */
257 void gmap_enable(struct gmap *gmap)
258 {
259 	S390_lowcore.gmap = (unsigned long) gmap;
260 }
261 EXPORT_SYMBOL_GPL(gmap_enable);
262 
263 /**
264  * gmap_disable - switch back to the standard primary address space
265  * @gmap: pointer to the guest address space structure
266  */
267 void gmap_disable(struct gmap *gmap)
268 {
269 	S390_lowcore.gmap = 0UL;
270 }
271 EXPORT_SYMBOL_GPL(gmap_disable);
272 
273 /**
274  * gmap_get_enabled - get a pointer to the currently enabled gmap
275  *
276  * Returns a pointer to the currently enabled gmap. 0 if none is enabled.
277  */
278 struct gmap *gmap_get_enabled(void)
279 {
280 	return (struct gmap *) S390_lowcore.gmap;
281 }
282 EXPORT_SYMBOL_GPL(gmap_get_enabled);
283 
284 /*
285  * gmap_alloc_table is assumed to be called with mmap_sem held
286  */
287 static int gmap_alloc_table(struct gmap *gmap, unsigned long *table,
288 			    unsigned long init, unsigned long gaddr)
289 {
290 	struct page *page;
291 	unsigned long *new;
292 
293 	/* since we dont free the gmap table until gmap_free we can unlock */
294 	page = alloc_pages(GFP_KERNEL, 2);
295 	if (!page)
296 		return -ENOMEM;
297 	new = (unsigned long *) page_to_phys(page);
298 	crst_table_init(new, init);
299 	spin_lock(&gmap->guest_table_lock);
300 	if (*table & _REGION_ENTRY_INVALID) {
301 		list_add(&page->lru, &gmap->crst_list);
302 		*table = (unsigned long) new | _REGION_ENTRY_LENGTH |
303 			(*table & _REGION_ENTRY_TYPE_MASK);
304 		page->index = gaddr;
305 		page = NULL;
306 	}
307 	spin_unlock(&gmap->guest_table_lock);
308 	if (page)
309 		__free_pages(page, 2);
310 	return 0;
311 }
312 
313 /**
314  * __gmap_segment_gaddr - find virtual address from segment pointer
315  * @entry: pointer to a segment table entry in the guest address space
316  *
317  * Returns the virtual address in the guest address space for the segment
318  */
319 static unsigned long __gmap_segment_gaddr(unsigned long *entry)
320 {
321 	struct page *page;
322 	unsigned long offset, mask;
323 
324 	offset = (unsigned long) entry / sizeof(unsigned long);
325 	offset = (offset & (PTRS_PER_PMD - 1)) * PMD_SIZE;
326 	mask = ~(PTRS_PER_PMD * sizeof(pmd_t) - 1);
327 	page = virt_to_page((void *)((unsigned long) entry & mask));
328 	return page->index + offset;
329 }
330 
331 /**
332  * __gmap_unlink_by_vmaddr - unlink a single segment via a host address
333  * @gmap: pointer to the guest address space structure
334  * @vmaddr: address in the host process address space
335  *
336  * Returns 1 if a TLB flush is required
337  */
338 static int __gmap_unlink_by_vmaddr(struct gmap *gmap, unsigned long vmaddr)
339 {
340 	unsigned long *entry;
341 	int flush = 0;
342 
343 	BUG_ON(gmap_is_shadow(gmap));
344 	spin_lock(&gmap->guest_table_lock);
345 	entry = radix_tree_delete(&gmap->host_to_guest, vmaddr >> PMD_SHIFT);
346 	if (entry) {
347 		flush = (*entry != _SEGMENT_ENTRY_INVALID);
348 		*entry = _SEGMENT_ENTRY_INVALID;
349 	}
350 	spin_unlock(&gmap->guest_table_lock);
351 	return flush;
352 }
353 
354 /**
355  * __gmap_unmap_by_gaddr - unmap a single segment via a guest address
356  * @gmap: pointer to the guest address space structure
357  * @gaddr: address in the guest address space
358  *
359  * Returns 1 if a TLB flush is required
360  */
361 static int __gmap_unmap_by_gaddr(struct gmap *gmap, unsigned long gaddr)
362 {
363 	unsigned long vmaddr;
364 
365 	vmaddr = (unsigned long) radix_tree_delete(&gmap->guest_to_host,
366 						   gaddr >> PMD_SHIFT);
367 	return vmaddr ? __gmap_unlink_by_vmaddr(gmap, vmaddr) : 0;
368 }
369 
370 /**
371  * gmap_unmap_segment - unmap segment from the guest address space
372  * @gmap: pointer to the guest address space structure
373  * @to: address in the guest address space
374  * @len: length of the memory area to unmap
375  *
376  * Returns 0 if the unmap succeeded, -EINVAL if not.
377  */
378 int gmap_unmap_segment(struct gmap *gmap, unsigned long to, unsigned long len)
379 {
380 	unsigned long off;
381 	int flush;
382 
383 	BUG_ON(gmap_is_shadow(gmap));
384 	if ((to | len) & (PMD_SIZE - 1))
385 		return -EINVAL;
386 	if (len == 0 || to + len < to)
387 		return -EINVAL;
388 
389 	flush = 0;
390 	down_write(&gmap->mm->mmap_sem);
391 	for (off = 0; off < len; off += PMD_SIZE)
392 		flush |= __gmap_unmap_by_gaddr(gmap, to + off);
393 	up_write(&gmap->mm->mmap_sem);
394 	if (flush)
395 		gmap_flush_tlb(gmap);
396 	return 0;
397 }
398 EXPORT_SYMBOL_GPL(gmap_unmap_segment);
399 
400 /**
401  * gmap_map_segment - map a segment to the guest address space
402  * @gmap: pointer to the guest address space structure
403  * @from: source address in the parent address space
404  * @to: target address in the guest address space
405  * @len: length of the memory area to map
406  *
407  * Returns 0 if the mmap succeeded, -EINVAL or -ENOMEM if not.
408  */
409 int gmap_map_segment(struct gmap *gmap, unsigned long from,
410 		     unsigned long to, unsigned long len)
411 {
412 	unsigned long off;
413 	int flush;
414 
415 	BUG_ON(gmap_is_shadow(gmap));
416 	if ((from | to | len) & (PMD_SIZE - 1))
417 		return -EINVAL;
418 	if (len == 0 || from + len < from || to + len < to ||
419 	    from + len - 1 > TASK_MAX_SIZE || to + len - 1 > gmap->asce_end)
420 		return -EINVAL;
421 
422 	flush = 0;
423 	down_write(&gmap->mm->mmap_sem);
424 	for (off = 0; off < len; off += PMD_SIZE) {
425 		/* Remove old translation */
426 		flush |= __gmap_unmap_by_gaddr(gmap, to + off);
427 		/* Store new translation */
428 		if (radix_tree_insert(&gmap->guest_to_host,
429 				      (to + off) >> PMD_SHIFT,
430 				      (void *) from + off))
431 			break;
432 	}
433 	up_write(&gmap->mm->mmap_sem);
434 	if (flush)
435 		gmap_flush_tlb(gmap);
436 	if (off >= len)
437 		return 0;
438 	gmap_unmap_segment(gmap, to, len);
439 	return -ENOMEM;
440 }
441 EXPORT_SYMBOL_GPL(gmap_map_segment);
442 
443 /**
444  * __gmap_translate - translate a guest address to a user space address
445  * @gmap: pointer to guest mapping meta data structure
446  * @gaddr: guest address
447  *
448  * Returns user space address which corresponds to the guest address or
449  * -EFAULT if no such mapping exists.
450  * This function does not establish potentially missing page table entries.
451  * The mmap_sem of the mm that belongs to the address space must be held
452  * when this function gets called.
453  *
454  * Note: Can also be called for shadow gmaps.
455  */
456 unsigned long __gmap_translate(struct gmap *gmap, unsigned long gaddr)
457 {
458 	unsigned long vmaddr;
459 
460 	vmaddr = (unsigned long)
461 		radix_tree_lookup(&gmap->guest_to_host, gaddr >> PMD_SHIFT);
462 	/* Note: guest_to_host is empty for a shadow gmap */
463 	return vmaddr ? (vmaddr | (gaddr & ~PMD_MASK)) : -EFAULT;
464 }
465 EXPORT_SYMBOL_GPL(__gmap_translate);
466 
467 /**
468  * gmap_translate - translate a guest address to a user space address
469  * @gmap: pointer to guest mapping meta data structure
470  * @gaddr: guest address
471  *
472  * Returns user space address which corresponds to the guest address or
473  * -EFAULT if no such mapping exists.
474  * This function does not establish potentially missing page table entries.
475  */
476 unsigned long gmap_translate(struct gmap *gmap, unsigned long gaddr)
477 {
478 	unsigned long rc;
479 
480 	down_read(&gmap->mm->mmap_sem);
481 	rc = __gmap_translate(gmap, gaddr);
482 	up_read(&gmap->mm->mmap_sem);
483 	return rc;
484 }
485 EXPORT_SYMBOL_GPL(gmap_translate);
486 
487 /**
488  * gmap_unlink - disconnect a page table from the gmap shadow tables
489  * @gmap: pointer to guest mapping meta data structure
490  * @table: pointer to the host page table
491  * @vmaddr: vm address associated with the host page table
492  */
493 void gmap_unlink(struct mm_struct *mm, unsigned long *table,
494 		 unsigned long vmaddr)
495 {
496 	struct gmap *gmap;
497 	int flush;
498 
499 	rcu_read_lock();
500 	list_for_each_entry_rcu(gmap, &mm->context.gmap_list, list) {
501 		flush = __gmap_unlink_by_vmaddr(gmap, vmaddr);
502 		if (flush)
503 			gmap_flush_tlb(gmap);
504 	}
505 	rcu_read_unlock();
506 }
507 
508 /**
509  * gmap_link - set up shadow page tables to connect a host to a guest address
510  * @gmap: pointer to guest mapping meta data structure
511  * @gaddr: guest address
512  * @vmaddr: vm address
513  *
514  * Returns 0 on success, -ENOMEM for out of memory conditions, and -EFAULT
515  * if the vm address is already mapped to a different guest segment.
516  * The mmap_sem of the mm that belongs to the address space must be held
517  * when this function gets called.
518  */
519 int __gmap_link(struct gmap *gmap, unsigned long gaddr, unsigned long vmaddr)
520 {
521 	struct mm_struct *mm;
522 	unsigned long *table;
523 	spinlock_t *ptl;
524 	pgd_t *pgd;
525 	pud_t *pud;
526 	pmd_t *pmd;
527 	int rc;
528 
529 	BUG_ON(gmap_is_shadow(gmap));
530 	/* Create higher level tables in the gmap page table */
531 	table = gmap->table;
532 	if ((gmap->asce & _ASCE_TYPE_MASK) >= _ASCE_TYPE_REGION1) {
533 		table += (gaddr >> 53) & 0x7ff;
534 		if ((*table & _REGION_ENTRY_INVALID) &&
535 		    gmap_alloc_table(gmap, table, _REGION2_ENTRY_EMPTY,
536 				     gaddr & 0xffe0000000000000UL))
537 			return -ENOMEM;
538 		table = (unsigned long *)(*table & _REGION_ENTRY_ORIGIN);
539 	}
540 	if ((gmap->asce & _ASCE_TYPE_MASK) >= _ASCE_TYPE_REGION2) {
541 		table += (gaddr >> 42) & 0x7ff;
542 		if ((*table & _REGION_ENTRY_INVALID) &&
543 		    gmap_alloc_table(gmap, table, _REGION3_ENTRY_EMPTY,
544 				     gaddr & 0xfffffc0000000000UL))
545 			return -ENOMEM;
546 		table = (unsigned long *)(*table & _REGION_ENTRY_ORIGIN);
547 	}
548 	if ((gmap->asce & _ASCE_TYPE_MASK) >= _ASCE_TYPE_REGION3) {
549 		table += (gaddr >> 31) & 0x7ff;
550 		if ((*table & _REGION_ENTRY_INVALID) &&
551 		    gmap_alloc_table(gmap, table, _SEGMENT_ENTRY_EMPTY,
552 				     gaddr & 0xffffffff80000000UL))
553 			return -ENOMEM;
554 		table = (unsigned long *)(*table & _REGION_ENTRY_ORIGIN);
555 	}
556 	table += (gaddr >> 20) & 0x7ff;
557 	/* Walk the parent mm page table */
558 	mm = gmap->mm;
559 	pgd = pgd_offset(mm, vmaddr);
560 	VM_BUG_ON(pgd_none(*pgd));
561 	pud = pud_offset(pgd, vmaddr);
562 	VM_BUG_ON(pud_none(*pud));
563 	/* large puds cannot yet be handled */
564 	if (pud_large(*pud))
565 		return -EFAULT;
566 	pmd = pmd_offset(pud, vmaddr);
567 	VM_BUG_ON(pmd_none(*pmd));
568 	/* large pmds cannot yet be handled */
569 	if (pmd_large(*pmd))
570 		return -EFAULT;
571 	/* Link gmap segment table entry location to page table. */
572 	rc = radix_tree_preload(GFP_KERNEL);
573 	if (rc)
574 		return rc;
575 	ptl = pmd_lock(mm, pmd);
576 	spin_lock(&gmap->guest_table_lock);
577 	if (*table == _SEGMENT_ENTRY_INVALID) {
578 		rc = radix_tree_insert(&gmap->host_to_guest,
579 				       vmaddr >> PMD_SHIFT, table);
580 		if (!rc)
581 			*table = pmd_val(*pmd);
582 	} else
583 		rc = 0;
584 	spin_unlock(&gmap->guest_table_lock);
585 	spin_unlock(ptl);
586 	radix_tree_preload_end();
587 	return rc;
588 }
589 
590 /**
591  * gmap_fault - resolve a fault on a guest address
592  * @gmap: pointer to guest mapping meta data structure
593  * @gaddr: guest address
594  * @fault_flags: flags to pass down to handle_mm_fault()
595  *
596  * Returns 0 on success, -ENOMEM for out of memory conditions, and -EFAULT
597  * if the vm address is already mapped to a different guest segment.
598  */
599 int gmap_fault(struct gmap *gmap, unsigned long gaddr,
600 	       unsigned int fault_flags)
601 {
602 	unsigned long vmaddr;
603 	int rc;
604 	bool unlocked;
605 
606 	down_read(&gmap->mm->mmap_sem);
607 
608 retry:
609 	unlocked = false;
610 	vmaddr = __gmap_translate(gmap, gaddr);
611 	if (IS_ERR_VALUE(vmaddr)) {
612 		rc = vmaddr;
613 		goto out_up;
614 	}
615 	if (fixup_user_fault(current, gmap->mm, vmaddr, fault_flags,
616 			     &unlocked)) {
617 		rc = -EFAULT;
618 		goto out_up;
619 	}
620 	/*
621 	 * In the case that fixup_user_fault unlocked the mmap_sem during
622 	 * faultin redo __gmap_translate to not race with a map/unmap_segment.
623 	 */
624 	if (unlocked)
625 		goto retry;
626 
627 	rc = __gmap_link(gmap, gaddr, vmaddr);
628 out_up:
629 	up_read(&gmap->mm->mmap_sem);
630 	return rc;
631 }
632 EXPORT_SYMBOL_GPL(gmap_fault);
633 
634 /*
635  * this function is assumed to be called with mmap_sem held
636  */
637 void __gmap_zap(struct gmap *gmap, unsigned long gaddr)
638 {
639 	unsigned long vmaddr;
640 	spinlock_t *ptl;
641 	pte_t *ptep;
642 
643 	/* Find the vm address for the guest address */
644 	vmaddr = (unsigned long) radix_tree_lookup(&gmap->guest_to_host,
645 						   gaddr >> PMD_SHIFT);
646 	if (vmaddr) {
647 		vmaddr |= gaddr & ~PMD_MASK;
648 		/* Get pointer to the page table entry */
649 		ptep = get_locked_pte(gmap->mm, vmaddr, &ptl);
650 		if (likely(ptep))
651 			ptep_zap_unused(gmap->mm, vmaddr, ptep, 0);
652 		pte_unmap_unlock(ptep, ptl);
653 	}
654 }
655 EXPORT_SYMBOL_GPL(__gmap_zap);
656 
657 void gmap_discard(struct gmap *gmap, unsigned long from, unsigned long to)
658 {
659 	unsigned long gaddr, vmaddr, size;
660 	struct vm_area_struct *vma;
661 
662 	down_read(&gmap->mm->mmap_sem);
663 	for (gaddr = from; gaddr < to;
664 	     gaddr = (gaddr + PMD_SIZE) & PMD_MASK) {
665 		/* Find the vm address for the guest address */
666 		vmaddr = (unsigned long)
667 			radix_tree_lookup(&gmap->guest_to_host,
668 					  gaddr >> PMD_SHIFT);
669 		if (!vmaddr)
670 			continue;
671 		vmaddr |= gaddr & ~PMD_MASK;
672 		/* Find vma in the parent mm */
673 		vma = find_vma(gmap->mm, vmaddr);
674 		size = min(to - gaddr, PMD_SIZE - (gaddr & ~PMD_MASK));
675 		zap_page_range(vma, vmaddr, size, NULL);
676 	}
677 	up_read(&gmap->mm->mmap_sem);
678 }
679 EXPORT_SYMBOL_GPL(gmap_discard);
680 
681 static LIST_HEAD(gmap_notifier_list);
682 static DEFINE_SPINLOCK(gmap_notifier_lock);
683 
684 /**
685  * gmap_register_pte_notifier - register a pte invalidation callback
686  * @nb: pointer to the gmap notifier block
687  */
688 void gmap_register_pte_notifier(struct gmap_notifier *nb)
689 {
690 	spin_lock(&gmap_notifier_lock);
691 	list_add_rcu(&nb->list, &gmap_notifier_list);
692 	spin_unlock(&gmap_notifier_lock);
693 }
694 EXPORT_SYMBOL_GPL(gmap_register_pte_notifier);
695 
696 /**
697  * gmap_unregister_pte_notifier - remove a pte invalidation callback
698  * @nb: pointer to the gmap notifier block
699  */
700 void gmap_unregister_pte_notifier(struct gmap_notifier *nb)
701 {
702 	spin_lock(&gmap_notifier_lock);
703 	list_del_rcu(&nb->list);
704 	spin_unlock(&gmap_notifier_lock);
705 	synchronize_rcu();
706 }
707 EXPORT_SYMBOL_GPL(gmap_unregister_pte_notifier);
708 
709 /**
710  * gmap_call_notifier - call all registered invalidation callbacks
711  * @gmap: pointer to guest mapping meta data structure
712  * @start: start virtual address in the guest address space
713  * @end: end virtual address in the guest address space
714  */
715 static void gmap_call_notifier(struct gmap *gmap, unsigned long start,
716 			       unsigned long end)
717 {
718 	struct gmap_notifier *nb;
719 
720 	list_for_each_entry(nb, &gmap_notifier_list, list)
721 		nb->notifier_call(gmap, start, end);
722 }
723 
724 /**
725  * gmap_table_walk - walk the gmap page tables
726  * @gmap: pointer to guest mapping meta data structure
727  * @gaddr: virtual address in the guest address space
728  * @level: page table level to stop at
729  *
730  * Returns a table entry pointer for the given guest address and @level
731  * @level=0 : returns a pointer to a page table table entry (or NULL)
732  * @level=1 : returns a pointer to a segment table entry (or NULL)
733  * @level=2 : returns a pointer to a region-3 table entry (or NULL)
734  * @level=3 : returns a pointer to a region-2 table entry (or NULL)
735  * @level=4 : returns a pointer to a region-1 table entry (or NULL)
736  *
737  * Returns NULL if the gmap page tables could not be walked to the
738  * requested level.
739  *
740  * Note: Can also be called for shadow gmaps.
741  */
742 static inline unsigned long *gmap_table_walk(struct gmap *gmap,
743 					     unsigned long gaddr, int level)
744 {
745 	unsigned long *table;
746 
747 	if ((gmap->asce & _ASCE_TYPE_MASK) + 4 < (level * 4))
748 		return NULL;
749 	if (gmap_is_shadow(gmap) && gmap->removed)
750 		return NULL;
751 	if (gaddr & (-1UL << (31 + ((gmap->asce & _ASCE_TYPE_MASK) >> 2)*11)))
752 		return NULL;
753 	table = gmap->table;
754 	switch (gmap->asce & _ASCE_TYPE_MASK) {
755 	case _ASCE_TYPE_REGION1:
756 		table += (gaddr >> 53) & 0x7ff;
757 		if (level == 4)
758 			break;
759 		if (*table & _REGION_ENTRY_INVALID)
760 			return NULL;
761 		table = (unsigned long *)(*table & _REGION_ENTRY_ORIGIN);
762 		/* Fallthrough */
763 	case _ASCE_TYPE_REGION2:
764 		table += (gaddr >> 42) & 0x7ff;
765 		if (level == 3)
766 			break;
767 		if (*table & _REGION_ENTRY_INVALID)
768 			return NULL;
769 		table = (unsigned long *)(*table & _REGION_ENTRY_ORIGIN);
770 		/* Fallthrough */
771 	case _ASCE_TYPE_REGION3:
772 		table += (gaddr >> 31) & 0x7ff;
773 		if (level == 2)
774 			break;
775 		if (*table & _REGION_ENTRY_INVALID)
776 			return NULL;
777 		table = (unsigned long *)(*table & _REGION_ENTRY_ORIGIN);
778 		/* Fallthrough */
779 	case _ASCE_TYPE_SEGMENT:
780 		table += (gaddr >> 20) & 0x7ff;
781 		if (level == 1)
782 			break;
783 		if (*table & _REGION_ENTRY_INVALID)
784 			return NULL;
785 		table = (unsigned long *)(*table & _SEGMENT_ENTRY_ORIGIN);
786 		table += (gaddr >> 12) & 0xff;
787 	}
788 	return table;
789 }
790 
791 /**
792  * gmap_pte_op_walk - walk the gmap page table, get the page table lock
793  *		      and return the pte pointer
794  * @gmap: pointer to guest mapping meta data structure
795  * @gaddr: virtual address in the guest address space
796  * @ptl: pointer to the spinlock pointer
797  *
798  * Returns a pointer to the locked pte for a guest address, or NULL
799  *
800  * Note: Can also be called for shadow gmaps.
801  */
802 static pte_t *gmap_pte_op_walk(struct gmap *gmap, unsigned long gaddr,
803 			       spinlock_t **ptl)
804 {
805 	unsigned long *table;
806 
807 	if (gmap_is_shadow(gmap))
808 		spin_lock(&gmap->guest_table_lock);
809 	/* Walk the gmap page table, lock and get pte pointer */
810 	table = gmap_table_walk(gmap, gaddr, 1); /* get segment pointer */
811 	if (!table || *table & _SEGMENT_ENTRY_INVALID) {
812 		if (gmap_is_shadow(gmap))
813 			spin_unlock(&gmap->guest_table_lock);
814 		return NULL;
815 	}
816 	if (gmap_is_shadow(gmap)) {
817 		*ptl = &gmap->guest_table_lock;
818 		return pte_offset_map((pmd_t *) table, gaddr);
819 	}
820 	return pte_alloc_map_lock(gmap->mm, (pmd_t *) table, gaddr, ptl);
821 }
822 
823 /**
824  * gmap_pte_op_fixup - force a page in and connect the gmap page table
825  * @gmap: pointer to guest mapping meta data structure
826  * @gaddr: virtual address in the guest address space
827  * @vmaddr: address in the host process address space
828  * @prot: indicates access rights: PROT_NONE, PROT_READ or PROT_WRITE
829  *
830  * Returns 0 if the caller can retry __gmap_translate (might fail again),
831  * -ENOMEM if out of memory and -EFAULT if anything goes wrong while fixing
832  * up or connecting the gmap page table.
833  */
834 static int gmap_pte_op_fixup(struct gmap *gmap, unsigned long gaddr,
835 			     unsigned long vmaddr, int prot)
836 {
837 	struct mm_struct *mm = gmap->mm;
838 	unsigned int fault_flags;
839 	bool unlocked = false;
840 
841 	BUG_ON(gmap_is_shadow(gmap));
842 	fault_flags = (prot == PROT_WRITE) ? FAULT_FLAG_WRITE : 0;
843 	if (fixup_user_fault(current, mm, vmaddr, fault_flags, &unlocked))
844 		return -EFAULT;
845 	if (unlocked)
846 		/* lost mmap_sem, caller has to retry __gmap_translate */
847 		return 0;
848 	/* Connect the page tables */
849 	return __gmap_link(gmap, gaddr, vmaddr);
850 }
851 
852 /**
853  * gmap_pte_op_end - release the page table lock
854  * @ptl: pointer to the spinlock pointer
855  */
856 static void gmap_pte_op_end(spinlock_t *ptl)
857 {
858 	spin_unlock(ptl);
859 }
860 
861 /*
862  * gmap_protect_range - remove access rights to memory and set pgste bits
863  * @gmap: pointer to guest mapping meta data structure
864  * @gaddr: virtual address in the guest address space
865  * @len: size of area
866  * @prot: indicates access rights: PROT_NONE, PROT_READ or PROT_WRITE
867  * @bits: pgste notification bits to set
868  *
869  * Returns 0 if successfully protected, -ENOMEM if out of memory and
870  * -EFAULT if gaddr is invalid (or mapping for shadows is missing).
871  *
872  * Called with sg->mm->mmap_sem in read.
873  *
874  * Note: Can also be called for shadow gmaps.
875  */
876 static int gmap_protect_range(struct gmap *gmap, unsigned long gaddr,
877 			      unsigned long len, int prot, unsigned long bits)
878 {
879 	unsigned long vmaddr;
880 	spinlock_t *ptl;
881 	pte_t *ptep;
882 	int rc;
883 
884 	while (len) {
885 		rc = -EAGAIN;
886 		ptep = gmap_pte_op_walk(gmap, gaddr, &ptl);
887 		if (ptep) {
888 			rc = ptep_force_prot(gmap->mm, gaddr, ptep, prot, bits);
889 			gmap_pte_op_end(ptl);
890 		}
891 		if (rc) {
892 			vmaddr = __gmap_translate(gmap, gaddr);
893 			if (IS_ERR_VALUE(vmaddr))
894 				return vmaddr;
895 			rc = gmap_pte_op_fixup(gmap, gaddr, vmaddr, prot);
896 			if (rc)
897 				return rc;
898 			continue;
899 		}
900 		gaddr += PAGE_SIZE;
901 		len -= PAGE_SIZE;
902 	}
903 	return 0;
904 }
905 
906 /**
907  * gmap_mprotect_notify - change access rights for a range of ptes and
908  *                        call the notifier if any pte changes again
909  * @gmap: pointer to guest mapping meta data structure
910  * @gaddr: virtual address in the guest address space
911  * @len: size of area
912  * @prot: indicates access rights: PROT_NONE, PROT_READ or PROT_WRITE
913  *
914  * Returns 0 if for each page in the given range a gmap mapping exists,
915  * the new access rights could be set and the notifier could be armed.
916  * If the gmap mapping is missing for one or more pages -EFAULT is
917  * returned. If no memory could be allocated -ENOMEM is returned.
918  * This function establishes missing page table entries.
919  */
920 int gmap_mprotect_notify(struct gmap *gmap, unsigned long gaddr,
921 			 unsigned long len, int prot)
922 {
923 	int rc;
924 
925 	if ((gaddr & ~PAGE_MASK) || (len & ~PAGE_MASK) || gmap_is_shadow(gmap))
926 		return -EINVAL;
927 	if (!MACHINE_HAS_ESOP && prot == PROT_READ)
928 		return -EINVAL;
929 	down_read(&gmap->mm->mmap_sem);
930 	rc = gmap_protect_range(gmap, gaddr, len, prot, PGSTE_IN_BIT);
931 	up_read(&gmap->mm->mmap_sem);
932 	return rc;
933 }
934 EXPORT_SYMBOL_GPL(gmap_mprotect_notify);
935 
936 /**
937  * gmap_read_table - get an unsigned long value from a guest page table using
938  *                   absolute addressing, without marking the page referenced.
939  * @gmap: pointer to guest mapping meta data structure
940  * @gaddr: virtual address in the guest address space
941  * @val: pointer to the unsigned long value to return
942  *
943  * Returns 0 if the value was read, -ENOMEM if out of memory and -EFAULT
944  * if reading using the virtual address failed.
945  *
946  * Called with gmap->mm->mmap_sem in read.
947  */
948 int gmap_read_table(struct gmap *gmap, unsigned long gaddr, unsigned long *val)
949 {
950 	unsigned long address, vmaddr;
951 	spinlock_t *ptl;
952 	pte_t *ptep, pte;
953 	int rc;
954 
955 	while (1) {
956 		rc = -EAGAIN;
957 		ptep = gmap_pte_op_walk(gmap, gaddr, &ptl);
958 		if (ptep) {
959 			pte = *ptep;
960 			if (pte_present(pte) && (pte_val(pte) & _PAGE_READ)) {
961 				address = pte_val(pte) & PAGE_MASK;
962 				address += gaddr & ~PAGE_MASK;
963 				*val = *(unsigned long *) address;
964 				pte_val(*ptep) |= _PAGE_YOUNG;
965 				/* Do *NOT* clear the _PAGE_INVALID bit! */
966 				rc = 0;
967 			}
968 			gmap_pte_op_end(ptl);
969 		}
970 		if (!rc)
971 			break;
972 		vmaddr = __gmap_translate(gmap, gaddr);
973 		if (IS_ERR_VALUE(vmaddr)) {
974 			rc = vmaddr;
975 			break;
976 		}
977 		rc = gmap_pte_op_fixup(gmap, gaddr, vmaddr, PROT_READ);
978 		if (rc)
979 			break;
980 	}
981 	return rc;
982 }
983 EXPORT_SYMBOL_GPL(gmap_read_table);
984 
985 /**
986  * gmap_insert_rmap - add a rmap to the host_to_rmap radix tree
987  * @sg: pointer to the shadow guest address space structure
988  * @vmaddr: vm address associated with the rmap
989  * @rmap: pointer to the rmap structure
990  *
991  * Called with the sg->guest_table_lock
992  */
993 static inline void gmap_insert_rmap(struct gmap *sg, unsigned long vmaddr,
994 				    struct gmap_rmap *rmap)
995 {
996 	void **slot;
997 
998 	BUG_ON(!gmap_is_shadow(sg));
999 	slot = radix_tree_lookup_slot(&sg->host_to_rmap, vmaddr >> PAGE_SHIFT);
1000 	if (slot) {
1001 		rmap->next = radix_tree_deref_slot_protected(slot,
1002 							&sg->guest_table_lock);
1003 		radix_tree_replace_slot(slot, rmap);
1004 	} else {
1005 		rmap->next = NULL;
1006 		radix_tree_insert(&sg->host_to_rmap, vmaddr >> PAGE_SHIFT,
1007 				  rmap);
1008 	}
1009 }
1010 
1011 /**
1012  * gmap_protect_rmap - modify access rights to memory and create an rmap
1013  * @sg: pointer to the shadow guest address space structure
1014  * @raddr: rmap address in the shadow gmap
1015  * @paddr: address in the parent guest address space
1016  * @len: length of the memory area to protect
1017  * @prot: indicates access rights: none, read-only or read-write
1018  *
1019  * Returns 0 if successfully protected and the rmap was created, -ENOMEM
1020  * if out of memory and -EFAULT if paddr is invalid.
1021  */
1022 static int gmap_protect_rmap(struct gmap *sg, unsigned long raddr,
1023 			     unsigned long paddr, unsigned long len, int prot)
1024 {
1025 	struct gmap *parent;
1026 	struct gmap_rmap *rmap;
1027 	unsigned long vmaddr;
1028 	spinlock_t *ptl;
1029 	pte_t *ptep;
1030 	int rc;
1031 
1032 	BUG_ON(!gmap_is_shadow(sg));
1033 	parent = sg->parent;
1034 	while (len) {
1035 		vmaddr = __gmap_translate(parent, paddr);
1036 		if (IS_ERR_VALUE(vmaddr))
1037 			return vmaddr;
1038 		rmap = kzalloc(sizeof(*rmap), GFP_KERNEL);
1039 		if (!rmap)
1040 			return -ENOMEM;
1041 		rmap->raddr = raddr;
1042 		rc = radix_tree_preload(GFP_KERNEL);
1043 		if (rc) {
1044 			kfree(rmap);
1045 			return rc;
1046 		}
1047 		rc = -EAGAIN;
1048 		ptep = gmap_pte_op_walk(parent, paddr, &ptl);
1049 		if (ptep) {
1050 			spin_lock(&sg->guest_table_lock);
1051 			rc = ptep_force_prot(parent->mm, paddr, ptep, prot,
1052 					     PGSTE_VSIE_BIT);
1053 			if (!rc)
1054 				gmap_insert_rmap(sg, vmaddr, rmap);
1055 			spin_unlock(&sg->guest_table_lock);
1056 			gmap_pte_op_end(ptl);
1057 		}
1058 		radix_tree_preload_end();
1059 		if (rc) {
1060 			kfree(rmap);
1061 			rc = gmap_pte_op_fixup(parent, paddr, vmaddr, prot);
1062 			if (rc)
1063 				return rc;
1064 			continue;
1065 		}
1066 		paddr += PAGE_SIZE;
1067 		len -= PAGE_SIZE;
1068 	}
1069 	return 0;
1070 }
1071 
1072 #define _SHADOW_RMAP_MASK	0x7
1073 #define _SHADOW_RMAP_REGION1	0x5
1074 #define _SHADOW_RMAP_REGION2	0x4
1075 #define _SHADOW_RMAP_REGION3	0x3
1076 #define _SHADOW_RMAP_SEGMENT	0x2
1077 #define _SHADOW_RMAP_PGTABLE	0x1
1078 
1079 /**
1080  * gmap_idte_one - invalidate a single region or segment table entry
1081  * @asce: region or segment table *origin* + table-type bits
1082  * @vaddr: virtual address to identify the table entry to flush
1083  *
1084  * The invalid bit of a single region or segment table entry is set
1085  * and the associated TLB entries depending on the entry are flushed.
1086  * The table-type of the @asce identifies the portion of the @vaddr
1087  * that is used as the invalidation index.
1088  */
1089 static inline void gmap_idte_one(unsigned long asce, unsigned long vaddr)
1090 {
1091 	asm volatile(
1092 		"	.insn	rrf,0xb98e0000,%0,%1,0,0"
1093 		: : "a" (asce), "a" (vaddr) : "cc", "memory");
1094 }
1095 
1096 /**
1097  * gmap_unshadow_page - remove a page from a shadow page table
1098  * @sg: pointer to the shadow guest address space structure
1099  * @raddr: rmap address in the shadow guest address space
1100  *
1101  * Called with the sg->guest_table_lock
1102  */
1103 static void gmap_unshadow_page(struct gmap *sg, unsigned long raddr)
1104 {
1105 	unsigned long *table;
1106 
1107 	BUG_ON(!gmap_is_shadow(sg));
1108 	table = gmap_table_walk(sg, raddr, 0); /* get page table pointer */
1109 	if (!table || *table & _PAGE_INVALID)
1110 		return;
1111 	gmap_call_notifier(sg, raddr, raddr + (1UL << 12) - 1);
1112 	ptep_unshadow_pte(sg->mm, raddr, (pte_t *) table);
1113 }
1114 
1115 /**
1116  * __gmap_unshadow_pgt - remove all entries from a shadow page table
1117  * @sg: pointer to the shadow guest address space structure
1118  * @raddr: rmap address in the shadow guest address space
1119  * @pgt: pointer to the start of a shadow page table
1120  *
1121  * Called with the sg->guest_table_lock
1122  */
1123 static void __gmap_unshadow_pgt(struct gmap *sg, unsigned long raddr,
1124 				unsigned long *pgt)
1125 {
1126 	int i;
1127 
1128 	BUG_ON(!gmap_is_shadow(sg));
1129 	for (i = 0; i < 256; i++, raddr += 1UL << 12)
1130 		pgt[i] = _PAGE_INVALID;
1131 }
1132 
1133 /**
1134  * gmap_unshadow_pgt - remove a shadow page table from a segment entry
1135  * @sg: pointer to the shadow guest address space structure
1136  * @raddr: address in the shadow guest address space
1137  *
1138  * Called with the sg->guest_table_lock
1139  */
1140 static void gmap_unshadow_pgt(struct gmap *sg, unsigned long raddr)
1141 {
1142 	unsigned long sto, *ste, *pgt;
1143 	struct page *page;
1144 
1145 	BUG_ON(!gmap_is_shadow(sg));
1146 	ste = gmap_table_walk(sg, raddr, 1); /* get segment pointer */
1147 	if (!ste || !(*ste & _SEGMENT_ENTRY_ORIGIN))
1148 		return;
1149 	gmap_call_notifier(sg, raddr, raddr + (1UL << 20) - 1);
1150 	sto = (unsigned long) (ste - ((raddr >> 20) & 0x7ff));
1151 	gmap_idte_one(sto | _ASCE_TYPE_SEGMENT, raddr);
1152 	pgt = (unsigned long *)(*ste & _SEGMENT_ENTRY_ORIGIN);
1153 	*ste = _SEGMENT_ENTRY_EMPTY;
1154 	__gmap_unshadow_pgt(sg, raddr, pgt);
1155 	/* Free page table */
1156 	page = pfn_to_page(__pa(pgt) >> PAGE_SHIFT);
1157 	list_del(&page->lru);
1158 	page_table_free_pgste(page);
1159 }
1160 
1161 /**
1162  * __gmap_unshadow_sgt - remove all entries from a shadow segment table
1163  * @sg: pointer to the shadow guest address space structure
1164  * @raddr: rmap address in the shadow guest address space
1165  * @sgt: pointer to the start of a shadow segment table
1166  *
1167  * Called with the sg->guest_table_lock
1168  */
1169 static void __gmap_unshadow_sgt(struct gmap *sg, unsigned long raddr,
1170 				unsigned long *sgt)
1171 {
1172 	unsigned long asce, *pgt;
1173 	struct page *page;
1174 	int i;
1175 
1176 	BUG_ON(!gmap_is_shadow(sg));
1177 	asce = (unsigned long) sgt | _ASCE_TYPE_SEGMENT;
1178 	for (i = 0; i < 2048; i++, raddr += 1UL << 20) {
1179 		if (!(sgt[i] & _SEGMENT_ENTRY_ORIGIN))
1180 			continue;
1181 		pgt = (unsigned long *)(sgt[i] & _REGION_ENTRY_ORIGIN);
1182 		sgt[i] = _SEGMENT_ENTRY_EMPTY;
1183 		__gmap_unshadow_pgt(sg, raddr, pgt);
1184 		/* Free page table */
1185 		page = pfn_to_page(__pa(pgt) >> PAGE_SHIFT);
1186 		list_del(&page->lru);
1187 		page_table_free_pgste(page);
1188 	}
1189 }
1190 
1191 /**
1192  * gmap_unshadow_sgt - remove a shadow segment table from a region-3 entry
1193  * @sg: pointer to the shadow guest address space structure
1194  * @raddr: rmap address in the shadow guest address space
1195  *
1196  * Called with the shadow->guest_table_lock
1197  */
1198 static void gmap_unshadow_sgt(struct gmap *sg, unsigned long raddr)
1199 {
1200 	unsigned long r3o, *r3e, *sgt;
1201 	struct page *page;
1202 
1203 	BUG_ON(!gmap_is_shadow(sg));
1204 	r3e = gmap_table_walk(sg, raddr, 2); /* get region-3 pointer */
1205 	if (!r3e || !(*r3e & _REGION_ENTRY_ORIGIN))
1206 		return;
1207 	gmap_call_notifier(sg, raddr, raddr + (1UL << 31) - 1);
1208 	r3o = (unsigned long) (r3e - ((raddr >> 31) & 0x7ff));
1209 	gmap_idte_one(r3o | _ASCE_TYPE_REGION3, raddr);
1210 	sgt = (unsigned long *)(*r3e & _REGION_ENTRY_ORIGIN);
1211 	*r3e = _REGION3_ENTRY_EMPTY;
1212 	__gmap_unshadow_sgt(sg, raddr, sgt);
1213 	/* Free segment table */
1214 	page = pfn_to_page(__pa(sgt) >> PAGE_SHIFT);
1215 	list_del(&page->lru);
1216 	__free_pages(page, 2);
1217 }
1218 
1219 /**
1220  * __gmap_unshadow_r3t - remove all entries from a shadow region-3 table
1221  * @sg: pointer to the shadow guest address space structure
1222  * @raddr: address in the shadow guest address space
1223  * @r3t: pointer to the start of a shadow region-3 table
1224  *
1225  * Called with the sg->guest_table_lock
1226  */
1227 static void __gmap_unshadow_r3t(struct gmap *sg, unsigned long raddr,
1228 				unsigned long *r3t)
1229 {
1230 	unsigned long asce, *sgt;
1231 	struct page *page;
1232 	int i;
1233 
1234 	BUG_ON(!gmap_is_shadow(sg));
1235 	asce = (unsigned long) r3t | _ASCE_TYPE_REGION3;
1236 	for (i = 0; i < 2048; i++, raddr += 1UL << 31) {
1237 		if (!(r3t[i] & _REGION_ENTRY_ORIGIN))
1238 			continue;
1239 		sgt = (unsigned long *)(r3t[i] & _REGION_ENTRY_ORIGIN);
1240 		r3t[i] = _REGION3_ENTRY_EMPTY;
1241 		__gmap_unshadow_sgt(sg, raddr, sgt);
1242 		/* Free segment table */
1243 		page = pfn_to_page(__pa(sgt) >> PAGE_SHIFT);
1244 		list_del(&page->lru);
1245 		__free_pages(page, 2);
1246 	}
1247 }
1248 
1249 /**
1250  * gmap_unshadow_r3t - remove a shadow region-3 table from a region-2 entry
1251  * @sg: pointer to the shadow guest address space structure
1252  * @raddr: rmap address in the shadow guest address space
1253  *
1254  * Called with the sg->guest_table_lock
1255  */
1256 static void gmap_unshadow_r3t(struct gmap *sg, unsigned long raddr)
1257 {
1258 	unsigned long r2o, *r2e, *r3t;
1259 	struct page *page;
1260 
1261 	BUG_ON(!gmap_is_shadow(sg));
1262 	r2e = gmap_table_walk(sg, raddr, 3); /* get region-2 pointer */
1263 	if (!r2e || !(*r2e & _REGION_ENTRY_ORIGIN))
1264 		return;
1265 	gmap_call_notifier(sg, raddr, raddr + (1UL << 42) - 1);
1266 	r2o = (unsigned long) (r2e - ((raddr >> 42) & 0x7ff));
1267 	gmap_idte_one(r2o | _ASCE_TYPE_REGION2, raddr);
1268 	r3t = (unsigned long *)(*r2e & _REGION_ENTRY_ORIGIN);
1269 	*r2e = _REGION2_ENTRY_EMPTY;
1270 	__gmap_unshadow_r3t(sg, raddr, r3t);
1271 	/* Free region 3 table */
1272 	page = pfn_to_page(__pa(r3t) >> PAGE_SHIFT);
1273 	list_del(&page->lru);
1274 	__free_pages(page, 2);
1275 }
1276 
1277 /**
1278  * __gmap_unshadow_r2t - remove all entries from a shadow region-2 table
1279  * @sg: pointer to the shadow guest address space structure
1280  * @raddr: rmap address in the shadow guest address space
1281  * @r2t: pointer to the start of a shadow region-2 table
1282  *
1283  * Called with the sg->guest_table_lock
1284  */
1285 static void __gmap_unshadow_r2t(struct gmap *sg, unsigned long raddr,
1286 				unsigned long *r2t)
1287 {
1288 	unsigned long asce, *r3t;
1289 	struct page *page;
1290 	int i;
1291 
1292 	BUG_ON(!gmap_is_shadow(sg));
1293 	asce = (unsigned long) r2t | _ASCE_TYPE_REGION2;
1294 	for (i = 0; i < 2048; i++, raddr += 1UL << 42) {
1295 		if (!(r2t[i] & _REGION_ENTRY_ORIGIN))
1296 			continue;
1297 		r3t = (unsigned long *)(r2t[i] & _REGION_ENTRY_ORIGIN);
1298 		r2t[i] = _REGION2_ENTRY_EMPTY;
1299 		__gmap_unshadow_r3t(sg, raddr, r3t);
1300 		/* Free region 3 table */
1301 		page = pfn_to_page(__pa(r3t) >> PAGE_SHIFT);
1302 		list_del(&page->lru);
1303 		__free_pages(page, 2);
1304 	}
1305 }
1306 
1307 /**
1308  * gmap_unshadow_r2t - remove a shadow region-2 table from a region-1 entry
1309  * @sg: pointer to the shadow guest address space structure
1310  * @raddr: rmap address in the shadow guest address space
1311  *
1312  * Called with the sg->guest_table_lock
1313  */
1314 static void gmap_unshadow_r2t(struct gmap *sg, unsigned long raddr)
1315 {
1316 	unsigned long r1o, *r1e, *r2t;
1317 	struct page *page;
1318 
1319 	BUG_ON(!gmap_is_shadow(sg));
1320 	r1e = gmap_table_walk(sg, raddr, 4); /* get region-1 pointer */
1321 	if (!r1e || !(*r1e & _REGION_ENTRY_ORIGIN))
1322 		return;
1323 	gmap_call_notifier(sg, raddr, raddr + (1UL << 53) - 1);
1324 	r1o = (unsigned long) (r1e - ((raddr >> 53) & 0x7ff));
1325 	gmap_idte_one(r1o | _ASCE_TYPE_REGION1, raddr);
1326 	r2t = (unsigned long *)(*r1e & _REGION_ENTRY_ORIGIN);
1327 	*r1e = _REGION1_ENTRY_EMPTY;
1328 	__gmap_unshadow_r2t(sg, raddr, r2t);
1329 	/* Free region 2 table */
1330 	page = pfn_to_page(__pa(r2t) >> PAGE_SHIFT);
1331 	list_del(&page->lru);
1332 	__free_pages(page, 2);
1333 }
1334 
1335 /**
1336  * __gmap_unshadow_r1t - remove all entries from a shadow region-1 table
1337  * @sg: pointer to the shadow guest address space structure
1338  * @raddr: rmap address in the shadow guest address space
1339  * @r1t: pointer to the start of a shadow region-1 table
1340  *
1341  * Called with the shadow->guest_table_lock
1342  */
1343 static void __gmap_unshadow_r1t(struct gmap *sg, unsigned long raddr,
1344 				unsigned long *r1t)
1345 {
1346 	unsigned long asce, *r2t;
1347 	struct page *page;
1348 	int i;
1349 
1350 	BUG_ON(!gmap_is_shadow(sg));
1351 	asce = (unsigned long) r1t | _ASCE_TYPE_REGION1;
1352 	for (i = 0; i < 2048; i++, raddr += 1UL << 53) {
1353 		if (!(r1t[i] & _REGION_ENTRY_ORIGIN))
1354 			continue;
1355 		r2t = (unsigned long *)(r1t[i] & _REGION_ENTRY_ORIGIN);
1356 		__gmap_unshadow_r2t(sg, raddr, r2t);
1357 		/* Clear entry and flush translation r1t -> r2t */
1358 		gmap_idte_one(asce, raddr);
1359 		r1t[i] = _REGION1_ENTRY_EMPTY;
1360 		/* Free region 2 table */
1361 		page = pfn_to_page(__pa(r2t) >> PAGE_SHIFT);
1362 		list_del(&page->lru);
1363 		__free_pages(page, 2);
1364 	}
1365 }
1366 
1367 /**
1368  * gmap_unshadow - remove a shadow page table completely
1369  * @sg: pointer to the shadow guest address space structure
1370  *
1371  * Called with sg->guest_table_lock
1372  */
1373 static void gmap_unshadow(struct gmap *sg)
1374 {
1375 	unsigned long *table;
1376 
1377 	BUG_ON(!gmap_is_shadow(sg));
1378 	if (sg->removed)
1379 		return;
1380 	sg->removed = 1;
1381 	gmap_call_notifier(sg, 0, -1UL);
1382 	gmap_flush_tlb(sg);
1383 	table = (unsigned long *)(sg->asce & _ASCE_ORIGIN);
1384 	switch (sg->asce & _ASCE_TYPE_MASK) {
1385 	case _ASCE_TYPE_REGION1:
1386 		__gmap_unshadow_r1t(sg, 0, table);
1387 		break;
1388 	case _ASCE_TYPE_REGION2:
1389 		__gmap_unshadow_r2t(sg, 0, table);
1390 		break;
1391 	case _ASCE_TYPE_REGION3:
1392 		__gmap_unshadow_r3t(sg, 0, table);
1393 		break;
1394 	case _ASCE_TYPE_SEGMENT:
1395 		__gmap_unshadow_sgt(sg, 0, table);
1396 		break;
1397 	}
1398 }
1399 
1400 /**
1401  * gmap_find_shadow - find a specific asce in the list of shadow tables
1402  * @parent: pointer to the parent gmap
1403  * @asce: ASCE for which the shadow table is created
1404  * @edat_level: edat level to be used for the shadow translation
1405  *
1406  * Returns the pointer to a gmap if a shadow table with the given asce is
1407  * already available, ERR_PTR(-EAGAIN) if another one is just being created,
1408  * otherwise NULL
1409  */
1410 static struct gmap *gmap_find_shadow(struct gmap *parent, unsigned long asce,
1411 				     int edat_level)
1412 {
1413 	struct gmap *sg;
1414 
1415 	list_for_each_entry(sg, &parent->children, list) {
1416 		if (sg->orig_asce != asce || sg->edat_level != edat_level ||
1417 		    sg->removed)
1418 			continue;
1419 		if (!sg->initialized)
1420 			return ERR_PTR(-EAGAIN);
1421 		atomic_inc(&sg->ref_count);
1422 		return sg;
1423 	}
1424 	return NULL;
1425 }
1426 
1427 /**
1428  * gmap_shadow_valid - check if a shadow guest address space matches the
1429  *                     given properties and is still valid
1430  * @sg: pointer to the shadow guest address space structure
1431  * @asce: ASCE for which the shadow table is requested
1432  * @edat_level: edat level to be used for the shadow translation
1433  *
1434  * Returns 1 if the gmap shadow is still valid and matches the given
1435  * properties, the caller can continue using it. Returns 0 otherwise, the
1436  * caller has to request a new shadow gmap in this case.
1437  *
1438  */
1439 int gmap_shadow_valid(struct gmap *sg, unsigned long asce, int edat_level)
1440 {
1441 	if (sg->removed)
1442 		return 0;
1443 	return sg->orig_asce == asce && sg->edat_level == edat_level;
1444 }
1445 EXPORT_SYMBOL_GPL(gmap_shadow_valid);
1446 
1447 /**
1448  * gmap_shadow - create/find a shadow guest address space
1449  * @parent: pointer to the parent gmap
1450  * @asce: ASCE for which the shadow table is created
1451  * @edat_level: edat level to be used for the shadow translation
1452  *
1453  * The pages of the top level page table referred by the asce parameter
1454  * will be set to read-only and marked in the PGSTEs of the kvm process.
1455  * The shadow table will be removed automatically on any change to the
1456  * PTE mapping for the source table.
1457  *
1458  * Returns a guest address space structure, ERR_PTR(-ENOMEM) if out of memory,
1459  * ERR_PTR(-EAGAIN) if the caller has to retry and ERR_PTR(-EFAULT) if the
1460  * parent gmap table could not be protected.
1461  */
1462 struct gmap *gmap_shadow(struct gmap *parent, unsigned long asce,
1463 			 int edat_level)
1464 {
1465 	struct gmap *sg, *new;
1466 	unsigned long limit;
1467 	int rc;
1468 
1469 	BUG_ON(gmap_is_shadow(parent));
1470 	spin_lock(&parent->shadow_lock);
1471 	sg = gmap_find_shadow(parent, asce, edat_level);
1472 	spin_unlock(&parent->shadow_lock);
1473 	if (sg)
1474 		return sg;
1475 	/* Create a new shadow gmap */
1476 	limit = -1UL >> (33 - (((asce & _ASCE_TYPE_MASK) >> 2) * 11));
1477 	if (asce & _ASCE_REAL_SPACE)
1478 		limit = -1UL;
1479 	new = gmap_alloc(limit);
1480 	if (!new)
1481 		return ERR_PTR(-ENOMEM);
1482 	new->mm = parent->mm;
1483 	new->parent = gmap_get(parent);
1484 	new->orig_asce = asce;
1485 	new->edat_level = edat_level;
1486 	new->initialized = false;
1487 	spin_lock(&parent->shadow_lock);
1488 	/* Recheck if another CPU created the same shadow */
1489 	sg = gmap_find_shadow(parent, asce, edat_level);
1490 	if (sg) {
1491 		spin_unlock(&parent->shadow_lock);
1492 		gmap_free(new);
1493 		return sg;
1494 	}
1495 	if (asce & _ASCE_REAL_SPACE) {
1496 		/* only allow one real-space gmap shadow */
1497 		list_for_each_entry(sg, &parent->children, list) {
1498 			if (sg->orig_asce & _ASCE_REAL_SPACE) {
1499 				spin_lock(&sg->guest_table_lock);
1500 				gmap_unshadow(sg);
1501 				spin_unlock(&sg->guest_table_lock);
1502 				list_del(&sg->list);
1503 				gmap_put(sg);
1504 				break;
1505 			}
1506 		}
1507 	}
1508 	atomic_set(&new->ref_count, 2);
1509 	list_add(&new->list, &parent->children);
1510 	if (asce & _ASCE_REAL_SPACE) {
1511 		/* nothing to protect, return right away */
1512 		new->initialized = true;
1513 		spin_unlock(&parent->shadow_lock);
1514 		return new;
1515 	}
1516 	spin_unlock(&parent->shadow_lock);
1517 	/* protect after insertion, so it will get properly invalidated */
1518 	down_read(&parent->mm->mmap_sem);
1519 	rc = gmap_protect_range(parent, asce & _ASCE_ORIGIN,
1520 				((asce & _ASCE_TABLE_LENGTH) + 1) * 4096,
1521 				PROT_READ, PGSTE_VSIE_BIT);
1522 	up_read(&parent->mm->mmap_sem);
1523 	spin_lock(&parent->shadow_lock);
1524 	new->initialized = true;
1525 	if (rc) {
1526 		list_del(&new->list);
1527 		gmap_free(new);
1528 		new = ERR_PTR(rc);
1529 	}
1530 	spin_unlock(&parent->shadow_lock);
1531 	return new;
1532 }
1533 EXPORT_SYMBOL_GPL(gmap_shadow);
1534 
1535 /**
1536  * gmap_shadow_r2t - create an empty shadow region 2 table
1537  * @sg: pointer to the shadow guest address space structure
1538  * @saddr: faulting address in the shadow gmap
1539  * @r2t: parent gmap address of the region 2 table to get shadowed
1540  * @fake: r2t references contiguous guest memory block, not a r2t
1541  *
1542  * The r2t parameter specifies the address of the source table. The
1543  * four pages of the source table are made read-only in the parent gmap
1544  * address space. A write to the source table area @r2t will automatically
1545  * remove the shadow r2 table and all of its decendents.
1546  *
1547  * Returns 0 if successfully shadowed or already shadowed, -EAGAIN if the
1548  * shadow table structure is incomplete, -ENOMEM if out of memory and
1549  * -EFAULT if an address in the parent gmap could not be resolved.
1550  *
1551  * Called with sg->mm->mmap_sem in read.
1552  */
1553 int gmap_shadow_r2t(struct gmap *sg, unsigned long saddr, unsigned long r2t,
1554 		    int fake)
1555 {
1556 	unsigned long raddr, origin, offset, len;
1557 	unsigned long *s_r2t, *table;
1558 	struct page *page;
1559 	int rc;
1560 
1561 	BUG_ON(!gmap_is_shadow(sg));
1562 	/* Allocate a shadow region second table */
1563 	page = alloc_pages(GFP_KERNEL, 2);
1564 	if (!page)
1565 		return -ENOMEM;
1566 	page->index = r2t & _REGION_ENTRY_ORIGIN;
1567 	if (fake)
1568 		page->index |= GMAP_SHADOW_FAKE_TABLE;
1569 	s_r2t = (unsigned long *) page_to_phys(page);
1570 	/* Install shadow region second table */
1571 	spin_lock(&sg->guest_table_lock);
1572 	table = gmap_table_walk(sg, saddr, 4); /* get region-1 pointer */
1573 	if (!table) {
1574 		rc = -EAGAIN;		/* Race with unshadow */
1575 		goto out_free;
1576 	}
1577 	if (!(*table & _REGION_ENTRY_INVALID)) {
1578 		rc = 0;			/* Already established */
1579 		goto out_free;
1580 	} else if (*table & _REGION_ENTRY_ORIGIN) {
1581 		rc = -EAGAIN;		/* Race with shadow */
1582 		goto out_free;
1583 	}
1584 	crst_table_init(s_r2t, _REGION2_ENTRY_EMPTY);
1585 	/* mark as invalid as long as the parent table is not protected */
1586 	*table = (unsigned long) s_r2t | _REGION_ENTRY_LENGTH |
1587 		 _REGION_ENTRY_TYPE_R1 | _REGION_ENTRY_INVALID;
1588 	if (sg->edat_level >= 1)
1589 		*table |= (r2t & _REGION_ENTRY_PROTECT);
1590 	list_add(&page->lru, &sg->crst_list);
1591 	if (fake) {
1592 		/* nothing to protect for fake tables */
1593 		*table &= ~_REGION_ENTRY_INVALID;
1594 		spin_unlock(&sg->guest_table_lock);
1595 		return 0;
1596 	}
1597 	spin_unlock(&sg->guest_table_lock);
1598 	/* Make r2t read-only in parent gmap page table */
1599 	raddr = (saddr & 0xffe0000000000000UL) | _SHADOW_RMAP_REGION1;
1600 	origin = r2t & _REGION_ENTRY_ORIGIN;
1601 	offset = ((r2t & _REGION_ENTRY_OFFSET) >> 6) * 4096;
1602 	len = ((r2t & _REGION_ENTRY_LENGTH) + 1) * 4096 - offset;
1603 	rc = gmap_protect_rmap(sg, raddr, origin + offset, len, PROT_READ);
1604 	spin_lock(&sg->guest_table_lock);
1605 	if (!rc) {
1606 		table = gmap_table_walk(sg, saddr, 4);
1607 		if (!table || (*table & _REGION_ENTRY_ORIGIN) !=
1608 			      (unsigned long) s_r2t)
1609 			rc = -EAGAIN;		/* Race with unshadow */
1610 		else
1611 			*table &= ~_REGION_ENTRY_INVALID;
1612 	} else {
1613 		gmap_unshadow_r2t(sg, raddr);
1614 	}
1615 	spin_unlock(&sg->guest_table_lock);
1616 	return rc;
1617 out_free:
1618 	spin_unlock(&sg->guest_table_lock);
1619 	__free_pages(page, 2);
1620 	return rc;
1621 }
1622 EXPORT_SYMBOL_GPL(gmap_shadow_r2t);
1623 
1624 /**
1625  * gmap_shadow_r3t - create a shadow region 3 table
1626  * @sg: pointer to the shadow guest address space structure
1627  * @saddr: faulting address in the shadow gmap
1628  * @r3t: parent gmap address of the region 3 table to get shadowed
1629  * @fake: r3t references contiguous guest memory block, not a r3t
1630  *
1631  * Returns 0 if successfully shadowed or already shadowed, -EAGAIN if the
1632  * shadow table structure is incomplete, -ENOMEM if out of memory and
1633  * -EFAULT if an address in the parent gmap could not be resolved.
1634  *
1635  * Called with sg->mm->mmap_sem in read.
1636  */
1637 int gmap_shadow_r3t(struct gmap *sg, unsigned long saddr, unsigned long r3t,
1638 		    int fake)
1639 {
1640 	unsigned long raddr, origin, offset, len;
1641 	unsigned long *s_r3t, *table;
1642 	struct page *page;
1643 	int rc;
1644 
1645 	BUG_ON(!gmap_is_shadow(sg));
1646 	/* Allocate a shadow region second table */
1647 	page = alloc_pages(GFP_KERNEL, 2);
1648 	if (!page)
1649 		return -ENOMEM;
1650 	page->index = r3t & _REGION_ENTRY_ORIGIN;
1651 	if (fake)
1652 		page->index |= GMAP_SHADOW_FAKE_TABLE;
1653 	s_r3t = (unsigned long *) page_to_phys(page);
1654 	/* Install shadow region second table */
1655 	spin_lock(&sg->guest_table_lock);
1656 	table = gmap_table_walk(sg, saddr, 3); /* get region-2 pointer */
1657 	if (!table) {
1658 		rc = -EAGAIN;		/* Race with unshadow */
1659 		goto out_free;
1660 	}
1661 	if (!(*table & _REGION_ENTRY_INVALID)) {
1662 		rc = 0;			/* Already established */
1663 		goto out_free;
1664 	} else if (*table & _REGION_ENTRY_ORIGIN) {
1665 		rc = -EAGAIN;		/* Race with shadow */
1666 	}
1667 	crst_table_init(s_r3t, _REGION3_ENTRY_EMPTY);
1668 	/* mark as invalid as long as the parent table is not protected */
1669 	*table = (unsigned long) s_r3t | _REGION_ENTRY_LENGTH |
1670 		 _REGION_ENTRY_TYPE_R2 | _REGION_ENTRY_INVALID;
1671 	if (sg->edat_level >= 1)
1672 		*table |= (r3t & _REGION_ENTRY_PROTECT);
1673 	list_add(&page->lru, &sg->crst_list);
1674 	if (fake) {
1675 		/* nothing to protect for fake tables */
1676 		*table &= ~_REGION_ENTRY_INVALID;
1677 		spin_unlock(&sg->guest_table_lock);
1678 		return 0;
1679 	}
1680 	spin_unlock(&sg->guest_table_lock);
1681 	/* Make r3t read-only in parent gmap page table */
1682 	raddr = (saddr & 0xfffffc0000000000UL) | _SHADOW_RMAP_REGION2;
1683 	origin = r3t & _REGION_ENTRY_ORIGIN;
1684 	offset = ((r3t & _REGION_ENTRY_OFFSET) >> 6) * 4096;
1685 	len = ((r3t & _REGION_ENTRY_LENGTH) + 1) * 4096 - offset;
1686 	rc = gmap_protect_rmap(sg, raddr, origin + offset, len, PROT_READ);
1687 	spin_lock(&sg->guest_table_lock);
1688 	if (!rc) {
1689 		table = gmap_table_walk(sg, saddr, 3);
1690 		if (!table || (*table & _REGION_ENTRY_ORIGIN) !=
1691 			      (unsigned long) s_r3t)
1692 			rc = -EAGAIN;		/* Race with unshadow */
1693 		else
1694 			*table &= ~_REGION_ENTRY_INVALID;
1695 	} else {
1696 		gmap_unshadow_r3t(sg, raddr);
1697 	}
1698 	spin_unlock(&sg->guest_table_lock);
1699 	return rc;
1700 out_free:
1701 	spin_unlock(&sg->guest_table_lock);
1702 	__free_pages(page, 2);
1703 	return rc;
1704 }
1705 EXPORT_SYMBOL_GPL(gmap_shadow_r3t);
1706 
1707 /**
1708  * gmap_shadow_sgt - create a shadow segment table
1709  * @sg: pointer to the shadow guest address space structure
1710  * @saddr: faulting address in the shadow gmap
1711  * @sgt: parent gmap address of the segment table to get shadowed
1712  * @fake: sgt references contiguous guest memory block, not a sgt
1713  *
1714  * Returns: 0 if successfully shadowed or already shadowed, -EAGAIN if the
1715  * shadow table structure is incomplete, -ENOMEM if out of memory and
1716  * -EFAULT if an address in the parent gmap could not be resolved.
1717  *
1718  * Called with sg->mm->mmap_sem in read.
1719  */
1720 int gmap_shadow_sgt(struct gmap *sg, unsigned long saddr, unsigned long sgt,
1721 		    int fake)
1722 {
1723 	unsigned long raddr, origin, offset, len;
1724 	unsigned long *s_sgt, *table;
1725 	struct page *page;
1726 	int rc;
1727 
1728 	BUG_ON(!gmap_is_shadow(sg) || (sgt & _REGION3_ENTRY_LARGE));
1729 	/* Allocate a shadow segment table */
1730 	page = alloc_pages(GFP_KERNEL, 2);
1731 	if (!page)
1732 		return -ENOMEM;
1733 	page->index = sgt & _REGION_ENTRY_ORIGIN;
1734 	if (fake)
1735 		page->index |= GMAP_SHADOW_FAKE_TABLE;
1736 	s_sgt = (unsigned long *) page_to_phys(page);
1737 	/* Install shadow region second table */
1738 	spin_lock(&sg->guest_table_lock);
1739 	table = gmap_table_walk(sg, saddr, 2); /* get region-3 pointer */
1740 	if (!table) {
1741 		rc = -EAGAIN;		/* Race with unshadow */
1742 		goto out_free;
1743 	}
1744 	if (!(*table & _REGION_ENTRY_INVALID)) {
1745 		rc = 0;			/* Already established */
1746 		goto out_free;
1747 	} else if (*table & _REGION_ENTRY_ORIGIN) {
1748 		rc = -EAGAIN;		/* Race with shadow */
1749 		goto out_free;
1750 	}
1751 	crst_table_init(s_sgt, _SEGMENT_ENTRY_EMPTY);
1752 	/* mark as invalid as long as the parent table is not protected */
1753 	*table = (unsigned long) s_sgt | _REGION_ENTRY_LENGTH |
1754 		 _REGION_ENTRY_TYPE_R3 | _REGION_ENTRY_INVALID;
1755 	if (sg->edat_level >= 1)
1756 		*table |= sgt & _REGION_ENTRY_PROTECT;
1757 	list_add(&page->lru, &sg->crst_list);
1758 	if (fake) {
1759 		/* nothing to protect for fake tables */
1760 		*table &= ~_REGION_ENTRY_INVALID;
1761 		spin_unlock(&sg->guest_table_lock);
1762 		return 0;
1763 	}
1764 	spin_unlock(&sg->guest_table_lock);
1765 	/* Make sgt read-only in parent gmap page table */
1766 	raddr = (saddr & 0xffffffff80000000UL) | _SHADOW_RMAP_REGION3;
1767 	origin = sgt & _REGION_ENTRY_ORIGIN;
1768 	offset = ((sgt & _REGION_ENTRY_OFFSET) >> 6) * 4096;
1769 	len = ((sgt & _REGION_ENTRY_LENGTH) + 1) * 4096 - offset;
1770 	rc = gmap_protect_rmap(sg, raddr, origin + offset, len, PROT_READ);
1771 	spin_lock(&sg->guest_table_lock);
1772 	if (!rc) {
1773 		table = gmap_table_walk(sg, saddr, 2);
1774 		if (!table || (*table & _REGION_ENTRY_ORIGIN) !=
1775 			      (unsigned long) s_sgt)
1776 			rc = -EAGAIN;		/* Race with unshadow */
1777 		else
1778 			*table &= ~_REGION_ENTRY_INVALID;
1779 	} else {
1780 		gmap_unshadow_sgt(sg, raddr);
1781 	}
1782 	spin_unlock(&sg->guest_table_lock);
1783 	return rc;
1784 out_free:
1785 	spin_unlock(&sg->guest_table_lock);
1786 	__free_pages(page, 2);
1787 	return rc;
1788 }
1789 EXPORT_SYMBOL_GPL(gmap_shadow_sgt);
1790 
1791 /**
1792  * gmap_shadow_lookup_pgtable - find a shadow page table
1793  * @sg: pointer to the shadow guest address space structure
1794  * @saddr: the address in the shadow aguest address space
1795  * @pgt: parent gmap address of the page table to get shadowed
1796  * @dat_protection: if the pgtable is marked as protected by dat
1797  * @fake: pgt references contiguous guest memory block, not a pgtable
1798  *
1799  * Returns 0 if the shadow page table was found and -EAGAIN if the page
1800  * table was not found.
1801  *
1802  * Called with sg->mm->mmap_sem in read.
1803  */
1804 int gmap_shadow_pgt_lookup(struct gmap *sg, unsigned long saddr,
1805 			   unsigned long *pgt, int *dat_protection,
1806 			   int *fake)
1807 {
1808 	unsigned long *table;
1809 	struct page *page;
1810 	int rc;
1811 
1812 	BUG_ON(!gmap_is_shadow(sg));
1813 	spin_lock(&sg->guest_table_lock);
1814 	table = gmap_table_walk(sg, saddr, 1); /* get segment pointer */
1815 	if (table && !(*table & _SEGMENT_ENTRY_INVALID)) {
1816 		/* Shadow page tables are full pages (pte+pgste) */
1817 		page = pfn_to_page(*table >> PAGE_SHIFT);
1818 		*pgt = page->index & ~GMAP_SHADOW_FAKE_TABLE;
1819 		*dat_protection = !!(*table & _SEGMENT_ENTRY_PROTECT);
1820 		*fake = !!(page->index & GMAP_SHADOW_FAKE_TABLE);
1821 		rc = 0;
1822 	} else  {
1823 		rc = -EAGAIN;
1824 	}
1825 	spin_unlock(&sg->guest_table_lock);
1826 	return rc;
1827 
1828 }
1829 EXPORT_SYMBOL_GPL(gmap_shadow_pgt_lookup);
1830 
1831 /**
1832  * gmap_shadow_pgt - instantiate a shadow page table
1833  * @sg: pointer to the shadow guest address space structure
1834  * @saddr: faulting address in the shadow gmap
1835  * @pgt: parent gmap address of the page table to get shadowed
1836  * @fake: pgt references contiguous guest memory block, not a pgtable
1837  *
1838  * Returns 0 if successfully shadowed or already shadowed, -EAGAIN if the
1839  * shadow table structure is incomplete, -ENOMEM if out of memory,
1840  * -EFAULT if an address in the parent gmap could not be resolved and
1841  *
1842  * Called with gmap->mm->mmap_sem in read
1843  */
1844 int gmap_shadow_pgt(struct gmap *sg, unsigned long saddr, unsigned long pgt,
1845 		    int fake)
1846 {
1847 	unsigned long raddr, origin;
1848 	unsigned long *s_pgt, *table;
1849 	struct page *page;
1850 	int rc;
1851 
1852 	BUG_ON(!gmap_is_shadow(sg) || (pgt & _SEGMENT_ENTRY_LARGE));
1853 	/* Allocate a shadow page table */
1854 	page = page_table_alloc_pgste(sg->mm);
1855 	if (!page)
1856 		return -ENOMEM;
1857 	page->index = pgt & _SEGMENT_ENTRY_ORIGIN;
1858 	if (fake)
1859 		page->index |= GMAP_SHADOW_FAKE_TABLE;
1860 	s_pgt = (unsigned long *) page_to_phys(page);
1861 	/* Install shadow page table */
1862 	spin_lock(&sg->guest_table_lock);
1863 	table = gmap_table_walk(sg, saddr, 1); /* get segment pointer */
1864 	if (!table) {
1865 		rc = -EAGAIN;		/* Race with unshadow */
1866 		goto out_free;
1867 	}
1868 	if (!(*table & _SEGMENT_ENTRY_INVALID)) {
1869 		rc = 0;			/* Already established */
1870 		goto out_free;
1871 	} else if (*table & _SEGMENT_ENTRY_ORIGIN) {
1872 		rc = -EAGAIN;		/* Race with shadow */
1873 		goto out_free;
1874 	}
1875 	/* mark as invalid as long as the parent table is not protected */
1876 	*table = (unsigned long) s_pgt | _SEGMENT_ENTRY |
1877 		 (pgt & _SEGMENT_ENTRY_PROTECT) | _SEGMENT_ENTRY_INVALID;
1878 	list_add(&page->lru, &sg->pt_list);
1879 	if (fake) {
1880 		/* nothing to protect for fake tables */
1881 		*table &= ~_SEGMENT_ENTRY_INVALID;
1882 		spin_unlock(&sg->guest_table_lock);
1883 		return 0;
1884 	}
1885 	spin_unlock(&sg->guest_table_lock);
1886 	/* Make pgt read-only in parent gmap page table (not the pgste) */
1887 	raddr = (saddr & 0xfffffffffff00000UL) | _SHADOW_RMAP_SEGMENT;
1888 	origin = pgt & _SEGMENT_ENTRY_ORIGIN & PAGE_MASK;
1889 	rc = gmap_protect_rmap(sg, raddr, origin, PAGE_SIZE, PROT_READ);
1890 	spin_lock(&sg->guest_table_lock);
1891 	if (!rc) {
1892 		table = gmap_table_walk(sg, saddr, 1);
1893 		if (!table || (*table & _SEGMENT_ENTRY_ORIGIN) !=
1894 			      (unsigned long) s_pgt)
1895 			rc = -EAGAIN;		/* Race with unshadow */
1896 		else
1897 			*table &= ~_SEGMENT_ENTRY_INVALID;
1898 	} else {
1899 		gmap_unshadow_pgt(sg, raddr);
1900 	}
1901 	spin_unlock(&sg->guest_table_lock);
1902 	return rc;
1903 out_free:
1904 	spin_unlock(&sg->guest_table_lock);
1905 	page_table_free_pgste(page);
1906 	return rc;
1907 
1908 }
1909 EXPORT_SYMBOL_GPL(gmap_shadow_pgt);
1910 
1911 /**
1912  * gmap_shadow_page - create a shadow page mapping
1913  * @sg: pointer to the shadow guest address space structure
1914  * @saddr: faulting address in the shadow gmap
1915  * @pte: pte in parent gmap address space to get shadowed
1916  *
1917  * Returns 0 if successfully shadowed or already shadowed, -EAGAIN if the
1918  * shadow table structure is incomplete, -ENOMEM if out of memory and
1919  * -EFAULT if an address in the parent gmap could not be resolved.
1920  *
1921  * Called with sg->mm->mmap_sem in read.
1922  */
1923 int gmap_shadow_page(struct gmap *sg, unsigned long saddr, pte_t pte)
1924 {
1925 	struct gmap *parent;
1926 	struct gmap_rmap *rmap;
1927 	unsigned long vmaddr, paddr;
1928 	spinlock_t *ptl;
1929 	pte_t *sptep, *tptep;
1930 	int prot;
1931 	int rc;
1932 
1933 	BUG_ON(!gmap_is_shadow(sg));
1934 	parent = sg->parent;
1935 	prot = (pte_val(pte) & _PAGE_PROTECT) ? PROT_READ : PROT_WRITE;
1936 
1937 	rmap = kzalloc(sizeof(*rmap), GFP_KERNEL);
1938 	if (!rmap)
1939 		return -ENOMEM;
1940 	rmap->raddr = (saddr & PAGE_MASK) | _SHADOW_RMAP_PGTABLE;
1941 
1942 	while (1) {
1943 		paddr = pte_val(pte) & PAGE_MASK;
1944 		vmaddr = __gmap_translate(parent, paddr);
1945 		if (IS_ERR_VALUE(vmaddr)) {
1946 			rc = vmaddr;
1947 			break;
1948 		}
1949 		rc = radix_tree_preload(GFP_KERNEL);
1950 		if (rc)
1951 			break;
1952 		rc = -EAGAIN;
1953 		sptep = gmap_pte_op_walk(parent, paddr, &ptl);
1954 		if (sptep) {
1955 			spin_lock(&sg->guest_table_lock);
1956 			/* Get page table pointer */
1957 			tptep = (pte_t *) gmap_table_walk(sg, saddr, 0);
1958 			if (!tptep) {
1959 				spin_unlock(&sg->guest_table_lock);
1960 				gmap_pte_op_end(ptl);
1961 				radix_tree_preload_end();
1962 				break;
1963 			}
1964 			rc = ptep_shadow_pte(sg->mm, saddr, sptep, tptep, pte);
1965 			if (rc > 0) {
1966 				/* Success and a new mapping */
1967 				gmap_insert_rmap(sg, vmaddr, rmap);
1968 				rmap = NULL;
1969 				rc = 0;
1970 			}
1971 			gmap_pte_op_end(ptl);
1972 			spin_unlock(&sg->guest_table_lock);
1973 		}
1974 		radix_tree_preload_end();
1975 		if (!rc)
1976 			break;
1977 		rc = gmap_pte_op_fixup(parent, paddr, vmaddr, prot);
1978 		if (rc)
1979 			break;
1980 	}
1981 	kfree(rmap);
1982 	return rc;
1983 }
1984 EXPORT_SYMBOL_GPL(gmap_shadow_page);
1985 
1986 /**
1987  * gmap_shadow_notify - handle notifications for shadow gmap
1988  *
1989  * Called with sg->parent->shadow_lock.
1990  */
1991 static void gmap_shadow_notify(struct gmap *sg, unsigned long vmaddr,
1992 			       unsigned long offset, pte_t *pte)
1993 {
1994 	struct gmap_rmap *rmap, *rnext, *head;
1995 	unsigned long gaddr, start, end, bits, raddr;
1996 	unsigned long *table;
1997 
1998 	BUG_ON(!gmap_is_shadow(sg));
1999 	spin_lock(&sg->parent->guest_table_lock);
2000 	table = radix_tree_lookup(&sg->parent->host_to_guest,
2001 				  vmaddr >> PMD_SHIFT);
2002 	gaddr = table ? __gmap_segment_gaddr(table) + offset : 0;
2003 	spin_unlock(&sg->parent->guest_table_lock);
2004 	if (!table)
2005 		return;
2006 
2007 	spin_lock(&sg->guest_table_lock);
2008 	if (sg->removed) {
2009 		spin_unlock(&sg->guest_table_lock);
2010 		return;
2011 	}
2012 	/* Check for top level table */
2013 	start = sg->orig_asce & _ASCE_ORIGIN;
2014 	end = start + ((sg->orig_asce & _ASCE_TABLE_LENGTH) + 1) * 4096;
2015 	if (!(sg->orig_asce & _ASCE_REAL_SPACE) && gaddr >= start &&
2016 	    gaddr < end) {
2017 		/* The complete shadow table has to go */
2018 		gmap_unshadow(sg);
2019 		spin_unlock(&sg->guest_table_lock);
2020 		list_del(&sg->list);
2021 		gmap_put(sg);
2022 		return;
2023 	}
2024 	/* Remove the page table tree from on specific entry */
2025 	head = radix_tree_delete(&sg->host_to_rmap, vmaddr >> 12);
2026 	gmap_for_each_rmap_safe(rmap, rnext, head) {
2027 		bits = rmap->raddr & _SHADOW_RMAP_MASK;
2028 		raddr = rmap->raddr ^ bits;
2029 		switch (bits) {
2030 		case _SHADOW_RMAP_REGION1:
2031 			gmap_unshadow_r2t(sg, raddr);
2032 			break;
2033 		case _SHADOW_RMAP_REGION2:
2034 			gmap_unshadow_r3t(sg, raddr);
2035 			break;
2036 		case _SHADOW_RMAP_REGION3:
2037 			gmap_unshadow_sgt(sg, raddr);
2038 			break;
2039 		case _SHADOW_RMAP_SEGMENT:
2040 			gmap_unshadow_pgt(sg, raddr);
2041 			break;
2042 		case _SHADOW_RMAP_PGTABLE:
2043 			gmap_unshadow_page(sg, raddr);
2044 			break;
2045 		}
2046 		kfree(rmap);
2047 	}
2048 	spin_unlock(&sg->guest_table_lock);
2049 }
2050 
2051 /**
2052  * ptep_notify - call all invalidation callbacks for a specific pte.
2053  * @mm: pointer to the process mm_struct
2054  * @addr: virtual address in the process address space
2055  * @pte: pointer to the page table entry
2056  * @bits: bits from the pgste that caused the notify call
2057  *
2058  * This function is assumed to be called with the page table lock held
2059  * for the pte to notify.
2060  */
2061 void ptep_notify(struct mm_struct *mm, unsigned long vmaddr,
2062 		 pte_t *pte, unsigned long bits)
2063 {
2064 	unsigned long offset, gaddr;
2065 	unsigned long *table;
2066 	struct gmap *gmap, *sg, *next;
2067 
2068 	offset = ((unsigned long) pte) & (255 * sizeof(pte_t));
2069 	offset = offset * (4096 / sizeof(pte_t));
2070 	rcu_read_lock();
2071 	list_for_each_entry_rcu(gmap, &mm->context.gmap_list, list) {
2072 		if (!list_empty(&gmap->children) && (bits & PGSTE_VSIE_BIT)) {
2073 			spin_lock(&gmap->shadow_lock);
2074 			list_for_each_entry_safe(sg, next,
2075 						 &gmap->children, list)
2076 				gmap_shadow_notify(sg, vmaddr, offset, pte);
2077 			spin_unlock(&gmap->shadow_lock);
2078 		}
2079 		if (!(bits & PGSTE_IN_BIT))
2080 			continue;
2081 		spin_lock(&gmap->guest_table_lock);
2082 		table = radix_tree_lookup(&gmap->host_to_guest,
2083 					  vmaddr >> PMD_SHIFT);
2084 		if (table)
2085 			gaddr = __gmap_segment_gaddr(table) + offset;
2086 		spin_unlock(&gmap->guest_table_lock);
2087 		if (table)
2088 			gmap_call_notifier(gmap, gaddr, gaddr + PAGE_SIZE - 1);
2089 	}
2090 	rcu_read_unlock();
2091 }
2092 EXPORT_SYMBOL_GPL(ptep_notify);
2093 
2094 static inline void thp_split_mm(struct mm_struct *mm)
2095 {
2096 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
2097 	struct vm_area_struct *vma;
2098 	unsigned long addr;
2099 
2100 	for (vma = mm->mmap; vma != NULL; vma = vma->vm_next) {
2101 		for (addr = vma->vm_start;
2102 		     addr < vma->vm_end;
2103 		     addr += PAGE_SIZE)
2104 			follow_page(vma, addr, FOLL_SPLIT);
2105 		vma->vm_flags &= ~VM_HUGEPAGE;
2106 		vma->vm_flags |= VM_NOHUGEPAGE;
2107 	}
2108 	mm->def_flags |= VM_NOHUGEPAGE;
2109 #endif
2110 }
2111 
2112 /*
2113  * switch on pgstes for its userspace process (for kvm)
2114  */
2115 int s390_enable_sie(void)
2116 {
2117 	struct mm_struct *mm = current->mm;
2118 
2119 	/* Do we have pgstes? if yes, we are done */
2120 	if (mm_has_pgste(mm))
2121 		return 0;
2122 	/* Fail if the page tables are 2K */
2123 	if (!mm_alloc_pgste(mm))
2124 		return -EINVAL;
2125 	down_write(&mm->mmap_sem);
2126 	mm->context.has_pgste = 1;
2127 	/* split thp mappings and disable thp for future mappings */
2128 	thp_split_mm(mm);
2129 	up_write(&mm->mmap_sem);
2130 	return 0;
2131 }
2132 EXPORT_SYMBOL_GPL(s390_enable_sie);
2133 
2134 /*
2135  * Enable storage key handling from now on and initialize the storage
2136  * keys with the default key.
2137  */
2138 static int __s390_enable_skey(pte_t *pte, unsigned long addr,
2139 			      unsigned long next, struct mm_walk *walk)
2140 {
2141 	/*
2142 	 * Remove all zero page mappings,
2143 	 * after establishing a policy to forbid zero page mappings
2144 	 * following faults for that page will get fresh anonymous pages
2145 	 */
2146 	if (is_zero_pfn(pte_pfn(*pte)))
2147 		ptep_xchg_direct(walk->mm, addr, pte, __pte(_PAGE_INVALID));
2148 	/* Clear storage key */
2149 	ptep_zap_key(walk->mm, addr, pte);
2150 	return 0;
2151 }
2152 
2153 int s390_enable_skey(void)
2154 {
2155 	struct mm_walk walk = { .pte_entry = __s390_enable_skey };
2156 	struct mm_struct *mm = current->mm;
2157 	struct vm_area_struct *vma;
2158 	int rc = 0;
2159 
2160 	down_write(&mm->mmap_sem);
2161 	if (mm_use_skey(mm))
2162 		goto out_up;
2163 
2164 	mm->context.use_skey = 1;
2165 	for (vma = mm->mmap; vma; vma = vma->vm_next) {
2166 		if (ksm_madvise(vma, vma->vm_start, vma->vm_end,
2167 				MADV_UNMERGEABLE, &vma->vm_flags)) {
2168 			mm->context.use_skey = 0;
2169 			rc = -ENOMEM;
2170 			goto out_up;
2171 		}
2172 	}
2173 	mm->def_flags &= ~VM_MERGEABLE;
2174 
2175 	walk.mm = mm;
2176 	walk_page_range(0, TASK_SIZE, &walk);
2177 
2178 out_up:
2179 	up_write(&mm->mmap_sem);
2180 	return rc;
2181 }
2182 EXPORT_SYMBOL_GPL(s390_enable_skey);
2183 
2184 /*
2185  * Reset CMMA state, make all pages stable again.
2186  */
2187 static int __s390_reset_cmma(pte_t *pte, unsigned long addr,
2188 			     unsigned long next, struct mm_walk *walk)
2189 {
2190 	ptep_zap_unused(walk->mm, addr, pte, 1);
2191 	return 0;
2192 }
2193 
2194 void s390_reset_cmma(struct mm_struct *mm)
2195 {
2196 	struct mm_walk walk = { .pte_entry = __s390_reset_cmma };
2197 
2198 	down_write(&mm->mmap_sem);
2199 	walk.mm = mm;
2200 	walk_page_range(0, TASK_SIZE, &walk);
2201 	up_write(&mm->mmap_sem);
2202 }
2203 EXPORT_SYMBOL_GPL(s390_reset_cmma);
2204