1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  *  KVM guest address space mapping code
4  *
5  *    Copyright IBM Corp. 2007, 2020
6  *    Author(s): Martin Schwidefsky <schwidefsky@de.ibm.com>
7  *		 David Hildenbrand <david@redhat.com>
8  *		 Janosch Frank <frankja@linux.vnet.ibm.com>
9  */
10 
11 #include <linux/kernel.h>
12 #include <linux/pagewalk.h>
13 #include <linux/swap.h>
14 #include <linux/smp.h>
15 #include <linux/spinlock.h>
16 #include <linux/slab.h>
17 #include <linux/swapops.h>
18 #include <linux/ksm.h>
19 #include <linux/mman.h>
20 #include <linux/pgtable.h>
21 
22 #include <asm/pgalloc.h>
23 #include <asm/gmap.h>
24 #include <asm/page.h>
25 #include <asm/tlb.h>
26 
27 #define GMAP_SHADOW_FAKE_TABLE 1ULL
28 
gmap_alloc_crst(void)29 static struct page *gmap_alloc_crst(void)
30 {
31 	struct page *page;
32 
33 	page = alloc_pages(GFP_KERNEL_ACCOUNT, CRST_ALLOC_ORDER);
34 	if (!page)
35 		return NULL;
36 	arch_set_page_dat(page, CRST_ALLOC_ORDER);
37 	return page;
38 }
39 
40 /**
41  * gmap_alloc - allocate and initialize a guest address space
42  * @limit: maximum address of the gmap address space
43  *
44  * Returns a guest address space structure.
45  */
gmap_alloc(unsigned long limit)46 static struct gmap *gmap_alloc(unsigned long limit)
47 {
48 	struct gmap *gmap;
49 	struct page *page;
50 	unsigned long *table;
51 	unsigned long etype, atype;
52 
53 	if (limit < _REGION3_SIZE) {
54 		limit = _REGION3_SIZE - 1;
55 		atype = _ASCE_TYPE_SEGMENT;
56 		etype = _SEGMENT_ENTRY_EMPTY;
57 	} else if (limit < _REGION2_SIZE) {
58 		limit = _REGION2_SIZE - 1;
59 		atype = _ASCE_TYPE_REGION3;
60 		etype = _REGION3_ENTRY_EMPTY;
61 	} else if (limit < _REGION1_SIZE) {
62 		limit = _REGION1_SIZE - 1;
63 		atype = _ASCE_TYPE_REGION2;
64 		etype = _REGION2_ENTRY_EMPTY;
65 	} else {
66 		limit = -1UL;
67 		atype = _ASCE_TYPE_REGION1;
68 		etype = _REGION1_ENTRY_EMPTY;
69 	}
70 	gmap = kzalloc(sizeof(struct gmap), GFP_KERNEL_ACCOUNT);
71 	if (!gmap)
72 		goto out;
73 	INIT_LIST_HEAD(&gmap->crst_list);
74 	INIT_LIST_HEAD(&gmap->children);
75 	INIT_LIST_HEAD(&gmap->pt_list);
76 	INIT_RADIX_TREE(&gmap->guest_to_host, GFP_KERNEL_ACCOUNT);
77 	INIT_RADIX_TREE(&gmap->host_to_guest, GFP_ATOMIC | __GFP_ACCOUNT);
78 	INIT_RADIX_TREE(&gmap->host_to_rmap, GFP_ATOMIC | __GFP_ACCOUNT);
79 	spin_lock_init(&gmap->guest_table_lock);
80 	spin_lock_init(&gmap->shadow_lock);
81 	refcount_set(&gmap->ref_count, 1);
82 	page = gmap_alloc_crst();
83 	if (!page)
84 		goto out_free;
85 	page->index = 0;
86 	list_add(&page->lru, &gmap->crst_list);
87 	table = page_to_virt(page);
88 	crst_table_init(table, etype);
89 	gmap->table = table;
90 	gmap->asce = atype | _ASCE_TABLE_LENGTH |
91 		_ASCE_USER_BITS | __pa(table);
92 	gmap->asce_end = limit;
93 	return gmap;
94 
95 out_free:
96 	kfree(gmap);
97 out:
98 	return NULL;
99 }
100 
101 /**
102  * gmap_create - create a guest address space
103  * @mm: pointer to the parent mm_struct
104  * @limit: maximum size of the gmap address space
105  *
106  * Returns a guest address space structure.
107  */
gmap_create(struct mm_struct * mm,unsigned long limit)108 struct gmap *gmap_create(struct mm_struct *mm, unsigned long limit)
109 {
110 	struct gmap *gmap;
111 	unsigned long gmap_asce;
112 
113 	gmap = gmap_alloc(limit);
114 	if (!gmap)
115 		return NULL;
116 	gmap->mm = mm;
117 	spin_lock(&mm->context.lock);
118 	list_add_rcu(&gmap->list, &mm->context.gmap_list);
119 	if (list_is_singular(&mm->context.gmap_list))
120 		gmap_asce = gmap->asce;
121 	else
122 		gmap_asce = -1UL;
123 	WRITE_ONCE(mm->context.gmap_asce, gmap_asce);
124 	spin_unlock(&mm->context.lock);
125 	return gmap;
126 }
127 EXPORT_SYMBOL_GPL(gmap_create);
128 
gmap_flush_tlb(struct gmap * gmap)129 static void gmap_flush_tlb(struct gmap *gmap)
130 {
131 	if (MACHINE_HAS_IDTE)
132 		__tlb_flush_idte(gmap->asce);
133 	else
134 		__tlb_flush_global();
135 }
136 
gmap_radix_tree_free(struct radix_tree_root * root)137 static void gmap_radix_tree_free(struct radix_tree_root *root)
138 {
139 	struct radix_tree_iter iter;
140 	unsigned long indices[16];
141 	unsigned long index;
142 	void __rcu **slot;
143 	int i, nr;
144 
145 	/* A radix tree is freed by deleting all of its entries */
146 	index = 0;
147 	do {
148 		nr = 0;
149 		radix_tree_for_each_slot(slot, root, &iter, index) {
150 			indices[nr] = iter.index;
151 			if (++nr == 16)
152 				break;
153 		}
154 		for (i = 0; i < nr; i++) {
155 			index = indices[i];
156 			radix_tree_delete(root, index);
157 		}
158 	} while (nr > 0);
159 }
160 
gmap_rmap_radix_tree_free(struct radix_tree_root * root)161 static void gmap_rmap_radix_tree_free(struct radix_tree_root *root)
162 {
163 	struct gmap_rmap *rmap, *rnext, *head;
164 	struct radix_tree_iter iter;
165 	unsigned long indices[16];
166 	unsigned long index;
167 	void __rcu **slot;
168 	int i, nr;
169 
170 	/* A radix tree is freed by deleting all of its entries */
171 	index = 0;
172 	do {
173 		nr = 0;
174 		radix_tree_for_each_slot(slot, root, &iter, index) {
175 			indices[nr] = iter.index;
176 			if (++nr == 16)
177 				break;
178 		}
179 		for (i = 0; i < nr; i++) {
180 			index = indices[i];
181 			head = radix_tree_delete(root, index);
182 			gmap_for_each_rmap_safe(rmap, rnext, head)
183 				kfree(rmap);
184 		}
185 	} while (nr > 0);
186 }
187 
188 /**
189  * gmap_free - free a guest address space
190  * @gmap: pointer to the guest address space structure
191  *
192  * No locks required. There are no references to this gmap anymore.
193  */
gmap_free(struct gmap * gmap)194 static void gmap_free(struct gmap *gmap)
195 {
196 	struct page *page, *next;
197 
198 	/* Flush tlb of all gmaps (if not already done for shadows) */
199 	if (!(gmap_is_shadow(gmap) && gmap->removed))
200 		gmap_flush_tlb(gmap);
201 	/* Free all segment & region tables. */
202 	list_for_each_entry_safe(page, next, &gmap->crst_list, lru)
203 		__free_pages(page, CRST_ALLOC_ORDER);
204 	gmap_radix_tree_free(&gmap->guest_to_host);
205 	gmap_radix_tree_free(&gmap->host_to_guest);
206 
207 	/* Free additional data for a shadow gmap */
208 	if (gmap_is_shadow(gmap)) {
209 		/* Free all page tables. */
210 		list_for_each_entry_safe(page, next, &gmap->pt_list, lru)
211 			page_table_free_pgste(page);
212 		gmap_rmap_radix_tree_free(&gmap->host_to_rmap);
213 		/* Release reference to the parent */
214 		gmap_put(gmap->parent);
215 	}
216 
217 	kfree(gmap);
218 }
219 
220 /**
221  * gmap_get - increase reference counter for guest address space
222  * @gmap: pointer to the guest address space structure
223  *
224  * Returns the gmap pointer
225  */
gmap_get(struct gmap * gmap)226 struct gmap *gmap_get(struct gmap *gmap)
227 {
228 	refcount_inc(&gmap->ref_count);
229 	return gmap;
230 }
231 EXPORT_SYMBOL_GPL(gmap_get);
232 
233 /**
234  * gmap_put - decrease reference counter for guest address space
235  * @gmap: pointer to the guest address space structure
236  *
237  * If the reference counter reaches zero the guest address space is freed.
238  */
gmap_put(struct gmap * gmap)239 void gmap_put(struct gmap *gmap)
240 {
241 	if (refcount_dec_and_test(&gmap->ref_count))
242 		gmap_free(gmap);
243 }
244 EXPORT_SYMBOL_GPL(gmap_put);
245 
246 /**
247  * gmap_remove - remove a guest address space but do not free it yet
248  * @gmap: pointer to the guest address space structure
249  */
gmap_remove(struct gmap * gmap)250 void gmap_remove(struct gmap *gmap)
251 {
252 	struct gmap *sg, *next;
253 	unsigned long gmap_asce;
254 
255 	/* Remove all shadow gmaps linked to this gmap */
256 	if (!list_empty(&gmap->children)) {
257 		spin_lock(&gmap->shadow_lock);
258 		list_for_each_entry_safe(sg, next, &gmap->children, list) {
259 			list_del(&sg->list);
260 			gmap_put(sg);
261 		}
262 		spin_unlock(&gmap->shadow_lock);
263 	}
264 	/* Remove gmap from the pre-mm list */
265 	spin_lock(&gmap->mm->context.lock);
266 	list_del_rcu(&gmap->list);
267 	if (list_empty(&gmap->mm->context.gmap_list))
268 		gmap_asce = 0;
269 	else if (list_is_singular(&gmap->mm->context.gmap_list))
270 		gmap_asce = list_first_entry(&gmap->mm->context.gmap_list,
271 					     struct gmap, list)->asce;
272 	else
273 		gmap_asce = -1UL;
274 	WRITE_ONCE(gmap->mm->context.gmap_asce, gmap_asce);
275 	spin_unlock(&gmap->mm->context.lock);
276 	synchronize_rcu();
277 	/* Put reference */
278 	gmap_put(gmap);
279 }
280 EXPORT_SYMBOL_GPL(gmap_remove);
281 
282 /**
283  * gmap_enable - switch primary space to the guest address space
284  * @gmap: pointer to the guest address space structure
285  */
gmap_enable(struct gmap * gmap)286 void gmap_enable(struct gmap *gmap)
287 {
288 	S390_lowcore.gmap = (unsigned long) gmap;
289 }
290 EXPORT_SYMBOL_GPL(gmap_enable);
291 
292 /**
293  * gmap_disable - switch back to the standard primary address space
294  * @gmap: pointer to the guest address space structure
295  */
gmap_disable(struct gmap * gmap)296 void gmap_disable(struct gmap *gmap)
297 {
298 	S390_lowcore.gmap = 0UL;
299 }
300 EXPORT_SYMBOL_GPL(gmap_disable);
301 
302 /**
303  * gmap_get_enabled - get a pointer to the currently enabled gmap
304  *
305  * Returns a pointer to the currently enabled gmap. 0 if none is enabled.
306  */
gmap_get_enabled(void)307 struct gmap *gmap_get_enabled(void)
308 {
309 	return (struct gmap *) S390_lowcore.gmap;
310 }
311 EXPORT_SYMBOL_GPL(gmap_get_enabled);
312 
313 /*
314  * gmap_alloc_table is assumed to be called with mmap_lock held
315  */
gmap_alloc_table(struct gmap * gmap,unsigned long * table,unsigned long init,unsigned long gaddr)316 static int gmap_alloc_table(struct gmap *gmap, unsigned long *table,
317 			    unsigned long init, unsigned long gaddr)
318 {
319 	struct page *page;
320 	unsigned long *new;
321 
322 	/* since we dont free the gmap table until gmap_free we can unlock */
323 	page = gmap_alloc_crst();
324 	if (!page)
325 		return -ENOMEM;
326 	new = page_to_virt(page);
327 	crst_table_init(new, init);
328 	spin_lock(&gmap->guest_table_lock);
329 	if (*table & _REGION_ENTRY_INVALID) {
330 		list_add(&page->lru, &gmap->crst_list);
331 		*table = __pa(new) | _REGION_ENTRY_LENGTH |
332 			(*table & _REGION_ENTRY_TYPE_MASK);
333 		page->index = gaddr;
334 		page = NULL;
335 	}
336 	spin_unlock(&gmap->guest_table_lock);
337 	if (page)
338 		__free_pages(page, CRST_ALLOC_ORDER);
339 	return 0;
340 }
341 
342 /**
343  * __gmap_segment_gaddr - find virtual address from segment pointer
344  * @entry: pointer to a segment table entry in the guest address space
345  *
346  * Returns the virtual address in the guest address space for the segment
347  */
__gmap_segment_gaddr(unsigned long * entry)348 static unsigned long __gmap_segment_gaddr(unsigned long *entry)
349 {
350 	struct page *page;
351 	unsigned long offset;
352 
353 	offset = (unsigned long) entry / sizeof(unsigned long);
354 	offset = (offset & (PTRS_PER_PMD - 1)) * PMD_SIZE;
355 	page = pmd_pgtable_page((pmd_t *) entry);
356 	return page->index + offset;
357 }
358 
359 /**
360  * __gmap_unlink_by_vmaddr - unlink a single segment via a host address
361  * @gmap: pointer to the guest address space structure
362  * @vmaddr: address in the host process address space
363  *
364  * Returns 1 if a TLB flush is required
365  */
__gmap_unlink_by_vmaddr(struct gmap * gmap,unsigned long vmaddr)366 static int __gmap_unlink_by_vmaddr(struct gmap *gmap, unsigned long vmaddr)
367 {
368 	unsigned long *entry;
369 	int flush = 0;
370 
371 	BUG_ON(gmap_is_shadow(gmap));
372 	spin_lock(&gmap->guest_table_lock);
373 	entry = radix_tree_delete(&gmap->host_to_guest, vmaddr >> PMD_SHIFT);
374 	if (entry) {
375 		flush = (*entry != _SEGMENT_ENTRY_EMPTY);
376 		*entry = _SEGMENT_ENTRY_EMPTY;
377 	}
378 	spin_unlock(&gmap->guest_table_lock);
379 	return flush;
380 }
381 
382 /**
383  * __gmap_unmap_by_gaddr - unmap a single segment via a guest address
384  * @gmap: pointer to the guest address space structure
385  * @gaddr: address in the guest address space
386  *
387  * Returns 1 if a TLB flush is required
388  */
__gmap_unmap_by_gaddr(struct gmap * gmap,unsigned long gaddr)389 static int __gmap_unmap_by_gaddr(struct gmap *gmap, unsigned long gaddr)
390 {
391 	unsigned long vmaddr;
392 
393 	vmaddr = (unsigned long) radix_tree_delete(&gmap->guest_to_host,
394 						   gaddr >> PMD_SHIFT);
395 	return vmaddr ? __gmap_unlink_by_vmaddr(gmap, vmaddr) : 0;
396 }
397 
398 /**
399  * gmap_unmap_segment - unmap segment from the guest address space
400  * @gmap: pointer to the guest address space structure
401  * @to: address in the guest address space
402  * @len: length of the memory area to unmap
403  *
404  * Returns 0 if the unmap succeeded, -EINVAL if not.
405  */
gmap_unmap_segment(struct gmap * gmap,unsigned long to,unsigned long len)406 int gmap_unmap_segment(struct gmap *gmap, unsigned long to, unsigned long len)
407 {
408 	unsigned long off;
409 	int flush;
410 
411 	BUG_ON(gmap_is_shadow(gmap));
412 	if ((to | len) & (PMD_SIZE - 1))
413 		return -EINVAL;
414 	if (len == 0 || to + len < to)
415 		return -EINVAL;
416 
417 	flush = 0;
418 	mmap_write_lock(gmap->mm);
419 	for (off = 0; off < len; off += PMD_SIZE)
420 		flush |= __gmap_unmap_by_gaddr(gmap, to + off);
421 	mmap_write_unlock(gmap->mm);
422 	if (flush)
423 		gmap_flush_tlb(gmap);
424 	return 0;
425 }
426 EXPORT_SYMBOL_GPL(gmap_unmap_segment);
427 
428 /**
429  * gmap_map_segment - map a segment to the guest address space
430  * @gmap: pointer to the guest address space structure
431  * @from: source address in the parent address space
432  * @to: target address in the guest address space
433  * @len: length of the memory area to map
434  *
435  * Returns 0 if the mmap succeeded, -EINVAL or -ENOMEM if not.
436  */
gmap_map_segment(struct gmap * gmap,unsigned long from,unsigned long to,unsigned long len)437 int gmap_map_segment(struct gmap *gmap, unsigned long from,
438 		     unsigned long to, unsigned long len)
439 {
440 	unsigned long off;
441 	int flush;
442 
443 	BUG_ON(gmap_is_shadow(gmap));
444 	if ((from | to | len) & (PMD_SIZE - 1))
445 		return -EINVAL;
446 	if (len == 0 || from + len < from || to + len < to ||
447 	    from + len - 1 > TASK_SIZE_MAX || to + len - 1 > gmap->asce_end)
448 		return -EINVAL;
449 
450 	flush = 0;
451 	mmap_write_lock(gmap->mm);
452 	for (off = 0; off < len; off += PMD_SIZE) {
453 		/* Remove old translation */
454 		flush |= __gmap_unmap_by_gaddr(gmap, to + off);
455 		/* Store new translation */
456 		if (radix_tree_insert(&gmap->guest_to_host,
457 				      (to + off) >> PMD_SHIFT,
458 				      (void *) from + off))
459 			break;
460 	}
461 	mmap_write_unlock(gmap->mm);
462 	if (flush)
463 		gmap_flush_tlb(gmap);
464 	if (off >= len)
465 		return 0;
466 	gmap_unmap_segment(gmap, to, len);
467 	return -ENOMEM;
468 }
469 EXPORT_SYMBOL_GPL(gmap_map_segment);
470 
471 /**
472  * __gmap_translate - translate a guest address to a user space address
473  * @gmap: pointer to guest mapping meta data structure
474  * @gaddr: guest address
475  *
476  * Returns user space address which corresponds to the guest address or
477  * -EFAULT if no such mapping exists.
478  * This function does not establish potentially missing page table entries.
479  * The mmap_lock of the mm that belongs to the address space must be held
480  * when this function gets called.
481  *
482  * Note: Can also be called for shadow gmaps.
483  */
__gmap_translate(struct gmap * gmap,unsigned long gaddr)484 unsigned long __gmap_translate(struct gmap *gmap, unsigned long gaddr)
485 {
486 	unsigned long vmaddr;
487 
488 	vmaddr = (unsigned long)
489 		radix_tree_lookup(&gmap->guest_to_host, gaddr >> PMD_SHIFT);
490 	/* Note: guest_to_host is empty for a shadow gmap */
491 	return vmaddr ? (vmaddr | (gaddr & ~PMD_MASK)) : -EFAULT;
492 }
493 EXPORT_SYMBOL_GPL(__gmap_translate);
494 
495 /**
496  * gmap_translate - translate a guest address to a user space address
497  * @gmap: pointer to guest mapping meta data structure
498  * @gaddr: guest address
499  *
500  * Returns user space address which corresponds to the guest address or
501  * -EFAULT if no such mapping exists.
502  * This function does not establish potentially missing page table entries.
503  */
gmap_translate(struct gmap * gmap,unsigned long gaddr)504 unsigned long gmap_translate(struct gmap *gmap, unsigned long gaddr)
505 {
506 	unsigned long rc;
507 
508 	mmap_read_lock(gmap->mm);
509 	rc = __gmap_translate(gmap, gaddr);
510 	mmap_read_unlock(gmap->mm);
511 	return rc;
512 }
513 EXPORT_SYMBOL_GPL(gmap_translate);
514 
515 /**
516  * gmap_unlink - disconnect a page table from the gmap shadow tables
517  * @mm: pointer to the parent mm_struct
518  * @table: pointer to the host page table
519  * @vmaddr: vm address associated with the host page table
520  */
gmap_unlink(struct mm_struct * mm,unsigned long * table,unsigned long vmaddr)521 void gmap_unlink(struct mm_struct *mm, unsigned long *table,
522 		 unsigned long vmaddr)
523 {
524 	struct gmap *gmap;
525 	int flush;
526 
527 	rcu_read_lock();
528 	list_for_each_entry_rcu(gmap, &mm->context.gmap_list, list) {
529 		flush = __gmap_unlink_by_vmaddr(gmap, vmaddr);
530 		if (flush)
531 			gmap_flush_tlb(gmap);
532 	}
533 	rcu_read_unlock();
534 }
535 
536 static void gmap_pmdp_xchg(struct gmap *gmap, pmd_t *old, pmd_t new,
537 			   unsigned long gaddr);
538 
539 /**
540  * __gmap_link - set up shadow page tables to connect a host to a guest address
541  * @gmap: pointer to guest mapping meta data structure
542  * @gaddr: guest address
543  * @vmaddr: vm address
544  *
545  * Returns 0 on success, -ENOMEM for out of memory conditions, and -EFAULT
546  * if the vm address is already mapped to a different guest segment.
547  * The mmap_lock of the mm that belongs to the address space must be held
548  * when this function gets called.
549  */
__gmap_link(struct gmap * gmap,unsigned long gaddr,unsigned long vmaddr)550 int __gmap_link(struct gmap *gmap, unsigned long gaddr, unsigned long vmaddr)
551 {
552 	struct mm_struct *mm;
553 	unsigned long *table;
554 	spinlock_t *ptl;
555 	pgd_t *pgd;
556 	p4d_t *p4d;
557 	pud_t *pud;
558 	pmd_t *pmd;
559 	u64 unprot;
560 	int rc;
561 
562 	BUG_ON(gmap_is_shadow(gmap));
563 	/* Create higher level tables in the gmap page table */
564 	table = gmap->table;
565 	if ((gmap->asce & _ASCE_TYPE_MASK) >= _ASCE_TYPE_REGION1) {
566 		table += (gaddr & _REGION1_INDEX) >> _REGION1_SHIFT;
567 		if ((*table & _REGION_ENTRY_INVALID) &&
568 		    gmap_alloc_table(gmap, table, _REGION2_ENTRY_EMPTY,
569 				     gaddr & _REGION1_MASK))
570 			return -ENOMEM;
571 		table = __va(*table & _REGION_ENTRY_ORIGIN);
572 	}
573 	if ((gmap->asce & _ASCE_TYPE_MASK) >= _ASCE_TYPE_REGION2) {
574 		table += (gaddr & _REGION2_INDEX) >> _REGION2_SHIFT;
575 		if ((*table & _REGION_ENTRY_INVALID) &&
576 		    gmap_alloc_table(gmap, table, _REGION3_ENTRY_EMPTY,
577 				     gaddr & _REGION2_MASK))
578 			return -ENOMEM;
579 		table = __va(*table & _REGION_ENTRY_ORIGIN);
580 	}
581 	if ((gmap->asce & _ASCE_TYPE_MASK) >= _ASCE_TYPE_REGION3) {
582 		table += (gaddr & _REGION3_INDEX) >> _REGION3_SHIFT;
583 		if ((*table & _REGION_ENTRY_INVALID) &&
584 		    gmap_alloc_table(gmap, table, _SEGMENT_ENTRY_EMPTY,
585 				     gaddr & _REGION3_MASK))
586 			return -ENOMEM;
587 		table = __va(*table & _REGION_ENTRY_ORIGIN);
588 	}
589 	table += (gaddr & _SEGMENT_INDEX) >> _SEGMENT_SHIFT;
590 	/* Walk the parent mm page table */
591 	mm = gmap->mm;
592 	pgd = pgd_offset(mm, vmaddr);
593 	VM_BUG_ON(pgd_none(*pgd));
594 	p4d = p4d_offset(pgd, vmaddr);
595 	VM_BUG_ON(p4d_none(*p4d));
596 	pud = pud_offset(p4d, vmaddr);
597 	VM_BUG_ON(pud_none(*pud));
598 	/* large puds cannot yet be handled */
599 	if (pud_large(*pud))
600 		return -EFAULT;
601 	pmd = pmd_offset(pud, vmaddr);
602 	VM_BUG_ON(pmd_none(*pmd));
603 	/* Are we allowed to use huge pages? */
604 	if (pmd_large(*pmd) && !gmap->mm->context.allow_gmap_hpage_1m)
605 		return -EFAULT;
606 	/* Link gmap segment table entry location to page table. */
607 	rc = radix_tree_preload(GFP_KERNEL_ACCOUNT);
608 	if (rc)
609 		return rc;
610 	ptl = pmd_lock(mm, pmd);
611 	spin_lock(&gmap->guest_table_lock);
612 	if (*table == _SEGMENT_ENTRY_EMPTY) {
613 		rc = radix_tree_insert(&gmap->host_to_guest,
614 				       vmaddr >> PMD_SHIFT, table);
615 		if (!rc) {
616 			if (pmd_large(*pmd)) {
617 				*table = (pmd_val(*pmd) &
618 					  _SEGMENT_ENTRY_HARDWARE_BITS_LARGE)
619 					| _SEGMENT_ENTRY_GMAP_UC;
620 			} else
621 				*table = pmd_val(*pmd) &
622 					_SEGMENT_ENTRY_HARDWARE_BITS;
623 		}
624 	} else if (*table & _SEGMENT_ENTRY_PROTECT &&
625 		   !(pmd_val(*pmd) & _SEGMENT_ENTRY_PROTECT)) {
626 		unprot = (u64)*table;
627 		unprot &= ~_SEGMENT_ENTRY_PROTECT;
628 		unprot |= _SEGMENT_ENTRY_GMAP_UC;
629 		gmap_pmdp_xchg(gmap, (pmd_t *)table, __pmd(unprot), gaddr);
630 	}
631 	spin_unlock(&gmap->guest_table_lock);
632 	spin_unlock(ptl);
633 	radix_tree_preload_end();
634 	return rc;
635 }
636 
637 /**
638  * gmap_fault - resolve a fault on a guest address
639  * @gmap: pointer to guest mapping meta data structure
640  * @gaddr: guest address
641  * @fault_flags: flags to pass down to handle_mm_fault()
642  *
643  * Returns 0 on success, -ENOMEM for out of memory conditions, and -EFAULT
644  * if the vm address is already mapped to a different guest segment.
645  */
gmap_fault(struct gmap * gmap,unsigned long gaddr,unsigned int fault_flags)646 int gmap_fault(struct gmap *gmap, unsigned long gaddr,
647 	       unsigned int fault_flags)
648 {
649 	unsigned long vmaddr;
650 	int rc;
651 	bool unlocked;
652 
653 	mmap_read_lock(gmap->mm);
654 
655 retry:
656 	unlocked = false;
657 	vmaddr = __gmap_translate(gmap, gaddr);
658 	if (IS_ERR_VALUE(vmaddr)) {
659 		rc = vmaddr;
660 		goto out_up;
661 	}
662 	if (fixup_user_fault(gmap->mm, vmaddr, fault_flags,
663 			     &unlocked)) {
664 		rc = -EFAULT;
665 		goto out_up;
666 	}
667 	/*
668 	 * In the case that fixup_user_fault unlocked the mmap_lock during
669 	 * faultin redo __gmap_translate to not race with a map/unmap_segment.
670 	 */
671 	if (unlocked)
672 		goto retry;
673 
674 	rc = __gmap_link(gmap, gaddr, vmaddr);
675 out_up:
676 	mmap_read_unlock(gmap->mm);
677 	return rc;
678 }
679 EXPORT_SYMBOL_GPL(gmap_fault);
680 
681 /*
682  * this function is assumed to be called with mmap_lock held
683  */
__gmap_zap(struct gmap * gmap,unsigned long gaddr)684 void __gmap_zap(struct gmap *gmap, unsigned long gaddr)
685 {
686 	struct vm_area_struct *vma;
687 	unsigned long vmaddr;
688 	spinlock_t *ptl;
689 	pte_t *ptep;
690 
691 	/* Find the vm address for the guest address */
692 	vmaddr = (unsigned long) radix_tree_lookup(&gmap->guest_to_host,
693 						   gaddr >> PMD_SHIFT);
694 	if (vmaddr) {
695 		vmaddr |= gaddr & ~PMD_MASK;
696 
697 		vma = vma_lookup(gmap->mm, vmaddr);
698 		if (!vma || is_vm_hugetlb_page(vma))
699 			return;
700 
701 		/* Get pointer to the page table entry */
702 		ptep = get_locked_pte(gmap->mm, vmaddr, &ptl);
703 		if (likely(ptep)) {
704 			ptep_zap_unused(gmap->mm, vmaddr, ptep, 0);
705 			pte_unmap_unlock(ptep, ptl);
706 		}
707 	}
708 }
709 EXPORT_SYMBOL_GPL(__gmap_zap);
710 
gmap_discard(struct gmap * gmap,unsigned long from,unsigned long to)711 void gmap_discard(struct gmap *gmap, unsigned long from, unsigned long to)
712 {
713 	unsigned long gaddr, vmaddr, size;
714 	struct vm_area_struct *vma;
715 
716 	mmap_read_lock(gmap->mm);
717 	for (gaddr = from; gaddr < to;
718 	     gaddr = (gaddr + PMD_SIZE) & PMD_MASK) {
719 		/* Find the vm address for the guest address */
720 		vmaddr = (unsigned long)
721 			radix_tree_lookup(&gmap->guest_to_host,
722 					  gaddr >> PMD_SHIFT);
723 		if (!vmaddr)
724 			continue;
725 		vmaddr |= gaddr & ~PMD_MASK;
726 		/* Find vma in the parent mm */
727 		vma = find_vma(gmap->mm, vmaddr);
728 		if (!vma)
729 			continue;
730 		/*
731 		 * We do not discard pages that are backed by
732 		 * hugetlbfs, so we don't have to refault them.
733 		 */
734 		if (is_vm_hugetlb_page(vma))
735 			continue;
736 		size = min(to - gaddr, PMD_SIZE - (gaddr & ~PMD_MASK));
737 		zap_page_range_single(vma, vmaddr, size, NULL);
738 	}
739 	mmap_read_unlock(gmap->mm);
740 }
741 EXPORT_SYMBOL_GPL(gmap_discard);
742 
743 static LIST_HEAD(gmap_notifier_list);
744 static DEFINE_SPINLOCK(gmap_notifier_lock);
745 
746 /**
747  * gmap_register_pte_notifier - register a pte invalidation callback
748  * @nb: pointer to the gmap notifier block
749  */
gmap_register_pte_notifier(struct gmap_notifier * nb)750 void gmap_register_pte_notifier(struct gmap_notifier *nb)
751 {
752 	spin_lock(&gmap_notifier_lock);
753 	list_add_rcu(&nb->list, &gmap_notifier_list);
754 	spin_unlock(&gmap_notifier_lock);
755 }
756 EXPORT_SYMBOL_GPL(gmap_register_pte_notifier);
757 
758 /**
759  * gmap_unregister_pte_notifier - remove a pte invalidation callback
760  * @nb: pointer to the gmap notifier block
761  */
gmap_unregister_pte_notifier(struct gmap_notifier * nb)762 void gmap_unregister_pte_notifier(struct gmap_notifier *nb)
763 {
764 	spin_lock(&gmap_notifier_lock);
765 	list_del_rcu(&nb->list);
766 	spin_unlock(&gmap_notifier_lock);
767 	synchronize_rcu();
768 }
769 EXPORT_SYMBOL_GPL(gmap_unregister_pte_notifier);
770 
771 /**
772  * gmap_call_notifier - call all registered invalidation callbacks
773  * @gmap: pointer to guest mapping meta data structure
774  * @start: start virtual address in the guest address space
775  * @end: end virtual address in the guest address space
776  */
gmap_call_notifier(struct gmap * gmap,unsigned long start,unsigned long end)777 static void gmap_call_notifier(struct gmap *gmap, unsigned long start,
778 			       unsigned long end)
779 {
780 	struct gmap_notifier *nb;
781 
782 	list_for_each_entry(nb, &gmap_notifier_list, list)
783 		nb->notifier_call(gmap, start, end);
784 }
785 
786 /**
787  * gmap_table_walk - walk the gmap page tables
788  * @gmap: pointer to guest mapping meta data structure
789  * @gaddr: virtual address in the guest address space
790  * @level: page table level to stop at
791  *
792  * Returns a table entry pointer for the given guest address and @level
793  * @level=0 : returns a pointer to a page table table entry (or NULL)
794  * @level=1 : returns a pointer to a segment table entry (or NULL)
795  * @level=2 : returns a pointer to a region-3 table entry (or NULL)
796  * @level=3 : returns a pointer to a region-2 table entry (or NULL)
797  * @level=4 : returns a pointer to a region-1 table entry (or NULL)
798  *
799  * Returns NULL if the gmap page tables could not be walked to the
800  * requested level.
801  *
802  * Note: Can also be called for shadow gmaps.
803  */
gmap_table_walk(struct gmap * gmap,unsigned long gaddr,int level)804 static inline unsigned long *gmap_table_walk(struct gmap *gmap,
805 					     unsigned long gaddr, int level)
806 {
807 	const int asce_type = gmap->asce & _ASCE_TYPE_MASK;
808 	unsigned long *table = gmap->table;
809 
810 	if (gmap_is_shadow(gmap) && gmap->removed)
811 		return NULL;
812 
813 	if (WARN_ON_ONCE(level > (asce_type >> 2) + 1))
814 		return NULL;
815 
816 	if (asce_type != _ASCE_TYPE_REGION1 &&
817 	    gaddr & (-1UL << (31 + (asce_type >> 2) * 11)))
818 		return NULL;
819 
820 	switch (asce_type) {
821 	case _ASCE_TYPE_REGION1:
822 		table += (gaddr & _REGION1_INDEX) >> _REGION1_SHIFT;
823 		if (level == 4)
824 			break;
825 		if (*table & _REGION_ENTRY_INVALID)
826 			return NULL;
827 		table = __va(*table & _REGION_ENTRY_ORIGIN);
828 		fallthrough;
829 	case _ASCE_TYPE_REGION2:
830 		table += (gaddr & _REGION2_INDEX) >> _REGION2_SHIFT;
831 		if (level == 3)
832 			break;
833 		if (*table & _REGION_ENTRY_INVALID)
834 			return NULL;
835 		table = __va(*table & _REGION_ENTRY_ORIGIN);
836 		fallthrough;
837 	case _ASCE_TYPE_REGION3:
838 		table += (gaddr & _REGION3_INDEX) >> _REGION3_SHIFT;
839 		if (level == 2)
840 			break;
841 		if (*table & _REGION_ENTRY_INVALID)
842 			return NULL;
843 		table = __va(*table & _REGION_ENTRY_ORIGIN);
844 		fallthrough;
845 	case _ASCE_TYPE_SEGMENT:
846 		table += (gaddr & _SEGMENT_INDEX) >> _SEGMENT_SHIFT;
847 		if (level == 1)
848 			break;
849 		if (*table & _REGION_ENTRY_INVALID)
850 			return NULL;
851 		table = __va(*table & _SEGMENT_ENTRY_ORIGIN);
852 		table += (gaddr & _PAGE_INDEX) >> _PAGE_SHIFT;
853 	}
854 	return table;
855 }
856 
857 /**
858  * gmap_pte_op_walk - walk the gmap page table, get the page table lock
859  *		      and return the pte pointer
860  * @gmap: pointer to guest mapping meta data structure
861  * @gaddr: virtual address in the guest address space
862  * @ptl: pointer to the spinlock pointer
863  *
864  * Returns a pointer to the locked pte for a guest address, or NULL
865  */
gmap_pte_op_walk(struct gmap * gmap,unsigned long gaddr,spinlock_t ** ptl)866 static pte_t *gmap_pte_op_walk(struct gmap *gmap, unsigned long gaddr,
867 			       spinlock_t **ptl)
868 {
869 	unsigned long *table;
870 
871 	BUG_ON(gmap_is_shadow(gmap));
872 	/* Walk the gmap page table, lock and get pte pointer */
873 	table = gmap_table_walk(gmap, gaddr, 1); /* get segment pointer */
874 	if (!table || *table & _SEGMENT_ENTRY_INVALID)
875 		return NULL;
876 	return pte_alloc_map_lock(gmap->mm, (pmd_t *) table, gaddr, ptl);
877 }
878 
879 /**
880  * gmap_pte_op_fixup - force a page in and connect the gmap page table
881  * @gmap: pointer to guest mapping meta data structure
882  * @gaddr: virtual address in the guest address space
883  * @vmaddr: address in the host process address space
884  * @prot: indicates access rights: PROT_NONE, PROT_READ or PROT_WRITE
885  *
886  * Returns 0 if the caller can retry __gmap_translate (might fail again),
887  * -ENOMEM if out of memory and -EFAULT if anything goes wrong while fixing
888  * up or connecting the gmap page table.
889  */
gmap_pte_op_fixup(struct gmap * gmap,unsigned long gaddr,unsigned long vmaddr,int prot)890 static int gmap_pte_op_fixup(struct gmap *gmap, unsigned long gaddr,
891 			     unsigned long vmaddr, int prot)
892 {
893 	struct mm_struct *mm = gmap->mm;
894 	unsigned int fault_flags;
895 	bool unlocked = false;
896 
897 	BUG_ON(gmap_is_shadow(gmap));
898 	fault_flags = (prot == PROT_WRITE) ? FAULT_FLAG_WRITE : 0;
899 	if (fixup_user_fault(mm, vmaddr, fault_flags, &unlocked))
900 		return -EFAULT;
901 	if (unlocked)
902 		/* lost mmap_lock, caller has to retry __gmap_translate */
903 		return 0;
904 	/* Connect the page tables */
905 	return __gmap_link(gmap, gaddr, vmaddr);
906 }
907 
908 /**
909  * gmap_pte_op_end - release the page table lock
910  * @ptep: pointer to the locked pte
911  * @ptl: pointer to the page table spinlock
912  */
gmap_pte_op_end(pte_t * ptep,spinlock_t * ptl)913 static void gmap_pte_op_end(pte_t *ptep, spinlock_t *ptl)
914 {
915 	pte_unmap_unlock(ptep, ptl);
916 }
917 
918 /**
919  * gmap_pmd_op_walk - walk the gmap tables, get the guest table lock
920  *		      and return the pmd pointer
921  * @gmap: pointer to guest mapping meta data structure
922  * @gaddr: virtual address in the guest address space
923  *
924  * Returns a pointer to the pmd for a guest address, or NULL
925  */
gmap_pmd_op_walk(struct gmap * gmap,unsigned long gaddr)926 static inline pmd_t *gmap_pmd_op_walk(struct gmap *gmap, unsigned long gaddr)
927 {
928 	pmd_t *pmdp;
929 
930 	BUG_ON(gmap_is_shadow(gmap));
931 	pmdp = (pmd_t *) gmap_table_walk(gmap, gaddr, 1);
932 	if (!pmdp)
933 		return NULL;
934 
935 	/* without huge pages, there is no need to take the table lock */
936 	if (!gmap->mm->context.allow_gmap_hpage_1m)
937 		return pmd_none(*pmdp) ? NULL : pmdp;
938 
939 	spin_lock(&gmap->guest_table_lock);
940 	if (pmd_none(*pmdp)) {
941 		spin_unlock(&gmap->guest_table_lock);
942 		return NULL;
943 	}
944 
945 	/* 4k page table entries are locked via the pte (pte_alloc_map_lock). */
946 	if (!pmd_large(*pmdp))
947 		spin_unlock(&gmap->guest_table_lock);
948 	return pmdp;
949 }
950 
951 /**
952  * gmap_pmd_op_end - release the guest_table_lock if needed
953  * @gmap: pointer to the guest mapping meta data structure
954  * @pmdp: pointer to the pmd
955  */
gmap_pmd_op_end(struct gmap * gmap,pmd_t * pmdp)956 static inline void gmap_pmd_op_end(struct gmap *gmap, pmd_t *pmdp)
957 {
958 	if (pmd_large(*pmdp))
959 		spin_unlock(&gmap->guest_table_lock);
960 }
961 
962 /*
963  * gmap_protect_pmd - remove access rights to memory and set pmd notification bits
964  * @pmdp: pointer to the pmd to be protected
965  * @prot: indicates access rights: PROT_NONE, PROT_READ or PROT_WRITE
966  * @bits: notification bits to set
967  *
968  * Returns:
969  * 0 if successfully protected
970  * -EAGAIN if a fixup is needed
971  * -EINVAL if unsupported notifier bits have been specified
972  *
973  * Expected to be called with sg->mm->mmap_lock in read and
974  * guest_table_lock held.
975  */
gmap_protect_pmd(struct gmap * gmap,unsigned long gaddr,pmd_t * pmdp,int prot,unsigned long bits)976 static int gmap_protect_pmd(struct gmap *gmap, unsigned long gaddr,
977 			    pmd_t *pmdp, int prot, unsigned long bits)
978 {
979 	int pmd_i = pmd_val(*pmdp) & _SEGMENT_ENTRY_INVALID;
980 	int pmd_p = pmd_val(*pmdp) & _SEGMENT_ENTRY_PROTECT;
981 	pmd_t new = *pmdp;
982 
983 	/* Fixup needed */
984 	if ((pmd_i && (prot != PROT_NONE)) || (pmd_p && (prot == PROT_WRITE)))
985 		return -EAGAIN;
986 
987 	if (prot == PROT_NONE && !pmd_i) {
988 		new = set_pmd_bit(new, __pgprot(_SEGMENT_ENTRY_INVALID));
989 		gmap_pmdp_xchg(gmap, pmdp, new, gaddr);
990 	}
991 
992 	if (prot == PROT_READ && !pmd_p) {
993 		new = clear_pmd_bit(new, __pgprot(_SEGMENT_ENTRY_INVALID));
994 		new = set_pmd_bit(new, __pgprot(_SEGMENT_ENTRY_PROTECT));
995 		gmap_pmdp_xchg(gmap, pmdp, new, gaddr);
996 	}
997 
998 	if (bits & GMAP_NOTIFY_MPROT)
999 		set_pmd(pmdp, set_pmd_bit(*pmdp, __pgprot(_SEGMENT_ENTRY_GMAP_IN)));
1000 
1001 	/* Shadow GMAP protection needs split PMDs */
1002 	if (bits & GMAP_NOTIFY_SHADOW)
1003 		return -EINVAL;
1004 
1005 	return 0;
1006 }
1007 
1008 /*
1009  * gmap_protect_pte - remove access rights to memory and set pgste bits
1010  * @gmap: pointer to guest mapping meta data structure
1011  * @gaddr: virtual address in the guest address space
1012  * @pmdp: pointer to the pmd associated with the pte
1013  * @prot: indicates access rights: PROT_NONE, PROT_READ or PROT_WRITE
1014  * @bits: notification bits to set
1015  *
1016  * Returns 0 if successfully protected, -ENOMEM if out of memory and
1017  * -EAGAIN if a fixup is needed.
1018  *
1019  * Expected to be called with sg->mm->mmap_lock in read
1020  */
gmap_protect_pte(struct gmap * gmap,unsigned long gaddr,pmd_t * pmdp,int prot,unsigned long bits)1021 static int gmap_protect_pte(struct gmap *gmap, unsigned long gaddr,
1022 			    pmd_t *pmdp, int prot, unsigned long bits)
1023 {
1024 	int rc;
1025 	pte_t *ptep;
1026 	spinlock_t *ptl;
1027 	unsigned long pbits = 0;
1028 
1029 	if (pmd_val(*pmdp) & _SEGMENT_ENTRY_INVALID)
1030 		return -EAGAIN;
1031 
1032 	ptep = pte_alloc_map_lock(gmap->mm, pmdp, gaddr, &ptl);
1033 	if (!ptep)
1034 		return -ENOMEM;
1035 
1036 	pbits |= (bits & GMAP_NOTIFY_MPROT) ? PGSTE_IN_BIT : 0;
1037 	pbits |= (bits & GMAP_NOTIFY_SHADOW) ? PGSTE_VSIE_BIT : 0;
1038 	/* Protect and unlock. */
1039 	rc = ptep_force_prot(gmap->mm, gaddr, ptep, prot, pbits);
1040 	gmap_pte_op_end(ptep, ptl);
1041 	return rc;
1042 }
1043 
1044 /*
1045  * gmap_protect_range - remove access rights to memory and set pgste bits
1046  * @gmap: pointer to guest mapping meta data structure
1047  * @gaddr: virtual address in the guest address space
1048  * @len: size of area
1049  * @prot: indicates access rights: PROT_NONE, PROT_READ or PROT_WRITE
1050  * @bits: pgste notification bits to set
1051  *
1052  * Returns 0 if successfully protected, -ENOMEM if out of memory and
1053  * -EFAULT if gaddr is invalid (or mapping for shadows is missing).
1054  *
1055  * Called with sg->mm->mmap_lock in read.
1056  */
gmap_protect_range(struct gmap * gmap,unsigned long gaddr,unsigned long len,int prot,unsigned long bits)1057 static int gmap_protect_range(struct gmap *gmap, unsigned long gaddr,
1058 			      unsigned long len, int prot, unsigned long bits)
1059 {
1060 	unsigned long vmaddr, dist;
1061 	pmd_t *pmdp;
1062 	int rc;
1063 
1064 	BUG_ON(gmap_is_shadow(gmap));
1065 	while (len) {
1066 		rc = -EAGAIN;
1067 		pmdp = gmap_pmd_op_walk(gmap, gaddr);
1068 		if (pmdp) {
1069 			if (!pmd_large(*pmdp)) {
1070 				rc = gmap_protect_pte(gmap, gaddr, pmdp, prot,
1071 						      bits);
1072 				if (!rc) {
1073 					len -= PAGE_SIZE;
1074 					gaddr += PAGE_SIZE;
1075 				}
1076 			} else {
1077 				rc = gmap_protect_pmd(gmap, gaddr, pmdp, prot,
1078 						      bits);
1079 				if (!rc) {
1080 					dist = HPAGE_SIZE - (gaddr & ~HPAGE_MASK);
1081 					len = len < dist ? 0 : len - dist;
1082 					gaddr = (gaddr & HPAGE_MASK) + HPAGE_SIZE;
1083 				}
1084 			}
1085 			gmap_pmd_op_end(gmap, pmdp);
1086 		}
1087 		if (rc) {
1088 			if (rc == -EINVAL)
1089 				return rc;
1090 
1091 			/* -EAGAIN, fixup of userspace mm and gmap */
1092 			vmaddr = __gmap_translate(gmap, gaddr);
1093 			if (IS_ERR_VALUE(vmaddr))
1094 				return vmaddr;
1095 			rc = gmap_pte_op_fixup(gmap, gaddr, vmaddr, prot);
1096 			if (rc)
1097 				return rc;
1098 		}
1099 	}
1100 	return 0;
1101 }
1102 
1103 /**
1104  * gmap_mprotect_notify - change access rights for a range of ptes and
1105  *                        call the notifier if any pte changes again
1106  * @gmap: pointer to guest mapping meta data structure
1107  * @gaddr: virtual address in the guest address space
1108  * @len: size of area
1109  * @prot: indicates access rights: PROT_NONE, PROT_READ or PROT_WRITE
1110  *
1111  * Returns 0 if for each page in the given range a gmap mapping exists,
1112  * the new access rights could be set and the notifier could be armed.
1113  * If the gmap mapping is missing for one or more pages -EFAULT is
1114  * returned. If no memory could be allocated -ENOMEM is returned.
1115  * This function establishes missing page table entries.
1116  */
gmap_mprotect_notify(struct gmap * gmap,unsigned long gaddr,unsigned long len,int prot)1117 int gmap_mprotect_notify(struct gmap *gmap, unsigned long gaddr,
1118 			 unsigned long len, int prot)
1119 {
1120 	int rc;
1121 
1122 	if ((gaddr & ~PAGE_MASK) || (len & ~PAGE_MASK) || gmap_is_shadow(gmap))
1123 		return -EINVAL;
1124 	if (!MACHINE_HAS_ESOP && prot == PROT_READ)
1125 		return -EINVAL;
1126 	mmap_read_lock(gmap->mm);
1127 	rc = gmap_protect_range(gmap, gaddr, len, prot, GMAP_NOTIFY_MPROT);
1128 	mmap_read_unlock(gmap->mm);
1129 	return rc;
1130 }
1131 EXPORT_SYMBOL_GPL(gmap_mprotect_notify);
1132 
1133 /**
1134  * gmap_read_table - get an unsigned long value from a guest page table using
1135  *                   absolute addressing, without marking the page referenced.
1136  * @gmap: pointer to guest mapping meta data structure
1137  * @gaddr: virtual address in the guest address space
1138  * @val: pointer to the unsigned long value to return
1139  *
1140  * Returns 0 if the value was read, -ENOMEM if out of memory and -EFAULT
1141  * if reading using the virtual address failed. -EINVAL if called on a gmap
1142  * shadow.
1143  *
1144  * Called with gmap->mm->mmap_lock in read.
1145  */
gmap_read_table(struct gmap * gmap,unsigned long gaddr,unsigned long * val)1146 int gmap_read_table(struct gmap *gmap, unsigned long gaddr, unsigned long *val)
1147 {
1148 	unsigned long address, vmaddr;
1149 	spinlock_t *ptl;
1150 	pte_t *ptep, pte;
1151 	int rc;
1152 
1153 	if (gmap_is_shadow(gmap))
1154 		return -EINVAL;
1155 
1156 	while (1) {
1157 		rc = -EAGAIN;
1158 		ptep = gmap_pte_op_walk(gmap, gaddr, &ptl);
1159 		if (ptep) {
1160 			pte = *ptep;
1161 			if (pte_present(pte) && (pte_val(pte) & _PAGE_READ)) {
1162 				address = pte_val(pte) & PAGE_MASK;
1163 				address += gaddr & ~PAGE_MASK;
1164 				*val = *(unsigned long *)__va(address);
1165 				set_pte(ptep, set_pte_bit(*ptep, __pgprot(_PAGE_YOUNG)));
1166 				/* Do *NOT* clear the _PAGE_INVALID bit! */
1167 				rc = 0;
1168 			}
1169 			gmap_pte_op_end(ptep, ptl);
1170 		}
1171 		if (!rc)
1172 			break;
1173 		vmaddr = __gmap_translate(gmap, gaddr);
1174 		if (IS_ERR_VALUE(vmaddr)) {
1175 			rc = vmaddr;
1176 			break;
1177 		}
1178 		rc = gmap_pte_op_fixup(gmap, gaddr, vmaddr, PROT_READ);
1179 		if (rc)
1180 			break;
1181 	}
1182 	return rc;
1183 }
1184 EXPORT_SYMBOL_GPL(gmap_read_table);
1185 
1186 /**
1187  * gmap_insert_rmap - add a rmap to the host_to_rmap radix tree
1188  * @sg: pointer to the shadow guest address space structure
1189  * @vmaddr: vm address associated with the rmap
1190  * @rmap: pointer to the rmap structure
1191  *
1192  * Called with the sg->guest_table_lock
1193  */
gmap_insert_rmap(struct gmap * sg,unsigned long vmaddr,struct gmap_rmap * rmap)1194 static inline void gmap_insert_rmap(struct gmap *sg, unsigned long vmaddr,
1195 				    struct gmap_rmap *rmap)
1196 {
1197 	struct gmap_rmap *temp;
1198 	void __rcu **slot;
1199 
1200 	BUG_ON(!gmap_is_shadow(sg));
1201 	slot = radix_tree_lookup_slot(&sg->host_to_rmap, vmaddr >> PAGE_SHIFT);
1202 	if (slot) {
1203 		rmap->next = radix_tree_deref_slot_protected(slot,
1204 							&sg->guest_table_lock);
1205 		for (temp = rmap->next; temp; temp = temp->next) {
1206 			if (temp->raddr == rmap->raddr) {
1207 				kfree(rmap);
1208 				return;
1209 			}
1210 		}
1211 		radix_tree_replace_slot(&sg->host_to_rmap, slot, rmap);
1212 	} else {
1213 		rmap->next = NULL;
1214 		radix_tree_insert(&sg->host_to_rmap, vmaddr >> PAGE_SHIFT,
1215 				  rmap);
1216 	}
1217 }
1218 
1219 /**
1220  * gmap_protect_rmap - restrict access rights to memory (RO) and create an rmap
1221  * @sg: pointer to the shadow guest address space structure
1222  * @raddr: rmap address in the shadow gmap
1223  * @paddr: address in the parent guest address space
1224  * @len: length of the memory area to protect
1225  *
1226  * Returns 0 if successfully protected and the rmap was created, -ENOMEM
1227  * if out of memory and -EFAULT if paddr is invalid.
1228  */
gmap_protect_rmap(struct gmap * sg,unsigned long raddr,unsigned long paddr,unsigned long len)1229 static int gmap_protect_rmap(struct gmap *sg, unsigned long raddr,
1230 			     unsigned long paddr, unsigned long len)
1231 {
1232 	struct gmap *parent;
1233 	struct gmap_rmap *rmap;
1234 	unsigned long vmaddr;
1235 	spinlock_t *ptl;
1236 	pte_t *ptep;
1237 	int rc;
1238 
1239 	BUG_ON(!gmap_is_shadow(sg));
1240 	parent = sg->parent;
1241 	while (len) {
1242 		vmaddr = __gmap_translate(parent, paddr);
1243 		if (IS_ERR_VALUE(vmaddr))
1244 			return vmaddr;
1245 		rmap = kzalloc(sizeof(*rmap), GFP_KERNEL_ACCOUNT);
1246 		if (!rmap)
1247 			return -ENOMEM;
1248 		rmap->raddr = raddr;
1249 		rc = radix_tree_preload(GFP_KERNEL_ACCOUNT);
1250 		if (rc) {
1251 			kfree(rmap);
1252 			return rc;
1253 		}
1254 		rc = -EAGAIN;
1255 		ptep = gmap_pte_op_walk(parent, paddr, &ptl);
1256 		if (ptep) {
1257 			spin_lock(&sg->guest_table_lock);
1258 			rc = ptep_force_prot(parent->mm, paddr, ptep, PROT_READ,
1259 					     PGSTE_VSIE_BIT);
1260 			if (!rc)
1261 				gmap_insert_rmap(sg, vmaddr, rmap);
1262 			spin_unlock(&sg->guest_table_lock);
1263 			gmap_pte_op_end(ptep, ptl);
1264 		}
1265 		radix_tree_preload_end();
1266 		if (rc) {
1267 			kfree(rmap);
1268 			rc = gmap_pte_op_fixup(parent, paddr, vmaddr, PROT_READ);
1269 			if (rc)
1270 				return rc;
1271 			continue;
1272 		}
1273 		paddr += PAGE_SIZE;
1274 		len -= PAGE_SIZE;
1275 	}
1276 	return 0;
1277 }
1278 
1279 #define _SHADOW_RMAP_MASK	0x7
1280 #define _SHADOW_RMAP_REGION1	0x5
1281 #define _SHADOW_RMAP_REGION2	0x4
1282 #define _SHADOW_RMAP_REGION3	0x3
1283 #define _SHADOW_RMAP_SEGMENT	0x2
1284 #define _SHADOW_RMAP_PGTABLE	0x1
1285 
1286 /**
1287  * gmap_idte_one - invalidate a single region or segment table entry
1288  * @asce: region or segment table *origin* + table-type bits
1289  * @vaddr: virtual address to identify the table entry to flush
1290  *
1291  * The invalid bit of a single region or segment table entry is set
1292  * and the associated TLB entries depending on the entry are flushed.
1293  * The table-type of the @asce identifies the portion of the @vaddr
1294  * that is used as the invalidation index.
1295  */
gmap_idte_one(unsigned long asce,unsigned long vaddr)1296 static inline void gmap_idte_one(unsigned long asce, unsigned long vaddr)
1297 {
1298 	asm volatile(
1299 		"	idte	%0,0,%1"
1300 		: : "a" (asce), "a" (vaddr) : "cc", "memory");
1301 }
1302 
1303 /**
1304  * gmap_unshadow_page - remove a page from a shadow page table
1305  * @sg: pointer to the shadow guest address space structure
1306  * @raddr: rmap address in the shadow guest address space
1307  *
1308  * Called with the sg->guest_table_lock
1309  */
gmap_unshadow_page(struct gmap * sg,unsigned long raddr)1310 static void gmap_unshadow_page(struct gmap *sg, unsigned long raddr)
1311 {
1312 	unsigned long *table;
1313 
1314 	BUG_ON(!gmap_is_shadow(sg));
1315 	table = gmap_table_walk(sg, raddr, 0); /* get page table pointer */
1316 	if (!table || *table & _PAGE_INVALID)
1317 		return;
1318 	gmap_call_notifier(sg, raddr, raddr + _PAGE_SIZE - 1);
1319 	ptep_unshadow_pte(sg->mm, raddr, (pte_t *) table);
1320 }
1321 
1322 /**
1323  * __gmap_unshadow_pgt - remove all entries from a shadow page table
1324  * @sg: pointer to the shadow guest address space structure
1325  * @raddr: rmap address in the shadow guest address space
1326  * @pgt: pointer to the start of a shadow page table
1327  *
1328  * Called with the sg->guest_table_lock
1329  */
__gmap_unshadow_pgt(struct gmap * sg,unsigned long raddr,unsigned long * pgt)1330 static void __gmap_unshadow_pgt(struct gmap *sg, unsigned long raddr,
1331 				unsigned long *pgt)
1332 {
1333 	int i;
1334 
1335 	BUG_ON(!gmap_is_shadow(sg));
1336 	for (i = 0; i < _PAGE_ENTRIES; i++, raddr += _PAGE_SIZE)
1337 		pgt[i] = _PAGE_INVALID;
1338 }
1339 
1340 /**
1341  * gmap_unshadow_pgt - remove a shadow page table from a segment entry
1342  * @sg: pointer to the shadow guest address space structure
1343  * @raddr: address in the shadow guest address space
1344  *
1345  * Called with the sg->guest_table_lock
1346  */
gmap_unshadow_pgt(struct gmap * sg,unsigned long raddr)1347 static void gmap_unshadow_pgt(struct gmap *sg, unsigned long raddr)
1348 {
1349 	unsigned long *ste;
1350 	phys_addr_t sto, pgt;
1351 	struct page *page;
1352 
1353 	BUG_ON(!gmap_is_shadow(sg));
1354 	ste = gmap_table_walk(sg, raddr, 1); /* get segment pointer */
1355 	if (!ste || !(*ste & _SEGMENT_ENTRY_ORIGIN))
1356 		return;
1357 	gmap_call_notifier(sg, raddr, raddr + _SEGMENT_SIZE - 1);
1358 	sto = __pa(ste - ((raddr & _SEGMENT_INDEX) >> _SEGMENT_SHIFT));
1359 	gmap_idte_one(sto | _ASCE_TYPE_SEGMENT, raddr);
1360 	pgt = *ste & _SEGMENT_ENTRY_ORIGIN;
1361 	*ste = _SEGMENT_ENTRY_EMPTY;
1362 	__gmap_unshadow_pgt(sg, raddr, __va(pgt));
1363 	/* Free page table */
1364 	page = phys_to_page(pgt);
1365 	list_del(&page->lru);
1366 	page_table_free_pgste(page);
1367 }
1368 
1369 /**
1370  * __gmap_unshadow_sgt - remove all entries from a shadow segment table
1371  * @sg: pointer to the shadow guest address space structure
1372  * @raddr: rmap address in the shadow guest address space
1373  * @sgt: pointer to the start of a shadow segment table
1374  *
1375  * Called with the sg->guest_table_lock
1376  */
__gmap_unshadow_sgt(struct gmap * sg,unsigned long raddr,unsigned long * sgt)1377 static void __gmap_unshadow_sgt(struct gmap *sg, unsigned long raddr,
1378 				unsigned long *sgt)
1379 {
1380 	struct page *page;
1381 	phys_addr_t pgt;
1382 	int i;
1383 
1384 	BUG_ON(!gmap_is_shadow(sg));
1385 	for (i = 0; i < _CRST_ENTRIES; i++, raddr += _SEGMENT_SIZE) {
1386 		if (!(sgt[i] & _SEGMENT_ENTRY_ORIGIN))
1387 			continue;
1388 		pgt = sgt[i] & _REGION_ENTRY_ORIGIN;
1389 		sgt[i] = _SEGMENT_ENTRY_EMPTY;
1390 		__gmap_unshadow_pgt(sg, raddr, __va(pgt));
1391 		/* Free page table */
1392 		page = phys_to_page(pgt);
1393 		list_del(&page->lru);
1394 		page_table_free_pgste(page);
1395 	}
1396 }
1397 
1398 /**
1399  * gmap_unshadow_sgt - remove a shadow segment table from a region-3 entry
1400  * @sg: pointer to the shadow guest address space structure
1401  * @raddr: rmap address in the shadow guest address space
1402  *
1403  * Called with the shadow->guest_table_lock
1404  */
gmap_unshadow_sgt(struct gmap * sg,unsigned long raddr)1405 static void gmap_unshadow_sgt(struct gmap *sg, unsigned long raddr)
1406 {
1407 	unsigned long r3o, *r3e;
1408 	phys_addr_t sgt;
1409 	struct page *page;
1410 
1411 	BUG_ON(!gmap_is_shadow(sg));
1412 	r3e = gmap_table_walk(sg, raddr, 2); /* get region-3 pointer */
1413 	if (!r3e || !(*r3e & _REGION_ENTRY_ORIGIN))
1414 		return;
1415 	gmap_call_notifier(sg, raddr, raddr + _REGION3_SIZE - 1);
1416 	r3o = (unsigned long) (r3e - ((raddr & _REGION3_INDEX) >> _REGION3_SHIFT));
1417 	gmap_idte_one(__pa(r3o) | _ASCE_TYPE_REGION3, raddr);
1418 	sgt = *r3e & _REGION_ENTRY_ORIGIN;
1419 	*r3e = _REGION3_ENTRY_EMPTY;
1420 	__gmap_unshadow_sgt(sg, raddr, __va(sgt));
1421 	/* Free segment table */
1422 	page = phys_to_page(sgt);
1423 	list_del(&page->lru);
1424 	__free_pages(page, CRST_ALLOC_ORDER);
1425 }
1426 
1427 /**
1428  * __gmap_unshadow_r3t - remove all entries from a shadow region-3 table
1429  * @sg: pointer to the shadow guest address space structure
1430  * @raddr: address in the shadow guest address space
1431  * @r3t: pointer to the start of a shadow region-3 table
1432  *
1433  * Called with the sg->guest_table_lock
1434  */
__gmap_unshadow_r3t(struct gmap * sg,unsigned long raddr,unsigned long * r3t)1435 static void __gmap_unshadow_r3t(struct gmap *sg, unsigned long raddr,
1436 				unsigned long *r3t)
1437 {
1438 	struct page *page;
1439 	phys_addr_t sgt;
1440 	int i;
1441 
1442 	BUG_ON(!gmap_is_shadow(sg));
1443 	for (i = 0; i < _CRST_ENTRIES; i++, raddr += _REGION3_SIZE) {
1444 		if (!(r3t[i] & _REGION_ENTRY_ORIGIN))
1445 			continue;
1446 		sgt = r3t[i] & _REGION_ENTRY_ORIGIN;
1447 		r3t[i] = _REGION3_ENTRY_EMPTY;
1448 		__gmap_unshadow_sgt(sg, raddr, __va(sgt));
1449 		/* Free segment table */
1450 		page = phys_to_page(sgt);
1451 		list_del(&page->lru);
1452 		__free_pages(page, CRST_ALLOC_ORDER);
1453 	}
1454 }
1455 
1456 /**
1457  * gmap_unshadow_r3t - remove a shadow region-3 table from a region-2 entry
1458  * @sg: pointer to the shadow guest address space structure
1459  * @raddr: rmap address in the shadow guest address space
1460  *
1461  * Called with the sg->guest_table_lock
1462  */
gmap_unshadow_r3t(struct gmap * sg,unsigned long raddr)1463 static void gmap_unshadow_r3t(struct gmap *sg, unsigned long raddr)
1464 {
1465 	unsigned long r2o, *r2e;
1466 	phys_addr_t r3t;
1467 	struct page *page;
1468 
1469 	BUG_ON(!gmap_is_shadow(sg));
1470 	r2e = gmap_table_walk(sg, raddr, 3); /* get region-2 pointer */
1471 	if (!r2e || !(*r2e & _REGION_ENTRY_ORIGIN))
1472 		return;
1473 	gmap_call_notifier(sg, raddr, raddr + _REGION2_SIZE - 1);
1474 	r2o = (unsigned long) (r2e - ((raddr & _REGION2_INDEX) >> _REGION2_SHIFT));
1475 	gmap_idte_one(__pa(r2o) | _ASCE_TYPE_REGION2, raddr);
1476 	r3t = *r2e & _REGION_ENTRY_ORIGIN;
1477 	*r2e = _REGION2_ENTRY_EMPTY;
1478 	__gmap_unshadow_r3t(sg, raddr, __va(r3t));
1479 	/* Free region 3 table */
1480 	page = phys_to_page(r3t);
1481 	list_del(&page->lru);
1482 	__free_pages(page, CRST_ALLOC_ORDER);
1483 }
1484 
1485 /**
1486  * __gmap_unshadow_r2t - remove all entries from a shadow region-2 table
1487  * @sg: pointer to the shadow guest address space structure
1488  * @raddr: rmap address in the shadow guest address space
1489  * @r2t: pointer to the start of a shadow region-2 table
1490  *
1491  * Called with the sg->guest_table_lock
1492  */
__gmap_unshadow_r2t(struct gmap * sg,unsigned long raddr,unsigned long * r2t)1493 static void __gmap_unshadow_r2t(struct gmap *sg, unsigned long raddr,
1494 				unsigned long *r2t)
1495 {
1496 	phys_addr_t r3t;
1497 	struct page *page;
1498 	int i;
1499 
1500 	BUG_ON(!gmap_is_shadow(sg));
1501 	for (i = 0; i < _CRST_ENTRIES; i++, raddr += _REGION2_SIZE) {
1502 		if (!(r2t[i] & _REGION_ENTRY_ORIGIN))
1503 			continue;
1504 		r3t = r2t[i] & _REGION_ENTRY_ORIGIN;
1505 		r2t[i] = _REGION2_ENTRY_EMPTY;
1506 		__gmap_unshadow_r3t(sg, raddr, __va(r3t));
1507 		/* Free region 3 table */
1508 		page = phys_to_page(r3t);
1509 		list_del(&page->lru);
1510 		__free_pages(page, CRST_ALLOC_ORDER);
1511 	}
1512 }
1513 
1514 /**
1515  * gmap_unshadow_r2t - remove a shadow region-2 table from a region-1 entry
1516  * @sg: pointer to the shadow guest address space structure
1517  * @raddr: rmap address in the shadow guest address space
1518  *
1519  * Called with the sg->guest_table_lock
1520  */
gmap_unshadow_r2t(struct gmap * sg,unsigned long raddr)1521 static void gmap_unshadow_r2t(struct gmap *sg, unsigned long raddr)
1522 {
1523 	unsigned long r1o, *r1e;
1524 	struct page *page;
1525 	phys_addr_t r2t;
1526 
1527 	BUG_ON(!gmap_is_shadow(sg));
1528 	r1e = gmap_table_walk(sg, raddr, 4); /* get region-1 pointer */
1529 	if (!r1e || !(*r1e & _REGION_ENTRY_ORIGIN))
1530 		return;
1531 	gmap_call_notifier(sg, raddr, raddr + _REGION1_SIZE - 1);
1532 	r1o = (unsigned long) (r1e - ((raddr & _REGION1_INDEX) >> _REGION1_SHIFT));
1533 	gmap_idte_one(__pa(r1o) | _ASCE_TYPE_REGION1, raddr);
1534 	r2t = *r1e & _REGION_ENTRY_ORIGIN;
1535 	*r1e = _REGION1_ENTRY_EMPTY;
1536 	__gmap_unshadow_r2t(sg, raddr, __va(r2t));
1537 	/* Free region 2 table */
1538 	page = phys_to_page(r2t);
1539 	list_del(&page->lru);
1540 	__free_pages(page, CRST_ALLOC_ORDER);
1541 }
1542 
1543 /**
1544  * __gmap_unshadow_r1t - remove all entries from a shadow region-1 table
1545  * @sg: pointer to the shadow guest address space structure
1546  * @raddr: rmap address in the shadow guest address space
1547  * @r1t: pointer to the start of a shadow region-1 table
1548  *
1549  * Called with the shadow->guest_table_lock
1550  */
__gmap_unshadow_r1t(struct gmap * sg,unsigned long raddr,unsigned long * r1t)1551 static void __gmap_unshadow_r1t(struct gmap *sg, unsigned long raddr,
1552 				unsigned long *r1t)
1553 {
1554 	unsigned long asce;
1555 	struct page *page;
1556 	phys_addr_t r2t;
1557 	int i;
1558 
1559 	BUG_ON(!gmap_is_shadow(sg));
1560 	asce = __pa(r1t) | _ASCE_TYPE_REGION1;
1561 	for (i = 0; i < _CRST_ENTRIES; i++, raddr += _REGION1_SIZE) {
1562 		if (!(r1t[i] & _REGION_ENTRY_ORIGIN))
1563 			continue;
1564 		r2t = r1t[i] & _REGION_ENTRY_ORIGIN;
1565 		__gmap_unshadow_r2t(sg, raddr, __va(r2t));
1566 		/* Clear entry and flush translation r1t -> r2t */
1567 		gmap_idte_one(asce, raddr);
1568 		r1t[i] = _REGION1_ENTRY_EMPTY;
1569 		/* Free region 2 table */
1570 		page = phys_to_page(r2t);
1571 		list_del(&page->lru);
1572 		__free_pages(page, CRST_ALLOC_ORDER);
1573 	}
1574 }
1575 
1576 /**
1577  * gmap_unshadow - remove a shadow page table completely
1578  * @sg: pointer to the shadow guest address space structure
1579  *
1580  * Called with sg->guest_table_lock
1581  */
gmap_unshadow(struct gmap * sg)1582 static void gmap_unshadow(struct gmap *sg)
1583 {
1584 	unsigned long *table;
1585 
1586 	BUG_ON(!gmap_is_shadow(sg));
1587 	if (sg->removed)
1588 		return;
1589 	sg->removed = 1;
1590 	gmap_call_notifier(sg, 0, -1UL);
1591 	gmap_flush_tlb(sg);
1592 	table = __va(sg->asce & _ASCE_ORIGIN);
1593 	switch (sg->asce & _ASCE_TYPE_MASK) {
1594 	case _ASCE_TYPE_REGION1:
1595 		__gmap_unshadow_r1t(sg, 0, table);
1596 		break;
1597 	case _ASCE_TYPE_REGION2:
1598 		__gmap_unshadow_r2t(sg, 0, table);
1599 		break;
1600 	case _ASCE_TYPE_REGION3:
1601 		__gmap_unshadow_r3t(sg, 0, table);
1602 		break;
1603 	case _ASCE_TYPE_SEGMENT:
1604 		__gmap_unshadow_sgt(sg, 0, table);
1605 		break;
1606 	}
1607 }
1608 
1609 /**
1610  * gmap_find_shadow - find a specific asce in the list of shadow tables
1611  * @parent: pointer to the parent gmap
1612  * @asce: ASCE for which the shadow table is created
1613  * @edat_level: edat level to be used for the shadow translation
1614  *
1615  * Returns the pointer to a gmap if a shadow table with the given asce is
1616  * already available, ERR_PTR(-EAGAIN) if another one is just being created,
1617  * otherwise NULL
1618  */
gmap_find_shadow(struct gmap * parent,unsigned long asce,int edat_level)1619 static struct gmap *gmap_find_shadow(struct gmap *parent, unsigned long asce,
1620 				     int edat_level)
1621 {
1622 	struct gmap *sg;
1623 
1624 	list_for_each_entry(sg, &parent->children, list) {
1625 		if (sg->orig_asce != asce || sg->edat_level != edat_level ||
1626 		    sg->removed)
1627 			continue;
1628 		if (!sg->initialized)
1629 			return ERR_PTR(-EAGAIN);
1630 		refcount_inc(&sg->ref_count);
1631 		return sg;
1632 	}
1633 	return NULL;
1634 }
1635 
1636 /**
1637  * gmap_shadow_valid - check if a shadow guest address space matches the
1638  *                     given properties and is still valid
1639  * @sg: pointer to the shadow guest address space structure
1640  * @asce: ASCE for which the shadow table is requested
1641  * @edat_level: edat level to be used for the shadow translation
1642  *
1643  * Returns 1 if the gmap shadow is still valid and matches the given
1644  * properties, the caller can continue using it. Returns 0 otherwise, the
1645  * caller has to request a new shadow gmap in this case.
1646  *
1647  */
gmap_shadow_valid(struct gmap * sg,unsigned long asce,int edat_level)1648 int gmap_shadow_valid(struct gmap *sg, unsigned long asce, int edat_level)
1649 {
1650 	if (sg->removed)
1651 		return 0;
1652 	return sg->orig_asce == asce && sg->edat_level == edat_level;
1653 }
1654 EXPORT_SYMBOL_GPL(gmap_shadow_valid);
1655 
1656 /**
1657  * gmap_shadow - create/find a shadow guest address space
1658  * @parent: pointer to the parent gmap
1659  * @asce: ASCE for which the shadow table is created
1660  * @edat_level: edat level to be used for the shadow translation
1661  *
1662  * The pages of the top level page table referred by the asce parameter
1663  * will be set to read-only and marked in the PGSTEs of the kvm process.
1664  * The shadow table will be removed automatically on any change to the
1665  * PTE mapping for the source table.
1666  *
1667  * Returns a guest address space structure, ERR_PTR(-ENOMEM) if out of memory,
1668  * ERR_PTR(-EAGAIN) if the caller has to retry and ERR_PTR(-EFAULT) if the
1669  * parent gmap table could not be protected.
1670  */
gmap_shadow(struct gmap * parent,unsigned long asce,int edat_level)1671 struct gmap *gmap_shadow(struct gmap *parent, unsigned long asce,
1672 			 int edat_level)
1673 {
1674 	struct gmap *sg, *new;
1675 	unsigned long limit;
1676 	int rc;
1677 
1678 	BUG_ON(parent->mm->context.allow_gmap_hpage_1m);
1679 	BUG_ON(gmap_is_shadow(parent));
1680 	spin_lock(&parent->shadow_lock);
1681 	sg = gmap_find_shadow(parent, asce, edat_level);
1682 	spin_unlock(&parent->shadow_lock);
1683 	if (sg)
1684 		return sg;
1685 	/* Create a new shadow gmap */
1686 	limit = -1UL >> (33 - (((asce & _ASCE_TYPE_MASK) >> 2) * 11));
1687 	if (asce & _ASCE_REAL_SPACE)
1688 		limit = -1UL;
1689 	new = gmap_alloc(limit);
1690 	if (!new)
1691 		return ERR_PTR(-ENOMEM);
1692 	new->mm = parent->mm;
1693 	new->parent = gmap_get(parent);
1694 	new->orig_asce = asce;
1695 	new->edat_level = edat_level;
1696 	new->initialized = false;
1697 	spin_lock(&parent->shadow_lock);
1698 	/* Recheck if another CPU created the same shadow */
1699 	sg = gmap_find_shadow(parent, asce, edat_level);
1700 	if (sg) {
1701 		spin_unlock(&parent->shadow_lock);
1702 		gmap_free(new);
1703 		return sg;
1704 	}
1705 	if (asce & _ASCE_REAL_SPACE) {
1706 		/* only allow one real-space gmap shadow */
1707 		list_for_each_entry(sg, &parent->children, list) {
1708 			if (sg->orig_asce & _ASCE_REAL_SPACE) {
1709 				spin_lock(&sg->guest_table_lock);
1710 				gmap_unshadow(sg);
1711 				spin_unlock(&sg->guest_table_lock);
1712 				list_del(&sg->list);
1713 				gmap_put(sg);
1714 				break;
1715 			}
1716 		}
1717 	}
1718 	refcount_set(&new->ref_count, 2);
1719 	list_add(&new->list, &parent->children);
1720 	if (asce & _ASCE_REAL_SPACE) {
1721 		/* nothing to protect, return right away */
1722 		new->initialized = true;
1723 		spin_unlock(&parent->shadow_lock);
1724 		return new;
1725 	}
1726 	spin_unlock(&parent->shadow_lock);
1727 	/* protect after insertion, so it will get properly invalidated */
1728 	mmap_read_lock(parent->mm);
1729 	rc = gmap_protect_range(parent, asce & _ASCE_ORIGIN,
1730 				((asce & _ASCE_TABLE_LENGTH) + 1) * PAGE_SIZE,
1731 				PROT_READ, GMAP_NOTIFY_SHADOW);
1732 	mmap_read_unlock(parent->mm);
1733 	spin_lock(&parent->shadow_lock);
1734 	new->initialized = true;
1735 	if (rc) {
1736 		list_del(&new->list);
1737 		gmap_free(new);
1738 		new = ERR_PTR(rc);
1739 	}
1740 	spin_unlock(&parent->shadow_lock);
1741 	return new;
1742 }
1743 EXPORT_SYMBOL_GPL(gmap_shadow);
1744 
1745 /**
1746  * gmap_shadow_r2t - create an empty shadow region 2 table
1747  * @sg: pointer to the shadow guest address space structure
1748  * @saddr: faulting address in the shadow gmap
1749  * @r2t: parent gmap address of the region 2 table to get shadowed
1750  * @fake: r2t references contiguous guest memory block, not a r2t
1751  *
1752  * The r2t parameter specifies the address of the source table. The
1753  * four pages of the source table are made read-only in the parent gmap
1754  * address space. A write to the source table area @r2t will automatically
1755  * remove the shadow r2 table and all of its descendants.
1756  *
1757  * Returns 0 if successfully shadowed or already shadowed, -EAGAIN if the
1758  * shadow table structure is incomplete, -ENOMEM if out of memory and
1759  * -EFAULT if an address in the parent gmap could not be resolved.
1760  *
1761  * Called with sg->mm->mmap_lock in read.
1762  */
gmap_shadow_r2t(struct gmap * sg,unsigned long saddr,unsigned long r2t,int fake)1763 int gmap_shadow_r2t(struct gmap *sg, unsigned long saddr, unsigned long r2t,
1764 		    int fake)
1765 {
1766 	unsigned long raddr, origin, offset, len;
1767 	unsigned long *table;
1768 	phys_addr_t s_r2t;
1769 	struct page *page;
1770 	int rc;
1771 
1772 	BUG_ON(!gmap_is_shadow(sg));
1773 	/* Allocate a shadow region second table */
1774 	page = gmap_alloc_crst();
1775 	if (!page)
1776 		return -ENOMEM;
1777 	page->index = r2t & _REGION_ENTRY_ORIGIN;
1778 	if (fake)
1779 		page->index |= GMAP_SHADOW_FAKE_TABLE;
1780 	s_r2t = page_to_phys(page);
1781 	/* Install shadow region second table */
1782 	spin_lock(&sg->guest_table_lock);
1783 	table = gmap_table_walk(sg, saddr, 4); /* get region-1 pointer */
1784 	if (!table) {
1785 		rc = -EAGAIN;		/* Race with unshadow */
1786 		goto out_free;
1787 	}
1788 	if (!(*table & _REGION_ENTRY_INVALID)) {
1789 		rc = 0;			/* Already established */
1790 		goto out_free;
1791 	} else if (*table & _REGION_ENTRY_ORIGIN) {
1792 		rc = -EAGAIN;		/* Race with shadow */
1793 		goto out_free;
1794 	}
1795 	crst_table_init(__va(s_r2t), _REGION2_ENTRY_EMPTY);
1796 	/* mark as invalid as long as the parent table is not protected */
1797 	*table = s_r2t | _REGION_ENTRY_LENGTH |
1798 		 _REGION_ENTRY_TYPE_R1 | _REGION_ENTRY_INVALID;
1799 	if (sg->edat_level >= 1)
1800 		*table |= (r2t & _REGION_ENTRY_PROTECT);
1801 	list_add(&page->lru, &sg->crst_list);
1802 	if (fake) {
1803 		/* nothing to protect for fake tables */
1804 		*table &= ~_REGION_ENTRY_INVALID;
1805 		spin_unlock(&sg->guest_table_lock);
1806 		return 0;
1807 	}
1808 	spin_unlock(&sg->guest_table_lock);
1809 	/* Make r2t read-only in parent gmap page table */
1810 	raddr = (saddr & _REGION1_MASK) | _SHADOW_RMAP_REGION1;
1811 	origin = r2t & _REGION_ENTRY_ORIGIN;
1812 	offset = ((r2t & _REGION_ENTRY_OFFSET) >> 6) * PAGE_SIZE;
1813 	len = ((r2t & _REGION_ENTRY_LENGTH) + 1) * PAGE_SIZE - offset;
1814 	rc = gmap_protect_rmap(sg, raddr, origin + offset, len);
1815 	spin_lock(&sg->guest_table_lock);
1816 	if (!rc) {
1817 		table = gmap_table_walk(sg, saddr, 4);
1818 		if (!table || (*table & _REGION_ENTRY_ORIGIN) != s_r2t)
1819 			rc = -EAGAIN;		/* Race with unshadow */
1820 		else
1821 			*table &= ~_REGION_ENTRY_INVALID;
1822 	} else {
1823 		gmap_unshadow_r2t(sg, raddr);
1824 	}
1825 	spin_unlock(&sg->guest_table_lock);
1826 	return rc;
1827 out_free:
1828 	spin_unlock(&sg->guest_table_lock);
1829 	__free_pages(page, CRST_ALLOC_ORDER);
1830 	return rc;
1831 }
1832 EXPORT_SYMBOL_GPL(gmap_shadow_r2t);
1833 
1834 /**
1835  * gmap_shadow_r3t - create a shadow region 3 table
1836  * @sg: pointer to the shadow guest address space structure
1837  * @saddr: faulting address in the shadow gmap
1838  * @r3t: parent gmap address of the region 3 table to get shadowed
1839  * @fake: r3t references contiguous guest memory block, not a r3t
1840  *
1841  * Returns 0 if successfully shadowed or already shadowed, -EAGAIN if the
1842  * shadow table structure is incomplete, -ENOMEM if out of memory and
1843  * -EFAULT if an address in the parent gmap could not be resolved.
1844  *
1845  * Called with sg->mm->mmap_lock in read.
1846  */
gmap_shadow_r3t(struct gmap * sg,unsigned long saddr,unsigned long r3t,int fake)1847 int gmap_shadow_r3t(struct gmap *sg, unsigned long saddr, unsigned long r3t,
1848 		    int fake)
1849 {
1850 	unsigned long raddr, origin, offset, len;
1851 	unsigned long *table;
1852 	phys_addr_t s_r3t;
1853 	struct page *page;
1854 	int rc;
1855 
1856 	BUG_ON(!gmap_is_shadow(sg));
1857 	/* Allocate a shadow region second table */
1858 	page = gmap_alloc_crst();
1859 	if (!page)
1860 		return -ENOMEM;
1861 	page->index = r3t & _REGION_ENTRY_ORIGIN;
1862 	if (fake)
1863 		page->index |= GMAP_SHADOW_FAKE_TABLE;
1864 	s_r3t = page_to_phys(page);
1865 	/* Install shadow region second table */
1866 	spin_lock(&sg->guest_table_lock);
1867 	table = gmap_table_walk(sg, saddr, 3); /* get region-2 pointer */
1868 	if (!table) {
1869 		rc = -EAGAIN;		/* Race with unshadow */
1870 		goto out_free;
1871 	}
1872 	if (!(*table & _REGION_ENTRY_INVALID)) {
1873 		rc = 0;			/* Already established */
1874 		goto out_free;
1875 	} else if (*table & _REGION_ENTRY_ORIGIN) {
1876 		rc = -EAGAIN;		/* Race with shadow */
1877 		goto out_free;
1878 	}
1879 	crst_table_init(__va(s_r3t), _REGION3_ENTRY_EMPTY);
1880 	/* mark as invalid as long as the parent table is not protected */
1881 	*table = s_r3t | _REGION_ENTRY_LENGTH |
1882 		 _REGION_ENTRY_TYPE_R2 | _REGION_ENTRY_INVALID;
1883 	if (sg->edat_level >= 1)
1884 		*table |= (r3t & _REGION_ENTRY_PROTECT);
1885 	list_add(&page->lru, &sg->crst_list);
1886 	if (fake) {
1887 		/* nothing to protect for fake tables */
1888 		*table &= ~_REGION_ENTRY_INVALID;
1889 		spin_unlock(&sg->guest_table_lock);
1890 		return 0;
1891 	}
1892 	spin_unlock(&sg->guest_table_lock);
1893 	/* Make r3t read-only in parent gmap page table */
1894 	raddr = (saddr & _REGION2_MASK) | _SHADOW_RMAP_REGION2;
1895 	origin = r3t & _REGION_ENTRY_ORIGIN;
1896 	offset = ((r3t & _REGION_ENTRY_OFFSET) >> 6) * PAGE_SIZE;
1897 	len = ((r3t & _REGION_ENTRY_LENGTH) + 1) * PAGE_SIZE - offset;
1898 	rc = gmap_protect_rmap(sg, raddr, origin + offset, len);
1899 	spin_lock(&sg->guest_table_lock);
1900 	if (!rc) {
1901 		table = gmap_table_walk(sg, saddr, 3);
1902 		if (!table || (*table & _REGION_ENTRY_ORIGIN) != s_r3t)
1903 			rc = -EAGAIN;		/* Race with unshadow */
1904 		else
1905 			*table &= ~_REGION_ENTRY_INVALID;
1906 	} else {
1907 		gmap_unshadow_r3t(sg, raddr);
1908 	}
1909 	spin_unlock(&sg->guest_table_lock);
1910 	return rc;
1911 out_free:
1912 	spin_unlock(&sg->guest_table_lock);
1913 	__free_pages(page, CRST_ALLOC_ORDER);
1914 	return rc;
1915 }
1916 EXPORT_SYMBOL_GPL(gmap_shadow_r3t);
1917 
1918 /**
1919  * gmap_shadow_sgt - create a shadow segment table
1920  * @sg: pointer to the shadow guest address space structure
1921  * @saddr: faulting address in the shadow gmap
1922  * @sgt: parent gmap address of the segment table to get shadowed
1923  * @fake: sgt references contiguous guest memory block, not a sgt
1924  *
1925  * Returns: 0 if successfully shadowed or already shadowed, -EAGAIN if the
1926  * shadow table structure is incomplete, -ENOMEM if out of memory and
1927  * -EFAULT if an address in the parent gmap could not be resolved.
1928  *
1929  * Called with sg->mm->mmap_lock in read.
1930  */
gmap_shadow_sgt(struct gmap * sg,unsigned long saddr,unsigned long sgt,int fake)1931 int gmap_shadow_sgt(struct gmap *sg, unsigned long saddr, unsigned long sgt,
1932 		    int fake)
1933 {
1934 	unsigned long raddr, origin, offset, len;
1935 	unsigned long *table;
1936 	phys_addr_t s_sgt;
1937 	struct page *page;
1938 	int rc;
1939 
1940 	BUG_ON(!gmap_is_shadow(sg) || (sgt & _REGION3_ENTRY_LARGE));
1941 	/* Allocate a shadow segment table */
1942 	page = gmap_alloc_crst();
1943 	if (!page)
1944 		return -ENOMEM;
1945 	page->index = sgt & _REGION_ENTRY_ORIGIN;
1946 	if (fake)
1947 		page->index |= GMAP_SHADOW_FAKE_TABLE;
1948 	s_sgt = page_to_phys(page);
1949 	/* Install shadow region second table */
1950 	spin_lock(&sg->guest_table_lock);
1951 	table = gmap_table_walk(sg, saddr, 2); /* get region-3 pointer */
1952 	if (!table) {
1953 		rc = -EAGAIN;		/* Race with unshadow */
1954 		goto out_free;
1955 	}
1956 	if (!(*table & _REGION_ENTRY_INVALID)) {
1957 		rc = 0;			/* Already established */
1958 		goto out_free;
1959 	} else if (*table & _REGION_ENTRY_ORIGIN) {
1960 		rc = -EAGAIN;		/* Race with shadow */
1961 		goto out_free;
1962 	}
1963 	crst_table_init(__va(s_sgt), _SEGMENT_ENTRY_EMPTY);
1964 	/* mark as invalid as long as the parent table is not protected */
1965 	*table = s_sgt | _REGION_ENTRY_LENGTH |
1966 		 _REGION_ENTRY_TYPE_R3 | _REGION_ENTRY_INVALID;
1967 	if (sg->edat_level >= 1)
1968 		*table |= sgt & _REGION_ENTRY_PROTECT;
1969 	list_add(&page->lru, &sg->crst_list);
1970 	if (fake) {
1971 		/* nothing to protect for fake tables */
1972 		*table &= ~_REGION_ENTRY_INVALID;
1973 		spin_unlock(&sg->guest_table_lock);
1974 		return 0;
1975 	}
1976 	spin_unlock(&sg->guest_table_lock);
1977 	/* Make sgt read-only in parent gmap page table */
1978 	raddr = (saddr & _REGION3_MASK) | _SHADOW_RMAP_REGION3;
1979 	origin = sgt & _REGION_ENTRY_ORIGIN;
1980 	offset = ((sgt & _REGION_ENTRY_OFFSET) >> 6) * PAGE_SIZE;
1981 	len = ((sgt & _REGION_ENTRY_LENGTH) + 1) * PAGE_SIZE - offset;
1982 	rc = gmap_protect_rmap(sg, raddr, origin + offset, len);
1983 	spin_lock(&sg->guest_table_lock);
1984 	if (!rc) {
1985 		table = gmap_table_walk(sg, saddr, 2);
1986 		if (!table || (*table & _REGION_ENTRY_ORIGIN) != s_sgt)
1987 			rc = -EAGAIN;		/* Race with unshadow */
1988 		else
1989 			*table &= ~_REGION_ENTRY_INVALID;
1990 	} else {
1991 		gmap_unshadow_sgt(sg, raddr);
1992 	}
1993 	spin_unlock(&sg->guest_table_lock);
1994 	return rc;
1995 out_free:
1996 	spin_unlock(&sg->guest_table_lock);
1997 	__free_pages(page, CRST_ALLOC_ORDER);
1998 	return rc;
1999 }
2000 EXPORT_SYMBOL_GPL(gmap_shadow_sgt);
2001 
2002 /**
2003  * gmap_shadow_pgt_lookup - find a shadow page table
2004  * @sg: pointer to the shadow guest address space structure
2005  * @saddr: the address in the shadow aguest address space
2006  * @pgt: parent gmap address of the page table to get shadowed
2007  * @dat_protection: if the pgtable is marked as protected by dat
2008  * @fake: pgt references contiguous guest memory block, not a pgtable
2009  *
2010  * Returns 0 if the shadow page table was found and -EAGAIN if the page
2011  * table was not found.
2012  *
2013  * Called with sg->mm->mmap_lock in read.
2014  */
gmap_shadow_pgt_lookup(struct gmap * sg,unsigned long saddr,unsigned long * pgt,int * dat_protection,int * fake)2015 int gmap_shadow_pgt_lookup(struct gmap *sg, unsigned long saddr,
2016 			   unsigned long *pgt, int *dat_protection,
2017 			   int *fake)
2018 {
2019 	unsigned long *table;
2020 	struct page *page;
2021 	int rc;
2022 
2023 	BUG_ON(!gmap_is_shadow(sg));
2024 	spin_lock(&sg->guest_table_lock);
2025 	table = gmap_table_walk(sg, saddr, 1); /* get segment pointer */
2026 	if (table && !(*table & _SEGMENT_ENTRY_INVALID)) {
2027 		/* Shadow page tables are full pages (pte+pgste) */
2028 		page = pfn_to_page(*table >> PAGE_SHIFT);
2029 		*pgt = page->index & ~GMAP_SHADOW_FAKE_TABLE;
2030 		*dat_protection = !!(*table & _SEGMENT_ENTRY_PROTECT);
2031 		*fake = !!(page->index & GMAP_SHADOW_FAKE_TABLE);
2032 		rc = 0;
2033 	} else  {
2034 		rc = -EAGAIN;
2035 	}
2036 	spin_unlock(&sg->guest_table_lock);
2037 	return rc;
2038 
2039 }
2040 EXPORT_SYMBOL_GPL(gmap_shadow_pgt_lookup);
2041 
2042 /**
2043  * gmap_shadow_pgt - instantiate a shadow page table
2044  * @sg: pointer to the shadow guest address space structure
2045  * @saddr: faulting address in the shadow gmap
2046  * @pgt: parent gmap address of the page table to get shadowed
2047  * @fake: pgt references contiguous guest memory block, not a pgtable
2048  *
2049  * Returns 0 if successfully shadowed or already shadowed, -EAGAIN if the
2050  * shadow table structure is incomplete, -ENOMEM if out of memory,
2051  * -EFAULT if an address in the parent gmap could not be resolved and
2052  *
2053  * Called with gmap->mm->mmap_lock in read
2054  */
gmap_shadow_pgt(struct gmap * sg,unsigned long saddr,unsigned long pgt,int fake)2055 int gmap_shadow_pgt(struct gmap *sg, unsigned long saddr, unsigned long pgt,
2056 		    int fake)
2057 {
2058 	unsigned long raddr, origin;
2059 	unsigned long *table;
2060 	struct page *page;
2061 	phys_addr_t s_pgt;
2062 	int rc;
2063 
2064 	BUG_ON(!gmap_is_shadow(sg) || (pgt & _SEGMENT_ENTRY_LARGE));
2065 	/* Allocate a shadow page table */
2066 	page = page_table_alloc_pgste(sg->mm);
2067 	if (!page)
2068 		return -ENOMEM;
2069 	page->index = pgt & _SEGMENT_ENTRY_ORIGIN;
2070 	if (fake)
2071 		page->index |= GMAP_SHADOW_FAKE_TABLE;
2072 	s_pgt = page_to_phys(page);
2073 	/* Install shadow page table */
2074 	spin_lock(&sg->guest_table_lock);
2075 	table = gmap_table_walk(sg, saddr, 1); /* get segment pointer */
2076 	if (!table) {
2077 		rc = -EAGAIN;		/* Race with unshadow */
2078 		goto out_free;
2079 	}
2080 	if (!(*table & _SEGMENT_ENTRY_INVALID)) {
2081 		rc = 0;			/* Already established */
2082 		goto out_free;
2083 	} else if (*table & _SEGMENT_ENTRY_ORIGIN) {
2084 		rc = -EAGAIN;		/* Race with shadow */
2085 		goto out_free;
2086 	}
2087 	/* mark as invalid as long as the parent table is not protected */
2088 	*table = (unsigned long) s_pgt | _SEGMENT_ENTRY |
2089 		 (pgt & _SEGMENT_ENTRY_PROTECT) | _SEGMENT_ENTRY_INVALID;
2090 	list_add(&page->lru, &sg->pt_list);
2091 	if (fake) {
2092 		/* nothing to protect for fake tables */
2093 		*table &= ~_SEGMENT_ENTRY_INVALID;
2094 		spin_unlock(&sg->guest_table_lock);
2095 		return 0;
2096 	}
2097 	spin_unlock(&sg->guest_table_lock);
2098 	/* Make pgt read-only in parent gmap page table (not the pgste) */
2099 	raddr = (saddr & _SEGMENT_MASK) | _SHADOW_RMAP_SEGMENT;
2100 	origin = pgt & _SEGMENT_ENTRY_ORIGIN & PAGE_MASK;
2101 	rc = gmap_protect_rmap(sg, raddr, origin, PAGE_SIZE);
2102 	spin_lock(&sg->guest_table_lock);
2103 	if (!rc) {
2104 		table = gmap_table_walk(sg, saddr, 1);
2105 		if (!table || (*table & _SEGMENT_ENTRY_ORIGIN) != s_pgt)
2106 			rc = -EAGAIN;		/* Race with unshadow */
2107 		else
2108 			*table &= ~_SEGMENT_ENTRY_INVALID;
2109 	} else {
2110 		gmap_unshadow_pgt(sg, raddr);
2111 	}
2112 	spin_unlock(&sg->guest_table_lock);
2113 	return rc;
2114 out_free:
2115 	spin_unlock(&sg->guest_table_lock);
2116 	page_table_free_pgste(page);
2117 	return rc;
2118 
2119 }
2120 EXPORT_SYMBOL_GPL(gmap_shadow_pgt);
2121 
2122 /**
2123  * gmap_shadow_page - create a shadow page mapping
2124  * @sg: pointer to the shadow guest address space structure
2125  * @saddr: faulting address in the shadow gmap
2126  * @pte: pte in parent gmap address space to get shadowed
2127  *
2128  * Returns 0 if successfully shadowed or already shadowed, -EAGAIN if the
2129  * shadow table structure is incomplete, -ENOMEM if out of memory and
2130  * -EFAULT if an address in the parent gmap could not be resolved.
2131  *
2132  * Called with sg->mm->mmap_lock in read.
2133  */
gmap_shadow_page(struct gmap * sg,unsigned long saddr,pte_t pte)2134 int gmap_shadow_page(struct gmap *sg, unsigned long saddr, pte_t pte)
2135 {
2136 	struct gmap *parent;
2137 	struct gmap_rmap *rmap;
2138 	unsigned long vmaddr, paddr;
2139 	spinlock_t *ptl;
2140 	pte_t *sptep, *tptep;
2141 	int prot;
2142 	int rc;
2143 
2144 	BUG_ON(!gmap_is_shadow(sg));
2145 	parent = sg->parent;
2146 	prot = (pte_val(pte) & _PAGE_PROTECT) ? PROT_READ : PROT_WRITE;
2147 
2148 	rmap = kzalloc(sizeof(*rmap), GFP_KERNEL_ACCOUNT);
2149 	if (!rmap)
2150 		return -ENOMEM;
2151 	rmap->raddr = (saddr & PAGE_MASK) | _SHADOW_RMAP_PGTABLE;
2152 
2153 	while (1) {
2154 		paddr = pte_val(pte) & PAGE_MASK;
2155 		vmaddr = __gmap_translate(parent, paddr);
2156 		if (IS_ERR_VALUE(vmaddr)) {
2157 			rc = vmaddr;
2158 			break;
2159 		}
2160 		rc = radix_tree_preload(GFP_KERNEL_ACCOUNT);
2161 		if (rc)
2162 			break;
2163 		rc = -EAGAIN;
2164 		sptep = gmap_pte_op_walk(parent, paddr, &ptl);
2165 		if (sptep) {
2166 			spin_lock(&sg->guest_table_lock);
2167 			/* Get page table pointer */
2168 			tptep = (pte_t *) gmap_table_walk(sg, saddr, 0);
2169 			if (!tptep) {
2170 				spin_unlock(&sg->guest_table_lock);
2171 				gmap_pte_op_end(sptep, ptl);
2172 				radix_tree_preload_end();
2173 				break;
2174 			}
2175 			rc = ptep_shadow_pte(sg->mm, saddr, sptep, tptep, pte);
2176 			if (rc > 0) {
2177 				/* Success and a new mapping */
2178 				gmap_insert_rmap(sg, vmaddr, rmap);
2179 				rmap = NULL;
2180 				rc = 0;
2181 			}
2182 			gmap_pte_op_end(sptep, ptl);
2183 			spin_unlock(&sg->guest_table_lock);
2184 		}
2185 		radix_tree_preload_end();
2186 		if (!rc)
2187 			break;
2188 		rc = gmap_pte_op_fixup(parent, paddr, vmaddr, prot);
2189 		if (rc)
2190 			break;
2191 	}
2192 	kfree(rmap);
2193 	return rc;
2194 }
2195 EXPORT_SYMBOL_GPL(gmap_shadow_page);
2196 
2197 /*
2198  * gmap_shadow_notify - handle notifications for shadow gmap
2199  *
2200  * Called with sg->parent->shadow_lock.
2201  */
gmap_shadow_notify(struct gmap * sg,unsigned long vmaddr,unsigned long gaddr)2202 static void gmap_shadow_notify(struct gmap *sg, unsigned long vmaddr,
2203 			       unsigned long gaddr)
2204 {
2205 	struct gmap_rmap *rmap, *rnext, *head;
2206 	unsigned long start, end, bits, raddr;
2207 
2208 	BUG_ON(!gmap_is_shadow(sg));
2209 
2210 	spin_lock(&sg->guest_table_lock);
2211 	if (sg->removed) {
2212 		spin_unlock(&sg->guest_table_lock);
2213 		return;
2214 	}
2215 	/* Check for top level table */
2216 	start = sg->orig_asce & _ASCE_ORIGIN;
2217 	end = start + ((sg->orig_asce & _ASCE_TABLE_LENGTH) + 1) * PAGE_SIZE;
2218 	if (!(sg->orig_asce & _ASCE_REAL_SPACE) && gaddr >= start &&
2219 	    gaddr < end) {
2220 		/* The complete shadow table has to go */
2221 		gmap_unshadow(sg);
2222 		spin_unlock(&sg->guest_table_lock);
2223 		list_del(&sg->list);
2224 		gmap_put(sg);
2225 		return;
2226 	}
2227 	/* Remove the page table tree from on specific entry */
2228 	head = radix_tree_delete(&sg->host_to_rmap, vmaddr >> PAGE_SHIFT);
2229 	gmap_for_each_rmap_safe(rmap, rnext, head) {
2230 		bits = rmap->raddr & _SHADOW_RMAP_MASK;
2231 		raddr = rmap->raddr ^ bits;
2232 		switch (bits) {
2233 		case _SHADOW_RMAP_REGION1:
2234 			gmap_unshadow_r2t(sg, raddr);
2235 			break;
2236 		case _SHADOW_RMAP_REGION2:
2237 			gmap_unshadow_r3t(sg, raddr);
2238 			break;
2239 		case _SHADOW_RMAP_REGION3:
2240 			gmap_unshadow_sgt(sg, raddr);
2241 			break;
2242 		case _SHADOW_RMAP_SEGMENT:
2243 			gmap_unshadow_pgt(sg, raddr);
2244 			break;
2245 		case _SHADOW_RMAP_PGTABLE:
2246 			gmap_unshadow_page(sg, raddr);
2247 			break;
2248 		}
2249 		kfree(rmap);
2250 	}
2251 	spin_unlock(&sg->guest_table_lock);
2252 }
2253 
2254 /**
2255  * ptep_notify - call all invalidation callbacks for a specific pte.
2256  * @mm: pointer to the process mm_struct
2257  * @vmaddr: virtual address in the process address space
2258  * @pte: pointer to the page table entry
2259  * @bits: bits from the pgste that caused the notify call
2260  *
2261  * This function is assumed to be called with the page table lock held
2262  * for the pte to notify.
2263  */
ptep_notify(struct mm_struct * mm,unsigned long vmaddr,pte_t * pte,unsigned long bits)2264 void ptep_notify(struct mm_struct *mm, unsigned long vmaddr,
2265 		 pte_t *pte, unsigned long bits)
2266 {
2267 	unsigned long offset, gaddr = 0;
2268 	unsigned long *table;
2269 	struct gmap *gmap, *sg, *next;
2270 
2271 	offset = ((unsigned long) pte) & (255 * sizeof(pte_t));
2272 	offset = offset * (PAGE_SIZE / sizeof(pte_t));
2273 	rcu_read_lock();
2274 	list_for_each_entry_rcu(gmap, &mm->context.gmap_list, list) {
2275 		spin_lock(&gmap->guest_table_lock);
2276 		table = radix_tree_lookup(&gmap->host_to_guest,
2277 					  vmaddr >> PMD_SHIFT);
2278 		if (table)
2279 			gaddr = __gmap_segment_gaddr(table) + offset;
2280 		spin_unlock(&gmap->guest_table_lock);
2281 		if (!table)
2282 			continue;
2283 
2284 		if (!list_empty(&gmap->children) && (bits & PGSTE_VSIE_BIT)) {
2285 			spin_lock(&gmap->shadow_lock);
2286 			list_for_each_entry_safe(sg, next,
2287 						 &gmap->children, list)
2288 				gmap_shadow_notify(sg, vmaddr, gaddr);
2289 			spin_unlock(&gmap->shadow_lock);
2290 		}
2291 		if (bits & PGSTE_IN_BIT)
2292 			gmap_call_notifier(gmap, gaddr, gaddr + PAGE_SIZE - 1);
2293 	}
2294 	rcu_read_unlock();
2295 }
2296 EXPORT_SYMBOL_GPL(ptep_notify);
2297 
pmdp_notify_gmap(struct gmap * gmap,pmd_t * pmdp,unsigned long gaddr)2298 static void pmdp_notify_gmap(struct gmap *gmap, pmd_t *pmdp,
2299 			     unsigned long gaddr)
2300 {
2301 	set_pmd(pmdp, clear_pmd_bit(*pmdp, __pgprot(_SEGMENT_ENTRY_GMAP_IN)));
2302 	gmap_call_notifier(gmap, gaddr, gaddr + HPAGE_SIZE - 1);
2303 }
2304 
2305 /**
2306  * gmap_pmdp_xchg - exchange a gmap pmd with another
2307  * @gmap: pointer to the guest address space structure
2308  * @pmdp: pointer to the pmd entry
2309  * @new: replacement entry
2310  * @gaddr: the affected guest address
2311  *
2312  * This function is assumed to be called with the guest_table_lock
2313  * held.
2314  */
gmap_pmdp_xchg(struct gmap * gmap,pmd_t * pmdp,pmd_t new,unsigned long gaddr)2315 static void gmap_pmdp_xchg(struct gmap *gmap, pmd_t *pmdp, pmd_t new,
2316 			   unsigned long gaddr)
2317 {
2318 	gaddr &= HPAGE_MASK;
2319 	pmdp_notify_gmap(gmap, pmdp, gaddr);
2320 	new = clear_pmd_bit(new, __pgprot(_SEGMENT_ENTRY_GMAP_IN));
2321 	if (MACHINE_HAS_TLB_GUEST)
2322 		__pmdp_idte(gaddr, (pmd_t *)pmdp, IDTE_GUEST_ASCE, gmap->asce,
2323 			    IDTE_GLOBAL);
2324 	else if (MACHINE_HAS_IDTE)
2325 		__pmdp_idte(gaddr, (pmd_t *)pmdp, 0, 0, IDTE_GLOBAL);
2326 	else
2327 		__pmdp_csp(pmdp);
2328 	set_pmd(pmdp, new);
2329 }
2330 
gmap_pmdp_clear(struct mm_struct * mm,unsigned long vmaddr,int purge)2331 static void gmap_pmdp_clear(struct mm_struct *mm, unsigned long vmaddr,
2332 			    int purge)
2333 {
2334 	pmd_t *pmdp;
2335 	struct gmap *gmap;
2336 	unsigned long gaddr;
2337 
2338 	rcu_read_lock();
2339 	list_for_each_entry_rcu(gmap, &mm->context.gmap_list, list) {
2340 		spin_lock(&gmap->guest_table_lock);
2341 		pmdp = (pmd_t *)radix_tree_delete(&gmap->host_to_guest,
2342 						  vmaddr >> PMD_SHIFT);
2343 		if (pmdp) {
2344 			gaddr = __gmap_segment_gaddr((unsigned long *)pmdp);
2345 			pmdp_notify_gmap(gmap, pmdp, gaddr);
2346 			WARN_ON(pmd_val(*pmdp) & ~(_SEGMENT_ENTRY_HARDWARE_BITS_LARGE |
2347 						   _SEGMENT_ENTRY_GMAP_UC));
2348 			if (purge)
2349 				__pmdp_csp(pmdp);
2350 			set_pmd(pmdp, __pmd(_SEGMENT_ENTRY_EMPTY));
2351 		}
2352 		spin_unlock(&gmap->guest_table_lock);
2353 	}
2354 	rcu_read_unlock();
2355 }
2356 
2357 /**
2358  * gmap_pmdp_invalidate - invalidate all affected guest pmd entries without
2359  *                        flushing
2360  * @mm: pointer to the process mm_struct
2361  * @vmaddr: virtual address in the process address space
2362  */
gmap_pmdp_invalidate(struct mm_struct * mm,unsigned long vmaddr)2363 void gmap_pmdp_invalidate(struct mm_struct *mm, unsigned long vmaddr)
2364 {
2365 	gmap_pmdp_clear(mm, vmaddr, 0);
2366 }
2367 EXPORT_SYMBOL_GPL(gmap_pmdp_invalidate);
2368 
2369 /**
2370  * gmap_pmdp_csp - csp all affected guest pmd entries
2371  * @mm: pointer to the process mm_struct
2372  * @vmaddr: virtual address in the process address space
2373  */
gmap_pmdp_csp(struct mm_struct * mm,unsigned long vmaddr)2374 void gmap_pmdp_csp(struct mm_struct *mm, unsigned long vmaddr)
2375 {
2376 	gmap_pmdp_clear(mm, vmaddr, 1);
2377 }
2378 EXPORT_SYMBOL_GPL(gmap_pmdp_csp);
2379 
2380 /**
2381  * gmap_pmdp_idte_local - invalidate and clear a guest pmd entry
2382  * @mm: pointer to the process mm_struct
2383  * @vmaddr: virtual address in the process address space
2384  */
gmap_pmdp_idte_local(struct mm_struct * mm,unsigned long vmaddr)2385 void gmap_pmdp_idte_local(struct mm_struct *mm, unsigned long vmaddr)
2386 {
2387 	unsigned long *entry, gaddr;
2388 	struct gmap *gmap;
2389 	pmd_t *pmdp;
2390 
2391 	rcu_read_lock();
2392 	list_for_each_entry_rcu(gmap, &mm->context.gmap_list, list) {
2393 		spin_lock(&gmap->guest_table_lock);
2394 		entry = radix_tree_delete(&gmap->host_to_guest,
2395 					  vmaddr >> PMD_SHIFT);
2396 		if (entry) {
2397 			pmdp = (pmd_t *)entry;
2398 			gaddr = __gmap_segment_gaddr(entry);
2399 			pmdp_notify_gmap(gmap, pmdp, gaddr);
2400 			WARN_ON(*entry & ~(_SEGMENT_ENTRY_HARDWARE_BITS_LARGE |
2401 					   _SEGMENT_ENTRY_GMAP_UC));
2402 			if (MACHINE_HAS_TLB_GUEST)
2403 				__pmdp_idte(gaddr, pmdp, IDTE_GUEST_ASCE,
2404 					    gmap->asce, IDTE_LOCAL);
2405 			else if (MACHINE_HAS_IDTE)
2406 				__pmdp_idte(gaddr, pmdp, 0, 0, IDTE_LOCAL);
2407 			*entry = _SEGMENT_ENTRY_EMPTY;
2408 		}
2409 		spin_unlock(&gmap->guest_table_lock);
2410 	}
2411 	rcu_read_unlock();
2412 }
2413 EXPORT_SYMBOL_GPL(gmap_pmdp_idte_local);
2414 
2415 /**
2416  * gmap_pmdp_idte_global - invalidate and clear a guest pmd entry
2417  * @mm: pointer to the process mm_struct
2418  * @vmaddr: virtual address in the process address space
2419  */
gmap_pmdp_idte_global(struct mm_struct * mm,unsigned long vmaddr)2420 void gmap_pmdp_idte_global(struct mm_struct *mm, unsigned long vmaddr)
2421 {
2422 	unsigned long *entry, gaddr;
2423 	struct gmap *gmap;
2424 	pmd_t *pmdp;
2425 
2426 	rcu_read_lock();
2427 	list_for_each_entry_rcu(gmap, &mm->context.gmap_list, list) {
2428 		spin_lock(&gmap->guest_table_lock);
2429 		entry = radix_tree_delete(&gmap->host_to_guest,
2430 					  vmaddr >> PMD_SHIFT);
2431 		if (entry) {
2432 			pmdp = (pmd_t *)entry;
2433 			gaddr = __gmap_segment_gaddr(entry);
2434 			pmdp_notify_gmap(gmap, pmdp, gaddr);
2435 			WARN_ON(*entry & ~(_SEGMENT_ENTRY_HARDWARE_BITS_LARGE |
2436 					   _SEGMENT_ENTRY_GMAP_UC));
2437 			if (MACHINE_HAS_TLB_GUEST)
2438 				__pmdp_idte(gaddr, pmdp, IDTE_GUEST_ASCE,
2439 					    gmap->asce, IDTE_GLOBAL);
2440 			else if (MACHINE_HAS_IDTE)
2441 				__pmdp_idte(gaddr, pmdp, 0, 0, IDTE_GLOBAL);
2442 			else
2443 				__pmdp_csp(pmdp);
2444 			*entry = _SEGMENT_ENTRY_EMPTY;
2445 		}
2446 		spin_unlock(&gmap->guest_table_lock);
2447 	}
2448 	rcu_read_unlock();
2449 }
2450 EXPORT_SYMBOL_GPL(gmap_pmdp_idte_global);
2451 
2452 /**
2453  * gmap_test_and_clear_dirty_pmd - test and reset segment dirty status
2454  * @gmap: pointer to guest address space
2455  * @pmdp: pointer to the pmd to be tested
2456  * @gaddr: virtual address in the guest address space
2457  *
2458  * This function is assumed to be called with the guest_table_lock
2459  * held.
2460  */
gmap_test_and_clear_dirty_pmd(struct gmap * gmap,pmd_t * pmdp,unsigned long gaddr)2461 static bool gmap_test_and_clear_dirty_pmd(struct gmap *gmap, pmd_t *pmdp,
2462 					  unsigned long gaddr)
2463 {
2464 	if (pmd_val(*pmdp) & _SEGMENT_ENTRY_INVALID)
2465 		return false;
2466 
2467 	/* Already protected memory, which did not change is clean */
2468 	if (pmd_val(*pmdp) & _SEGMENT_ENTRY_PROTECT &&
2469 	    !(pmd_val(*pmdp) & _SEGMENT_ENTRY_GMAP_UC))
2470 		return false;
2471 
2472 	/* Clear UC indication and reset protection */
2473 	set_pmd(pmdp, clear_pmd_bit(*pmdp, __pgprot(_SEGMENT_ENTRY_GMAP_UC)));
2474 	gmap_protect_pmd(gmap, gaddr, pmdp, PROT_READ, 0);
2475 	return true;
2476 }
2477 
2478 /**
2479  * gmap_sync_dirty_log_pmd - set bitmap based on dirty status of segment
2480  * @gmap: pointer to guest address space
2481  * @bitmap: dirty bitmap for this pmd
2482  * @gaddr: virtual address in the guest address space
2483  * @vmaddr: virtual address in the host address space
2484  *
2485  * This function is assumed to be called with the guest_table_lock
2486  * held.
2487  */
gmap_sync_dirty_log_pmd(struct gmap * gmap,unsigned long bitmap[4],unsigned long gaddr,unsigned long vmaddr)2488 void gmap_sync_dirty_log_pmd(struct gmap *gmap, unsigned long bitmap[4],
2489 			     unsigned long gaddr, unsigned long vmaddr)
2490 {
2491 	int i;
2492 	pmd_t *pmdp;
2493 	pte_t *ptep;
2494 	spinlock_t *ptl;
2495 
2496 	pmdp = gmap_pmd_op_walk(gmap, gaddr);
2497 	if (!pmdp)
2498 		return;
2499 
2500 	if (pmd_large(*pmdp)) {
2501 		if (gmap_test_and_clear_dirty_pmd(gmap, pmdp, gaddr))
2502 			bitmap_fill(bitmap, _PAGE_ENTRIES);
2503 	} else {
2504 		for (i = 0; i < _PAGE_ENTRIES; i++, vmaddr += PAGE_SIZE) {
2505 			ptep = pte_alloc_map_lock(gmap->mm, pmdp, vmaddr, &ptl);
2506 			if (!ptep)
2507 				continue;
2508 			if (ptep_test_and_clear_uc(gmap->mm, vmaddr, ptep))
2509 				set_bit(i, bitmap);
2510 			pte_unmap_unlock(ptep, ptl);
2511 		}
2512 	}
2513 	gmap_pmd_op_end(gmap, pmdp);
2514 }
2515 EXPORT_SYMBOL_GPL(gmap_sync_dirty_log_pmd);
2516 
2517 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
thp_split_walk_pmd_entry(pmd_t * pmd,unsigned long addr,unsigned long end,struct mm_walk * walk)2518 static int thp_split_walk_pmd_entry(pmd_t *pmd, unsigned long addr,
2519 				    unsigned long end, struct mm_walk *walk)
2520 {
2521 	struct vm_area_struct *vma = walk->vma;
2522 
2523 	split_huge_pmd(vma, pmd, addr);
2524 	return 0;
2525 }
2526 
2527 static const struct mm_walk_ops thp_split_walk_ops = {
2528 	.pmd_entry	= thp_split_walk_pmd_entry,
2529 	.walk_lock	= PGWALK_WRLOCK_VERIFY,
2530 };
2531 
thp_split_mm(struct mm_struct * mm)2532 static inline void thp_split_mm(struct mm_struct *mm)
2533 {
2534 	struct vm_area_struct *vma;
2535 	VMA_ITERATOR(vmi, mm, 0);
2536 
2537 	for_each_vma(vmi, vma) {
2538 		vm_flags_mod(vma, VM_NOHUGEPAGE, VM_HUGEPAGE);
2539 		walk_page_vma(vma, &thp_split_walk_ops, NULL);
2540 	}
2541 	mm->def_flags |= VM_NOHUGEPAGE;
2542 }
2543 #else
thp_split_mm(struct mm_struct * mm)2544 static inline void thp_split_mm(struct mm_struct *mm)
2545 {
2546 }
2547 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
2548 
2549 /*
2550  * Remove all empty zero pages from the mapping for lazy refaulting
2551  * - This must be called after mm->context.has_pgste is set, to avoid
2552  *   future creation of zero pages
2553  * - This must be called after THP was disabled.
2554  *
2555  * mm contracts with s390, that even if mm were to remove a page table,
2556  * racing with the loop below and so causing pte_offset_map_lock() to fail,
2557  * it will never insert a page table containing empty zero pages once
2558  * mm_forbids_zeropage(mm) i.e. mm->context.has_pgste is set.
2559  */
__zap_zero_pages(pmd_t * pmd,unsigned long start,unsigned long end,struct mm_walk * walk)2560 static int __zap_zero_pages(pmd_t *pmd, unsigned long start,
2561 			   unsigned long end, struct mm_walk *walk)
2562 {
2563 	unsigned long addr;
2564 
2565 	for (addr = start; addr != end; addr += PAGE_SIZE) {
2566 		pte_t *ptep;
2567 		spinlock_t *ptl;
2568 
2569 		ptep = pte_offset_map_lock(walk->mm, pmd, addr, &ptl);
2570 		if (!ptep)
2571 			break;
2572 		if (is_zero_pfn(pte_pfn(*ptep)))
2573 			ptep_xchg_direct(walk->mm, addr, ptep, __pte(_PAGE_INVALID));
2574 		pte_unmap_unlock(ptep, ptl);
2575 	}
2576 	return 0;
2577 }
2578 
2579 static const struct mm_walk_ops zap_zero_walk_ops = {
2580 	.pmd_entry	= __zap_zero_pages,
2581 	.walk_lock	= PGWALK_WRLOCK,
2582 };
2583 
2584 /*
2585  * switch on pgstes for its userspace process (for kvm)
2586  */
s390_enable_sie(void)2587 int s390_enable_sie(void)
2588 {
2589 	struct mm_struct *mm = current->mm;
2590 
2591 	/* Do we have pgstes? if yes, we are done */
2592 	if (mm_has_pgste(mm))
2593 		return 0;
2594 	/* Fail if the page tables are 2K */
2595 	if (!mm_alloc_pgste(mm))
2596 		return -EINVAL;
2597 	mmap_write_lock(mm);
2598 	mm->context.has_pgste = 1;
2599 	/* split thp mappings and disable thp for future mappings */
2600 	thp_split_mm(mm);
2601 	walk_page_range(mm, 0, TASK_SIZE, &zap_zero_walk_ops, NULL);
2602 	mmap_write_unlock(mm);
2603 	return 0;
2604 }
2605 EXPORT_SYMBOL_GPL(s390_enable_sie);
2606 
gmap_mark_unmergeable(void)2607 int gmap_mark_unmergeable(void)
2608 {
2609 	/*
2610 	 * Make sure to disable KSM (if enabled for the whole process or
2611 	 * individual VMAs). Note that nothing currently hinders user space
2612 	 * from re-enabling it.
2613 	 */
2614 	return ksm_disable(current->mm);
2615 }
2616 EXPORT_SYMBOL_GPL(gmap_mark_unmergeable);
2617 
2618 /*
2619  * Enable storage key handling from now on and initialize the storage
2620  * keys with the default key.
2621  */
__s390_enable_skey_pte(pte_t * pte,unsigned long addr,unsigned long next,struct mm_walk * walk)2622 static int __s390_enable_skey_pte(pte_t *pte, unsigned long addr,
2623 				  unsigned long next, struct mm_walk *walk)
2624 {
2625 	/* Clear storage key */
2626 	ptep_zap_key(walk->mm, addr, pte);
2627 	return 0;
2628 }
2629 
2630 /*
2631  * Give a chance to schedule after setting a key to 256 pages.
2632  * We only hold the mm lock, which is a rwsem and the kvm srcu.
2633  * Both can sleep.
2634  */
__s390_enable_skey_pmd(pmd_t * pmd,unsigned long addr,unsigned long next,struct mm_walk * walk)2635 static int __s390_enable_skey_pmd(pmd_t *pmd, unsigned long addr,
2636 				  unsigned long next, struct mm_walk *walk)
2637 {
2638 	cond_resched();
2639 	return 0;
2640 }
2641 
__s390_enable_skey_hugetlb(pte_t * pte,unsigned long addr,unsigned long hmask,unsigned long next,struct mm_walk * walk)2642 static int __s390_enable_skey_hugetlb(pte_t *pte, unsigned long addr,
2643 				      unsigned long hmask, unsigned long next,
2644 				      struct mm_walk *walk)
2645 {
2646 	pmd_t *pmd = (pmd_t *)pte;
2647 	unsigned long start, end;
2648 	struct page *page = pmd_page(*pmd);
2649 
2650 	/*
2651 	 * The write check makes sure we do not set a key on shared
2652 	 * memory. This is needed as the walker does not differentiate
2653 	 * between actual guest memory and the process executable or
2654 	 * shared libraries.
2655 	 */
2656 	if (pmd_val(*pmd) & _SEGMENT_ENTRY_INVALID ||
2657 	    !(pmd_val(*pmd) & _SEGMENT_ENTRY_WRITE))
2658 		return 0;
2659 
2660 	start = pmd_val(*pmd) & HPAGE_MASK;
2661 	end = start + HPAGE_SIZE - 1;
2662 	__storage_key_init_range(start, end);
2663 	set_bit(PG_arch_1, &page->flags);
2664 	cond_resched();
2665 	return 0;
2666 }
2667 
2668 static const struct mm_walk_ops enable_skey_walk_ops = {
2669 	.hugetlb_entry		= __s390_enable_skey_hugetlb,
2670 	.pte_entry		= __s390_enable_skey_pte,
2671 	.pmd_entry		= __s390_enable_skey_pmd,
2672 	.walk_lock		= PGWALK_WRLOCK,
2673 };
2674 
s390_enable_skey(void)2675 int s390_enable_skey(void)
2676 {
2677 	struct mm_struct *mm = current->mm;
2678 	int rc = 0;
2679 
2680 	mmap_write_lock(mm);
2681 	if (mm_uses_skeys(mm))
2682 		goto out_up;
2683 
2684 	mm->context.uses_skeys = 1;
2685 	rc = gmap_mark_unmergeable();
2686 	if (rc) {
2687 		mm->context.uses_skeys = 0;
2688 		goto out_up;
2689 	}
2690 	walk_page_range(mm, 0, TASK_SIZE, &enable_skey_walk_ops, NULL);
2691 
2692 out_up:
2693 	mmap_write_unlock(mm);
2694 	return rc;
2695 }
2696 EXPORT_SYMBOL_GPL(s390_enable_skey);
2697 
2698 /*
2699  * Reset CMMA state, make all pages stable again.
2700  */
__s390_reset_cmma(pte_t * pte,unsigned long addr,unsigned long next,struct mm_walk * walk)2701 static int __s390_reset_cmma(pte_t *pte, unsigned long addr,
2702 			     unsigned long next, struct mm_walk *walk)
2703 {
2704 	ptep_zap_unused(walk->mm, addr, pte, 1);
2705 	return 0;
2706 }
2707 
2708 static const struct mm_walk_ops reset_cmma_walk_ops = {
2709 	.pte_entry		= __s390_reset_cmma,
2710 	.walk_lock		= PGWALK_WRLOCK,
2711 };
2712 
s390_reset_cmma(struct mm_struct * mm)2713 void s390_reset_cmma(struct mm_struct *mm)
2714 {
2715 	mmap_write_lock(mm);
2716 	walk_page_range(mm, 0, TASK_SIZE, &reset_cmma_walk_ops, NULL);
2717 	mmap_write_unlock(mm);
2718 }
2719 EXPORT_SYMBOL_GPL(s390_reset_cmma);
2720 
2721 #define GATHER_GET_PAGES 32
2722 
2723 struct reset_walk_state {
2724 	unsigned long next;
2725 	unsigned long count;
2726 	unsigned long pfns[GATHER_GET_PAGES];
2727 };
2728 
s390_gather_pages(pte_t * ptep,unsigned long addr,unsigned long next,struct mm_walk * walk)2729 static int s390_gather_pages(pte_t *ptep, unsigned long addr,
2730 			     unsigned long next, struct mm_walk *walk)
2731 {
2732 	struct reset_walk_state *p = walk->private;
2733 	pte_t pte = READ_ONCE(*ptep);
2734 
2735 	if (pte_present(pte)) {
2736 		/* we have a reference from the mapping, take an extra one */
2737 		get_page(phys_to_page(pte_val(pte)));
2738 		p->pfns[p->count] = phys_to_pfn(pte_val(pte));
2739 		p->next = next;
2740 		p->count++;
2741 	}
2742 	return p->count >= GATHER_GET_PAGES;
2743 }
2744 
2745 static const struct mm_walk_ops gather_pages_ops = {
2746 	.pte_entry = s390_gather_pages,
2747 	.walk_lock = PGWALK_RDLOCK,
2748 };
2749 
2750 /*
2751  * Call the Destroy secure page UVC on each page in the given array of PFNs.
2752  * Each page needs to have an extra reference, which will be released here.
2753  */
s390_uv_destroy_pfns(unsigned long count,unsigned long * pfns)2754 void s390_uv_destroy_pfns(unsigned long count, unsigned long *pfns)
2755 {
2756 	unsigned long i;
2757 
2758 	for (i = 0; i < count; i++) {
2759 		/* we always have an extra reference */
2760 		uv_destroy_owned_page(pfn_to_phys(pfns[i]));
2761 		/* get rid of the extra reference */
2762 		put_page(pfn_to_page(pfns[i]));
2763 		cond_resched();
2764 	}
2765 }
2766 EXPORT_SYMBOL_GPL(s390_uv_destroy_pfns);
2767 
2768 /**
2769  * __s390_uv_destroy_range - Call the destroy secure page UVC on each page
2770  * in the given range of the given address space.
2771  * @mm: the mm to operate on
2772  * @start: the start of the range
2773  * @end: the end of the range
2774  * @interruptible: if not 0, stop when a fatal signal is received
2775  *
2776  * Walk the given range of the given address space and call the destroy
2777  * secure page UVC on each page. Optionally exit early if a fatal signal is
2778  * pending.
2779  *
2780  * Return: 0 on success, -EINTR if the function stopped before completing
2781  */
__s390_uv_destroy_range(struct mm_struct * mm,unsigned long start,unsigned long end,bool interruptible)2782 int __s390_uv_destroy_range(struct mm_struct *mm, unsigned long start,
2783 			    unsigned long end, bool interruptible)
2784 {
2785 	struct reset_walk_state state = { .next = start };
2786 	int r = 1;
2787 
2788 	while (r > 0) {
2789 		state.count = 0;
2790 		mmap_read_lock(mm);
2791 		r = walk_page_range(mm, state.next, end, &gather_pages_ops, &state);
2792 		mmap_read_unlock(mm);
2793 		cond_resched();
2794 		s390_uv_destroy_pfns(state.count, state.pfns);
2795 		if (interruptible && fatal_signal_pending(current))
2796 			return -EINTR;
2797 	}
2798 	return 0;
2799 }
2800 EXPORT_SYMBOL_GPL(__s390_uv_destroy_range);
2801 
2802 /**
2803  * s390_unlist_old_asce - Remove the topmost level of page tables from the
2804  * list of page tables of the gmap.
2805  * @gmap: the gmap whose table is to be removed
2806  *
2807  * On s390x, KVM keeps a list of all pages containing the page tables of the
2808  * gmap (the CRST list). This list is used at tear down time to free all
2809  * pages that are now not needed anymore.
2810  *
2811  * This function removes the topmost page of the tree (the one pointed to by
2812  * the ASCE) from the CRST list.
2813  *
2814  * This means that it will not be freed when the VM is torn down, and needs
2815  * to be handled separately by the caller, unless a leak is actually
2816  * intended. Notice that this function will only remove the page from the
2817  * list, the page will still be used as a top level page table (and ASCE).
2818  */
s390_unlist_old_asce(struct gmap * gmap)2819 void s390_unlist_old_asce(struct gmap *gmap)
2820 {
2821 	struct page *old;
2822 
2823 	old = virt_to_page(gmap->table);
2824 	spin_lock(&gmap->guest_table_lock);
2825 	list_del(&old->lru);
2826 	/*
2827 	 * Sometimes the topmost page might need to be "removed" multiple
2828 	 * times, for example if the VM is rebooted into secure mode several
2829 	 * times concurrently, or if s390_replace_asce fails after calling
2830 	 * s390_remove_old_asce and is attempted again later. In that case
2831 	 * the old asce has been removed from the list, and therefore it
2832 	 * will not be freed when the VM terminates, but the ASCE is still
2833 	 * in use and still pointed to.
2834 	 * A subsequent call to replace_asce will follow the pointer and try
2835 	 * to remove the same page from the list again.
2836 	 * Therefore it's necessary that the page of the ASCE has valid
2837 	 * pointers, so list_del can work (and do nothing) without
2838 	 * dereferencing stale or invalid pointers.
2839 	 */
2840 	INIT_LIST_HEAD(&old->lru);
2841 	spin_unlock(&gmap->guest_table_lock);
2842 }
2843 EXPORT_SYMBOL_GPL(s390_unlist_old_asce);
2844 
2845 /**
2846  * s390_replace_asce - Try to replace the current ASCE of a gmap with a copy
2847  * @gmap: the gmap whose ASCE needs to be replaced
2848  *
2849  * If the ASCE is a SEGMENT type then this function will return -EINVAL,
2850  * otherwise the pointers in the host_to_guest radix tree will keep pointing
2851  * to the wrong pages, causing use-after-free and memory corruption.
2852  * If the allocation of the new top level page table fails, the ASCE is not
2853  * replaced.
2854  * In any case, the old ASCE is always removed from the gmap CRST list.
2855  * Therefore the caller has to make sure to save a pointer to it
2856  * beforehand, unless a leak is actually intended.
2857  */
s390_replace_asce(struct gmap * gmap)2858 int s390_replace_asce(struct gmap *gmap)
2859 {
2860 	unsigned long asce;
2861 	struct page *page;
2862 	void *table;
2863 
2864 	s390_unlist_old_asce(gmap);
2865 
2866 	/* Replacing segment type ASCEs would cause serious issues */
2867 	if ((gmap->asce & _ASCE_TYPE_MASK) == _ASCE_TYPE_SEGMENT)
2868 		return -EINVAL;
2869 
2870 	page = gmap_alloc_crst();
2871 	if (!page)
2872 		return -ENOMEM;
2873 	page->index = 0;
2874 	table = page_to_virt(page);
2875 	memcpy(table, gmap->table, 1UL << (CRST_ALLOC_ORDER + PAGE_SHIFT));
2876 
2877 	/*
2878 	 * The caller has to deal with the old ASCE, but here we make sure
2879 	 * the new one is properly added to the CRST list, so that
2880 	 * it will be freed when the VM is torn down.
2881 	 */
2882 	spin_lock(&gmap->guest_table_lock);
2883 	list_add(&page->lru, &gmap->crst_list);
2884 	spin_unlock(&gmap->guest_table_lock);
2885 
2886 	/* Set new table origin while preserving existing ASCE control bits */
2887 	asce = (gmap->asce & ~_ASCE_ORIGIN) | __pa(table);
2888 	WRITE_ONCE(gmap->asce, asce);
2889 	WRITE_ONCE(gmap->mm->context.gmap_asce, asce);
2890 	WRITE_ONCE(gmap->table, table);
2891 
2892 	return 0;
2893 }
2894 EXPORT_SYMBOL_GPL(s390_replace_asce);
2895