xref: /openbmc/linux/arch/s390/mm/gmap.c (revision 840565b1)
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_leaf(*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->private = parent->private;
1695 	new->orig_asce = asce;
1696 	new->edat_level = edat_level;
1697 	new->initialized = false;
1698 	spin_lock(&parent->shadow_lock);
1699 	/* Recheck if another CPU created the same shadow */
1700 	sg = gmap_find_shadow(parent, asce, edat_level);
1701 	if (sg) {
1702 		spin_unlock(&parent->shadow_lock);
1703 		gmap_free(new);
1704 		return sg;
1705 	}
1706 	if (asce & _ASCE_REAL_SPACE) {
1707 		/* only allow one real-space gmap shadow */
1708 		list_for_each_entry(sg, &parent->children, list) {
1709 			if (sg->orig_asce & _ASCE_REAL_SPACE) {
1710 				spin_lock(&sg->guest_table_lock);
1711 				gmap_unshadow(sg);
1712 				spin_unlock(&sg->guest_table_lock);
1713 				list_del(&sg->list);
1714 				gmap_put(sg);
1715 				break;
1716 			}
1717 		}
1718 	}
1719 	refcount_set(&new->ref_count, 2);
1720 	list_add(&new->list, &parent->children);
1721 	if (asce & _ASCE_REAL_SPACE) {
1722 		/* nothing to protect, return right away */
1723 		new->initialized = true;
1724 		spin_unlock(&parent->shadow_lock);
1725 		return new;
1726 	}
1727 	spin_unlock(&parent->shadow_lock);
1728 	/* protect after insertion, so it will get properly invalidated */
1729 	mmap_read_lock(parent->mm);
1730 	rc = gmap_protect_range(parent, asce & _ASCE_ORIGIN,
1731 				((asce & _ASCE_TABLE_LENGTH) + 1) * PAGE_SIZE,
1732 				PROT_READ, GMAP_NOTIFY_SHADOW);
1733 	mmap_read_unlock(parent->mm);
1734 	spin_lock(&parent->shadow_lock);
1735 	new->initialized = true;
1736 	if (rc) {
1737 		list_del(&new->list);
1738 		gmap_free(new);
1739 		new = ERR_PTR(rc);
1740 	}
1741 	spin_unlock(&parent->shadow_lock);
1742 	return new;
1743 }
1744 EXPORT_SYMBOL_GPL(gmap_shadow);
1745 
1746 /**
1747  * gmap_shadow_r2t - create an empty shadow region 2 table
1748  * @sg: pointer to the shadow guest address space structure
1749  * @saddr: faulting address in the shadow gmap
1750  * @r2t: parent gmap address of the region 2 table to get shadowed
1751  * @fake: r2t references contiguous guest memory block, not a r2t
1752  *
1753  * The r2t parameter specifies the address of the source table. The
1754  * four pages of the source table are made read-only in the parent gmap
1755  * address space. A write to the source table area @r2t will automatically
1756  * remove the shadow r2 table and all of its descendants.
1757  *
1758  * Returns 0 if successfully shadowed or already shadowed, -EAGAIN if the
1759  * shadow table structure is incomplete, -ENOMEM if out of memory and
1760  * -EFAULT if an address in the parent gmap could not be resolved.
1761  *
1762  * Called with sg->mm->mmap_lock in read.
1763  */
gmap_shadow_r2t(struct gmap * sg,unsigned long saddr,unsigned long r2t,int fake)1764 int gmap_shadow_r2t(struct gmap *sg, unsigned long saddr, unsigned long r2t,
1765 		    int fake)
1766 {
1767 	unsigned long raddr, origin, offset, len;
1768 	unsigned long *table;
1769 	phys_addr_t s_r2t;
1770 	struct page *page;
1771 	int rc;
1772 
1773 	BUG_ON(!gmap_is_shadow(sg));
1774 	/* Allocate a shadow region second table */
1775 	page = gmap_alloc_crst();
1776 	if (!page)
1777 		return -ENOMEM;
1778 	page->index = r2t & _REGION_ENTRY_ORIGIN;
1779 	if (fake)
1780 		page->index |= GMAP_SHADOW_FAKE_TABLE;
1781 	s_r2t = page_to_phys(page);
1782 	/* Install shadow region second table */
1783 	spin_lock(&sg->guest_table_lock);
1784 	table = gmap_table_walk(sg, saddr, 4); /* get region-1 pointer */
1785 	if (!table) {
1786 		rc = -EAGAIN;		/* Race with unshadow */
1787 		goto out_free;
1788 	}
1789 	if (!(*table & _REGION_ENTRY_INVALID)) {
1790 		rc = 0;			/* Already established */
1791 		goto out_free;
1792 	} else if (*table & _REGION_ENTRY_ORIGIN) {
1793 		rc = -EAGAIN;		/* Race with shadow */
1794 		goto out_free;
1795 	}
1796 	crst_table_init(__va(s_r2t), _REGION2_ENTRY_EMPTY);
1797 	/* mark as invalid as long as the parent table is not protected */
1798 	*table = s_r2t | _REGION_ENTRY_LENGTH |
1799 		 _REGION_ENTRY_TYPE_R1 | _REGION_ENTRY_INVALID;
1800 	if (sg->edat_level >= 1)
1801 		*table |= (r2t & _REGION_ENTRY_PROTECT);
1802 	list_add(&page->lru, &sg->crst_list);
1803 	if (fake) {
1804 		/* nothing to protect for fake tables */
1805 		*table &= ~_REGION_ENTRY_INVALID;
1806 		spin_unlock(&sg->guest_table_lock);
1807 		return 0;
1808 	}
1809 	spin_unlock(&sg->guest_table_lock);
1810 	/* Make r2t read-only in parent gmap page table */
1811 	raddr = (saddr & _REGION1_MASK) | _SHADOW_RMAP_REGION1;
1812 	origin = r2t & _REGION_ENTRY_ORIGIN;
1813 	offset = ((r2t & _REGION_ENTRY_OFFSET) >> 6) * PAGE_SIZE;
1814 	len = ((r2t & _REGION_ENTRY_LENGTH) + 1) * PAGE_SIZE - offset;
1815 	rc = gmap_protect_rmap(sg, raddr, origin + offset, len);
1816 	spin_lock(&sg->guest_table_lock);
1817 	if (!rc) {
1818 		table = gmap_table_walk(sg, saddr, 4);
1819 		if (!table || (*table & _REGION_ENTRY_ORIGIN) != s_r2t)
1820 			rc = -EAGAIN;		/* Race with unshadow */
1821 		else
1822 			*table &= ~_REGION_ENTRY_INVALID;
1823 	} else {
1824 		gmap_unshadow_r2t(sg, raddr);
1825 	}
1826 	spin_unlock(&sg->guest_table_lock);
1827 	return rc;
1828 out_free:
1829 	spin_unlock(&sg->guest_table_lock);
1830 	__free_pages(page, CRST_ALLOC_ORDER);
1831 	return rc;
1832 }
1833 EXPORT_SYMBOL_GPL(gmap_shadow_r2t);
1834 
1835 /**
1836  * gmap_shadow_r3t - create a shadow region 3 table
1837  * @sg: pointer to the shadow guest address space structure
1838  * @saddr: faulting address in the shadow gmap
1839  * @r3t: parent gmap address of the region 3 table to get shadowed
1840  * @fake: r3t references contiguous guest memory block, not a r3t
1841  *
1842  * Returns 0 if successfully shadowed or already shadowed, -EAGAIN if the
1843  * shadow table structure is incomplete, -ENOMEM if out of memory and
1844  * -EFAULT if an address in the parent gmap could not be resolved.
1845  *
1846  * Called with sg->mm->mmap_lock in read.
1847  */
gmap_shadow_r3t(struct gmap * sg,unsigned long saddr,unsigned long r3t,int fake)1848 int gmap_shadow_r3t(struct gmap *sg, unsigned long saddr, unsigned long r3t,
1849 		    int fake)
1850 {
1851 	unsigned long raddr, origin, offset, len;
1852 	unsigned long *table;
1853 	phys_addr_t s_r3t;
1854 	struct page *page;
1855 	int rc;
1856 
1857 	BUG_ON(!gmap_is_shadow(sg));
1858 	/* Allocate a shadow region second table */
1859 	page = gmap_alloc_crst();
1860 	if (!page)
1861 		return -ENOMEM;
1862 	page->index = r3t & _REGION_ENTRY_ORIGIN;
1863 	if (fake)
1864 		page->index |= GMAP_SHADOW_FAKE_TABLE;
1865 	s_r3t = page_to_phys(page);
1866 	/* Install shadow region second table */
1867 	spin_lock(&sg->guest_table_lock);
1868 	table = gmap_table_walk(sg, saddr, 3); /* get region-2 pointer */
1869 	if (!table) {
1870 		rc = -EAGAIN;		/* Race with unshadow */
1871 		goto out_free;
1872 	}
1873 	if (!(*table & _REGION_ENTRY_INVALID)) {
1874 		rc = 0;			/* Already established */
1875 		goto out_free;
1876 	} else if (*table & _REGION_ENTRY_ORIGIN) {
1877 		rc = -EAGAIN;		/* Race with shadow */
1878 		goto out_free;
1879 	}
1880 	crst_table_init(__va(s_r3t), _REGION3_ENTRY_EMPTY);
1881 	/* mark as invalid as long as the parent table is not protected */
1882 	*table = s_r3t | _REGION_ENTRY_LENGTH |
1883 		 _REGION_ENTRY_TYPE_R2 | _REGION_ENTRY_INVALID;
1884 	if (sg->edat_level >= 1)
1885 		*table |= (r3t & _REGION_ENTRY_PROTECT);
1886 	list_add(&page->lru, &sg->crst_list);
1887 	if (fake) {
1888 		/* nothing to protect for fake tables */
1889 		*table &= ~_REGION_ENTRY_INVALID;
1890 		spin_unlock(&sg->guest_table_lock);
1891 		return 0;
1892 	}
1893 	spin_unlock(&sg->guest_table_lock);
1894 	/* Make r3t read-only in parent gmap page table */
1895 	raddr = (saddr & _REGION2_MASK) | _SHADOW_RMAP_REGION2;
1896 	origin = r3t & _REGION_ENTRY_ORIGIN;
1897 	offset = ((r3t & _REGION_ENTRY_OFFSET) >> 6) * PAGE_SIZE;
1898 	len = ((r3t & _REGION_ENTRY_LENGTH) + 1) * PAGE_SIZE - offset;
1899 	rc = gmap_protect_rmap(sg, raddr, origin + offset, len);
1900 	spin_lock(&sg->guest_table_lock);
1901 	if (!rc) {
1902 		table = gmap_table_walk(sg, saddr, 3);
1903 		if (!table || (*table & _REGION_ENTRY_ORIGIN) != s_r3t)
1904 			rc = -EAGAIN;		/* Race with unshadow */
1905 		else
1906 			*table &= ~_REGION_ENTRY_INVALID;
1907 	} else {
1908 		gmap_unshadow_r3t(sg, raddr);
1909 	}
1910 	spin_unlock(&sg->guest_table_lock);
1911 	return rc;
1912 out_free:
1913 	spin_unlock(&sg->guest_table_lock);
1914 	__free_pages(page, CRST_ALLOC_ORDER);
1915 	return rc;
1916 }
1917 EXPORT_SYMBOL_GPL(gmap_shadow_r3t);
1918 
1919 /**
1920  * gmap_shadow_sgt - create a shadow segment table
1921  * @sg: pointer to the shadow guest address space structure
1922  * @saddr: faulting address in the shadow gmap
1923  * @sgt: parent gmap address of the segment table to get shadowed
1924  * @fake: sgt references contiguous guest memory block, not a sgt
1925  *
1926  * Returns: 0 if successfully shadowed or already shadowed, -EAGAIN if the
1927  * shadow table structure is incomplete, -ENOMEM if out of memory and
1928  * -EFAULT if an address in the parent gmap could not be resolved.
1929  *
1930  * Called with sg->mm->mmap_lock in read.
1931  */
gmap_shadow_sgt(struct gmap * sg,unsigned long saddr,unsigned long sgt,int fake)1932 int gmap_shadow_sgt(struct gmap *sg, unsigned long saddr, unsigned long sgt,
1933 		    int fake)
1934 {
1935 	unsigned long raddr, origin, offset, len;
1936 	unsigned long *table;
1937 	phys_addr_t s_sgt;
1938 	struct page *page;
1939 	int rc;
1940 
1941 	BUG_ON(!gmap_is_shadow(sg) || (sgt & _REGION3_ENTRY_LARGE));
1942 	/* Allocate a shadow segment table */
1943 	page = gmap_alloc_crst();
1944 	if (!page)
1945 		return -ENOMEM;
1946 	page->index = sgt & _REGION_ENTRY_ORIGIN;
1947 	if (fake)
1948 		page->index |= GMAP_SHADOW_FAKE_TABLE;
1949 	s_sgt = page_to_phys(page);
1950 	/* Install shadow region second table */
1951 	spin_lock(&sg->guest_table_lock);
1952 	table = gmap_table_walk(sg, saddr, 2); /* get region-3 pointer */
1953 	if (!table) {
1954 		rc = -EAGAIN;		/* Race with unshadow */
1955 		goto out_free;
1956 	}
1957 	if (!(*table & _REGION_ENTRY_INVALID)) {
1958 		rc = 0;			/* Already established */
1959 		goto out_free;
1960 	} else if (*table & _REGION_ENTRY_ORIGIN) {
1961 		rc = -EAGAIN;		/* Race with shadow */
1962 		goto out_free;
1963 	}
1964 	crst_table_init(__va(s_sgt), _SEGMENT_ENTRY_EMPTY);
1965 	/* mark as invalid as long as the parent table is not protected */
1966 	*table = s_sgt | _REGION_ENTRY_LENGTH |
1967 		 _REGION_ENTRY_TYPE_R3 | _REGION_ENTRY_INVALID;
1968 	if (sg->edat_level >= 1)
1969 		*table |= sgt & _REGION_ENTRY_PROTECT;
1970 	list_add(&page->lru, &sg->crst_list);
1971 	if (fake) {
1972 		/* nothing to protect for fake tables */
1973 		*table &= ~_REGION_ENTRY_INVALID;
1974 		spin_unlock(&sg->guest_table_lock);
1975 		return 0;
1976 	}
1977 	spin_unlock(&sg->guest_table_lock);
1978 	/* Make sgt read-only in parent gmap page table */
1979 	raddr = (saddr & _REGION3_MASK) | _SHADOW_RMAP_REGION3;
1980 	origin = sgt & _REGION_ENTRY_ORIGIN;
1981 	offset = ((sgt & _REGION_ENTRY_OFFSET) >> 6) * PAGE_SIZE;
1982 	len = ((sgt & _REGION_ENTRY_LENGTH) + 1) * PAGE_SIZE - offset;
1983 	rc = gmap_protect_rmap(sg, raddr, origin + offset, len);
1984 	spin_lock(&sg->guest_table_lock);
1985 	if (!rc) {
1986 		table = gmap_table_walk(sg, saddr, 2);
1987 		if (!table || (*table & _REGION_ENTRY_ORIGIN) != s_sgt)
1988 			rc = -EAGAIN;		/* Race with unshadow */
1989 		else
1990 			*table &= ~_REGION_ENTRY_INVALID;
1991 	} else {
1992 		gmap_unshadow_sgt(sg, raddr);
1993 	}
1994 	spin_unlock(&sg->guest_table_lock);
1995 	return rc;
1996 out_free:
1997 	spin_unlock(&sg->guest_table_lock);
1998 	__free_pages(page, CRST_ALLOC_ORDER);
1999 	return rc;
2000 }
2001 EXPORT_SYMBOL_GPL(gmap_shadow_sgt);
2002 
2003 /**
2004  * gmap_shadow_pgt_lookup - find a shadow page table
2005  * @sg: pointer to the shadow guest address space structure
2006  * @saddr: the address in the shadow aguest address space
2007  * @pgt: parent gmap address of the page table to get shadowed
2008  * @dat_protection: if the pgtable is marked as protected by dat
2009  * @fake: pgt references contiguous guest memory block, not a pgtable
2010  *
2011  * Returns 0 if the shadow page table was found and -EAGAIN if the page
2012  * table was not found.
2013  *
2014  * Called with sg->mm->mmap_lock in read.
2015  */
gmap_shadow_pgt_lookup(struct gmap * sg,unsigned long saddr,unsigned long * pgt,int * dat_protection,int * fake)2016 int gmap_shadow_pgt_lookup(struct gmap *sg, unsigned long saddr,
2017 			   unsigned long *pgt, int *dat_protection,
2018 			   int *fake)
2019 {
2020 	unsigned long *table;
2021 	struct page *page;
2022 	int rc;
2023 
2024 	BUG_ON(!gmap_is_shadow(sg));
2025 	spin_lock(&sg->guest_table_lock);
2026 	table = gmap_table_walk(sg, saddr, 1); /* get segment pointer */
2027 	if (table && !(*table & _SEGMENT_ENTRY_INVALID)) {
2028 		/* Shadow page tables are full pages (pte+pgste) */
2029 		page = pfn_to_page(*table >> PAGE_SHIFT);
2030 		*pgt = page->index & ~GMAP_SHADOW_FAKE_TABLE;
2031 		*dat_protection = !!(*table & _SEGMENT_ENTRY_PROTECT);
2032 		*fake = !!(page->index & GMAP_SHADOW_FAKE_TABLE);
2033 		rc = 0;
2034 	} else  {
2035 		rc = -EAGAIN;
2036 	}
2037 	spin_unlock(&sg->guest_table_lock);
2038 	return rc;
2039 
2040 }
2041 EXPORT_SYMBOL_GPL(gmap_shadow_pgt_lookup);
2042 
2043 /**
2044  * gmap_shadow_pgt - instantiate a shadow page table
2045  * @sg: pointer to the shadow guest address space structure
2046  * @saddr: faulting address in the shadow gmap
2047  * @pgt: parent gmap address of the page table to get shadowed
2048  * @fake: pgt references contiguous guest memory block, not a pgtable
2049  *
2050  * Returns 0 if successfully shadowed or already shadowed, -EAGAIN if the
2051  * shadow table structure is incomplete, -ENOMEM if out of memory,
2052  * -EFAULT if an address in the parent gmap could not be resolved and
2053  *
2054  * Called with gmap->mm->mmap_lock in read
2055  */
gmap_shadow_pgt(struct gmap * sg,unsigned long saddr,unsigned long pgt,int fake)2056 int gmap_shadow_pgt(struct gmap *sg, unsigned long saddr, unsigned long pgt,
2057 		    int fake)
2058 {
2059 	unsigned long raddr, origin;
2060 	unsigned long *table;
2061 	struct page *page;
2062 	phys_addr_t s_pgt;
2063 	int rc;
2064 
2065 	BUG_ON(!gmap_is_shadow(sg) || (pgt & _SEGMENT_ENTRY_LARGE));
2066 	/* Allocate a shadow page table */
2067 	page = page_table_alloc_pgste(sg->mm);
2068 	if (!page)
2069 		return -ENOMEM;
2070 	page->index = pgt & _SEGMENT_ENTRY_ORIGIN;
2071 	if (fake)
2072 		page->index |= GMAP_SHADOW_FAKE_TABLE;
2073 	s_pgt = page_to_phys(page);
2074 	/* Install shadow page table */
2075 	spin_lock(&sg->guest_table_lock);
2076 	table = gmap_table_walk(sg, saddr, 1); /* get segment pointer */
2077 	if (!table) {
2078 		rc = -EAGAIN;		/* Race with unshadow */
2079 		goto out_free;
2080 	}
2081 	if (!(*table & _SEGMENT_ENTRY_INVALID)) {
2082 		rc = 0;			/* Already established */
2083 		goto out_free;
2084 	} else if (*table & _SEGMENT_ENTRY_ORIGIN) {
2085 		rc = -EAGAIN;		/* Race with shadow */
2086 		goto out_free;
2087 	}
2088 	/* mark as invalid as long as the parent table is not protected */
2089 	*table = (unsigned long) s_pgt | _SEGMENT_ENTRY |
2090 		 (pgt & _SEGMENT_ENTRY_PROTECT) | _SEGMENT_ENTRY_INVALID;
2091 	list_add(&page->lru, &sg->pt_list);
2092 	if (fake) {
2093 		/* nothing to protect for fake tables */
2094 		*table &= ~_SEGMENT_ENTRY_INVALID;
2095 		spin_unlock(&sg->guest_table_lock);
2096 		return 0;
2097 	}
2098 	spin_unlock(&sg->guest_table_lock);
2099 	/* Make pgt read-only in parent gmap page table (not the pgste) */
2100 	raddr = (saddr & _SEGMENT_MASK) | _SHADOW_RMAP_SEGMENT;
2101 	origin = pgt & _SEGMENT_ENTRY_ORIGIN & PAGE_MASK;
2102 	rc = gmap_protect_rmap(sg, raddr, origin, PAGE_SIZE);
2103 	spin_lock(&sg->guest_table_lock);
2104 	if (!rc) {
2105 		table = gmap_table_walk(sg, saddr, 1);
2106 		if (!table || (*table & _SEGMENT_ENTRY_ORIGIN) != s_pgt)
2107 			rc = -EAGAIN;		/* Race with unshadow */
2108 		else
2109 			*table &= ~_SEGMENT_ENTRY_INVALID;
2110 	} else {
2111 		gmap_unshadow_pgt(sg, raddr);
2112 	}
2113 	spin_unlock(&sg->guest_table_lock);
2114 	return rc;
2115 out_free:
2116 	spin_unlock(&sg->guest_table_lock);
2117 	page_table_free_pgste(page);
2118 	return rc;
2119 
2120 }
2121 EXPORT_SYMBOL_GPL(gmap_shadow_pgt);
2122 
2123 /**
2124  * gmap_shadow_page - create a shadow page mapping
2125  * @sg: pointer to the shadow guest address space structure
2126  * @saddr: faulting address in the shadow gmap
2127  * @pte: pte in parent gmap address space to get shadowed
2128  *
2129  * Returns 0 if successfully shadowed or already shadowed, -EAGAIN if the
2130  * shadow table structure is incomplete, -ENOMEM if out of memory and
2131  * -EFAULT if an address in the parent gmap could not be resolved.
2132  *
2133  * Called with sg->mm->mmap_lock in read.
2134  */
gmap_shadow_page(struct gmap * sg,unsigned long saddr,pte_t pte)2135 int gmap_shadow_page(struct gmap *sg, unsigned long saddr, pte_t pte)
2136 {
2137 	struct gmap *parent;
2138 	struct gmap_rmap *rmap;
2139 	unsigned long vmaddr, paddr;
2140 	spinlock_t *ptl;
2141 	pte_t *sptep, *tptep;
2142 	int prot;
2143 	int rc;
2144 
2145 	BUG_ON(!gmap_is_shadow(sg));
2146 	parent = sg->parent;
2147 	prot = (pte_val(pte) & _PAGE_PROTECT) ? PROT_READ : PROT_WRITE;
2148 
2149 	rmap = kzalloc(sizeof(*rmap), GFP_KERNEL_ACCOUNT);
2150 	if (!rmap)
2151 		return -ENOMEM;
2152 	rmap->raddr = (saddr & PAGE_MASK) | _SHADOW_RMAP_PGTABLE;
2153 
2154 	while (1) {
2155 		paddr = pte_val(pte) & PAGE_MASK;
2156 		vmaddr = __gmap_translate(parent, paddr);
2157 		if (IS_ERR_VALUE(vmaddr)) {
2158 			rc = vmaddr;
2159 			break;
2160 		}
2161 		rc = radix_tree_preload(GFP_KERNEL_ACCOUNT);
2162 		if (rc)
2163 			break;
2164 		rc = -EAGAIN;
2165 		sptep = gmap_pte_op_walk(parent, paddr, &ptl);
2166 		if (sptep) {
2167 			spin_lock(&sg->guest_table_lock);
2168 			/* Get page table pointer */
2169 			tptep = (pte_t *) gmap_table_walk(sg, saddr, 0);
2170 			if (!tptep) {
2171 				spin_unlock(&sg->guest_table_lock);
2172 				gmap_pte_op_end(sptep, ptl);
2173 				radix_tree_preload_end();
2174 				break;
2175 			}
2176 			rc = ptep_shadow_pte(sg->mm, saddr, sptep, tptep, pte);
2177 			if (rc > 0) {
2178 				/* Success and a new mapping */
2179 				gmap_insert_rmap(sg, vmaddr, rmap);
2180 				rmap = NULL;
2181 				rc = 0;
2182 			}
2183 			gmap_pte_op_end(sptep, ptl);
2184 			spin_unlock(&sg->guest_table_lock);
2185 		}
2186 		radix_tree_preload_end();
2187 		if (!rc)
2188 			break;
2189 		rc = gmap_pte_op_fixup(parent, paddr, vmaddr, prot);
2190 		if (rc)
2191 			break;
2192 	}
2193 	kfree(rmap);
2194 	return rc;
2195 }
2196 EXPORT_SYMBOL_GPL(gmap_shadow_page);
2197 
2198 /*
2199  * gmap_shadow_notify - handle notifications for shadow gmap
2200  *
2201  * Called with sg->parent->shadow_lock.
2202  */
gmap_shadow_notify(struct gmap * sg,unsigned long vmaddr,unsigned long gaddr)2203 static void gmap_shadow_notify(struct gmap *sg, unsigned long vmaddr,
2204 			       unsigned long gaddr)
2205 {
2206 	struct gmap_rmap *rmap, *rnext, *head;
2207 	unsigned long start, end, bits, raddr;
2208 
2209 	BUG_ON(!gmap_is_shadow(sg));
2210 
2211 	spin_lock(&sg->guest_table_lock);
2212 	if (sg->removed) {
2213 		spin_unlock(&sg->guest_table_lock);
2214 		return;
2215 	}
2216 	/* Check for top level table */
2217 	start = sg->orig_asce & _ASCE_ORIGIN;
2218 	end = start + ((sg->orig_asce & _ASCE_TABLE_LENGTH) + 1) * PAGE_SIZE;
2219 	if (!(sg->orig_asce & _ASCE_REAL_SPACE) && gaddr >= start &&
2220 	    gaddr < end) {
2221 		/* The complete shadow table has to go */
2222 		gmap_unshadow(sg);
2223 		spin_unlock(&sg->guest_table_lock);
2224 		list_del(&sg->list);
2225 		gmap_put(sg);
2226 		return;
2227 	}
2228 	/* Remove the page table tree from on specific entry */
2229 	head = radix_tree_delete(&sg->host_to_rmap, vmaddr >> PAGE_SHIFT);
2230 	gmap_for_each_rmap_safe(rmap, rnext, head) {
2231 		bits = rmap->raddr & _SHADOW_RMAP_MASK;
2232 		raddr = rmap->raddr ^ bits;
2233 		switch (bits) {
2234 		case _SHADOW_RMAP_REGION1:
2235 			gmap_unshadow_r2t(sg, raddr);
2236 			break;
2237 		case _SHADOW_RMAP_REGION2:
2238 			gmap_unshadow_r3t(sg, raddr);
2239 			break;
2240 		case _SHADOW_RMAP_REGION3:
2241 			gmap_unshadow_sgt(sg, raddr);
2242 			break;
2243 		case _SHADOW_RMAP_SEGMENT:
2244 			gmap_unshadow_pgt(sg, raddr);
2245 			break;
2246 		case _SHADOW_RMAP_PGTABLE:
2247 			gmap_unshadow_page(sg, raddr);
2248 			break;
2249 		}
2250 		kfree(rmap);
2251 	}
2252 	spin_unlock(&sg->guest_table_lock);
2253 }
2254 
2255 /**
2256  * ptep_notify - call all invalidation callbacks for a specific pte.
2257  * @mm: pointer to the process mm_struct
2258  * @vmaddr: virtual address in the process address space
2259  * @pte: pointer to the page table entry
2260  * @bits: bits from the pgste that caused the notify call
2261  *
2262  * This function is assumed to be called with the page table lock held
2263  * for the pte to notify.
2264  */
ptep_notify(struct mm_struct * mm,unsigned long vmaddr,pte_t * pte,unsigned long bits)2265 void ptep_notify(struct mm_struct *mm, unsigned long vmaddr,
2266 		 pte_t *pte, unsigned long bits)
2267 {
2268 	unsigned long offset, gaddr = 0;
2269 	unsigned long *table;
2270 	struct gmap *gmap, *sg, *next;
2271 
2272 	offset = ((unsigned long) pte) & (255 * sizeof(pte_t));
2273 	offset = offset * (PAGE_SIZE / sizeof(pte_t));
2274 	rcu_read_lock();
2275 	list_for_each_entry_rcu(gmap, &mm->context.gmap_list, list) {
2276 		spin_lock(&gmap->guest_table_lock);
2277 		table = radix_tree_lookup(&gmap->host_to_guest,
2278 					  vmaddr >> PMD_SHIFT);
2279 		if (table)
2280 			gaddr = __gmap_segment_gaddr(table) + offset;
2281 		spin_unlock(&gmap->guest_table_lock);
2282 		if (!table)
2283 			continue;
2284 
2285 		if (!list_empty(&gmap->children) && (bits & PGSTE_VSIE_BIT)) {
2286 			spin_lock(&gmap->shadow_lock);
2287 			list_for_each_entry_safe(sg, next,
2288 						 &gmap->children, list)
2289 				gmap_shadow_notify(sg, vmaddr, gaddr);
2290 			spin_unlock(&gmap->shadow_lock);
2291 		}
2292 		if (bits & PGSTE_IN_BIT)
2293 			gmap_call_notifier(gmap, gaddr, gaddr + PAGE_SIZE - 1);
2294 	}
2295 	rcu_read_unlock();
2296 }
2297 EXPORT_SYMBOL_GPL(ptep_notify);
2298 
pmdp_notify_gmap(struct gmap * gmap,pmd_t * pmdp,unsigned long gaddr)2299 static void pmdp_notify_gmap(struct gmap *gmap, pmd_t *pmdp,
2300 			     unsigned long gaddr)
2301 {
2302 	set_pmd(pmdp, clear_pmd_bit(*pmdp, __pgprot(_SEGMENT_ENTRY_GMAP_IN)));
2303 	gmap_call_notifier(gmap, gaddr, gaddr + HPAGE_SIZE - 1);
2304 }
2305 
2306 /**
2307  * gmap_pmdp_xchg - exchange a gmap pmd with another
2308  * @gmap: pointer to the guest address space structure
2309  * @pmdp: pointer to the pmd entry
2310  * @new: replacement entry
2311  * @gaddr: the affected guest address
2312  *
2313  * This function is assumed to be called with the guest_table_lock
2314  * held.
2315  */
gmap_pmdp_xchg(struct gmap * gmap,pmd_t * pmdp,pmd_t new,unsigned long gaddr)2316 static void gmap_pmdp_xchg(struct gmap *gmap, pmd_t *pmdp, pmd_t new,
2317 			   unsigned long gaddr)
2318 {
2319 	gaddr &= HPAGE_MASK;
2320 	pmdp_notify_gmap(gmap, pmdp, gaddr);
2321 	new = clear_pmd_bit(new, __pgprot(_SEGMENT_ENTRY_GMAP_IN));
2322 	if (MACHINE_HAS_TLB_GUEST)
2323 		__pmdp_idte(gaddr, (pmd_t *)pmdp, IDTE_GUEST_ASCE, gmap->asce,
2324 			    IDTE_GLOBAL);
2325 	else if (MACHINE_HAS_IDTE)
2326 		__pmdp_idte(gaddr, (pmd_t *)pmdp, 0, 0, IDTE_GLOBAL);
2327 	else
2328 		__pmdp_csp(pmdp);
2329 	set_pmd(pmdp, new);
2330 }
2331 
gmap_pmdp_clear(struct mm_struct * mm,unsigned long vmaddr,int purge)2332 static void gmap_pmdp_clear(struct mm_struct *mm, unsigned long vmaddr,
2333 			    int purge)
2334 {
2335 	pmd_t *pmdp;
2336 	struct gmap *gmap;
2337 	unsigned long gaddr;
2338 
2339 	rcu_read_lock();
2340 	list_for_each_entry_rcu(gmap, &mm->context.gmap_list, list) {
2341 		spin_lock(&gmap->guest_table_lock);
2342 		pmdp = (pmd_t *)radix_tree_delete(&gmap->host_to_guest,
2343 						  vmaddr >> PMD_SHIFT);
2344 		if (pmdp) {
2345 			gaddr = __gmap_segment_gaddr((unsigned long *)pmdp);
2346 			pmdp_notify_gmap(gmap, pmdp, gaddr);
2347 			WARN_ON(pmd_val(*pmdp) & ~(_SEGMENT_ENTRY_HARDWARE_BITS_LARGE |
2348 						   _SEGMENT_ENTRY_GMAP_UC));
2349 			if (purge)
2350 				__pmdp_csp(pmdp);
2351 			set_pmd(pmdp, __pmd(_SEGMENT_ENTRY_EMPTY));
2352 		}
2353 		spin_unlock(&gmap->guest_table_lock);
2354 	}
2355 	rcu_read_unlock();
2356 }
2357 
2358 /**
2359  * gmap_pmdp_invalidate - invalidate all affected guest pmd entries without
2360  *                        flushing
2361  * @mm: pointer to the process mm_struct
2362  * @vmaddr: virtual address in the process address space
2363  */
gmap_pmdp_invalidate(struct mm_struct * mm,unsigned long vmaddr)2364 void gmap_pmdp_invalidate(struct mm_struct *mm, unsigned long vmaddr)
2365 {
2366 	gmap_pmdp_clear(mm, vmaddr, 0);
2367 }
2368 EXPORT_SYMBOL_GPL(gmap_pmdp_invalidate);
2369 
2370 /**
2371  * gmap_pmdp_csp - csp all affected guest pmd entries
2372  * @mm: pointer to the process mm_struct
2373  * @vmaddr: virtual address in the process address space
2374  */
gmap_pmdp_csp(struct mm_struct * mm,unsigned long vmaddr)2375 void gmap_pmdp_csp(struct mm_struct *mm, unsigned long vmaddr)
2376 {
2377 	gmap_pmdp_clear(mm, vmaddr, 1);
2378 }
2379 EXPORT_SYMBOL_GPL(gmap_pmdp_csp);
2380 
2381 /**
2382  * gmap_pmdp_idte_local - invalidate and clear a guest pmd entry
2383  * @mm: pointer to the process mm_struct
2384  * @vmaddr: virtual address in the process address space
2385  */
gmap_pmdp_idte_local(struct mm_struct * mm,unsigned long vmaddr)2386 void gmap_pmdp_idte_local(struct mm_struct *mm, unsigned long vmaddr)
2387 {
2388 	unsigned long *entry, gaddr;
2389 	struct gmap *gmap;
2390 	pmd_t *pmdp;
2391 
2392 	rcu_read_lock();
2393 	list_for_each_entry_rcu(gmap, &mm->context.gmap_list, list) {
2394 		spin_lock(&gmap->guest_table_lock);
2395 		entry = radix_tree_delete(&gmap->host_to_guest,
2396 					  vmaddr >> PMD_SHIFT);
2397 		if (entry) {
2398 			pmdp = (pmd_t *)entry;
2399 			gaddr = __gmap_segment_gaddr(entry);
2400 			pmdp_notify_gmap(gmap, pmdp, gaddr);
2401 			WARN_ON(*entry & ~(_SEGMENT_ENTRY_HARDWARE_BITS_LARGE |
2402 					   _SEGMENT_ENTRY_GMAP_UC));
2403 			if (MACHINE_HAS_TLB_GUEST)
2404 				__pmdp_idte(gaddr, pmdp, IDTE_GUEST_ASCE,
2405 					    gmap->asce, IDTE_LOCAL);
2406 			else if (MACHINE_HAS_IDTE)
2407 				__pmdp_idte(gaddr, pmdp, 0, 0, IDTE_LOCAL);
2408 			*entry = _SEGMENT_ENTRY_EMPTY;
2409 		}
2410 		spin_unlock(&gmap->guest_table_lock);
2411 	}
2412 	rcu_read_unlock();
2413 }
2414 EXPORT_SYMBOL_GPL(gmap_pmdp_idte_local);
2415 
2416 /**
2417  * gmap_pmdp_idte_global - invalidate and clear a guest pmd entry
2418  * @mm: pointer to the process mm_struct
2419  * @vmaddr: virtual address in the process address space
2420  */
gmap_pmdp_idte_global(struct mm_struct * mm,unsigned long vmaddr)2421 void gmap_pmdp_idte_global(struct mm_struct *mm, unsigned long vmaddr)
2422 {
2423 	unsigned long *entry, gaddr;
2424 	struct gmap *gmap;
2425 	pmd_t *pmdp;
2426 
2427 	rcu_read_lock();
2428 	list_for_each_entry_rcu(gmap, &mm->context.gmap_list, list) {
2429 		spin_lock(&gmap->guest_table_lock);
2430 		entry = radix_tree_delete(&gmap->host_to_guest,
2431 					  vmaddr >> PMD_SHIFT);
2432 		if (entry) {
2433 			pmdp = (pmd_t *)entry;
2434 			gaddr = __gmap_segment_gaddr(entry);
2435 			pmdp_notify_gmap(gmap, pmdp, gaddr);
2436 			WARN_ON(*entry & ~(_SEGMENT_ENTRY_HARDWARE_BITS_LARGE |
2437 					   _SEGMENT_ENTRY_GMAP_UC));
2438 			if (MACHINE_HAS_TLB_GUEST)
2439 				__pmdp_idte(gaddr, pmdp, IDTE_GUEST_ASCE,
2440 					    gmap->asce, IDTE_GLOBAL);
2441 			else if (MACHINE_HAS_IDTE)
2442 				__pmdp_idte(gaddr, pmdp, 0, 0, IDTE_GLOBAL);
2443 			else
2444 				__pmdp_csp(pmdp);
2445 			*entry = _SEGMENT_ENTRY_EMPTY;
2446 		}
2447 		spin_unlock(&gmap->guest_table_lock);
2448 	}
2449 	rcu_read_unlock();
2450 }
2451 EXPORT_SYMBOL_GPL(gmap_pmdp_idte_global);
2452 
2453 /**
2454  * gmap_test_and_clear_dirty_pmd - test and reset segment dirty status
2455  * @gmap: pointer to guest address space
2456  * @pmdp: pointer to the pmd to be tested
2457  * @gaddr: virtual address in the guest address space
2458  *
2459  * This function is assumed to be called with the guest_table_lock
2460  * held.
2461  */
gmap_test_and_clear_dirty_pmd(struct gmap * gmap,pmd_t * pmdp,unsigned long gaddr)2462 static bool gmap_test_and_clear_dirty_pmd(struct gmap *gmap, pmd_t *pmdp,
2463 					  unsigned long gaddr)
2464 {
2465 	if (pmd_val(*pmdp) & _SEGMENT_ENTRY_INVALID)
2466 		return false;
2467 
2468 	/* Already protected memory, which did not change is clean */
2469 	if (pmd_val(*pmdp) & _SEGMENT_ENTRY_PROTECT &&
2470 	    !(pmd_val(*pmdp) & _SEGMENT_ENTRY_GMAP_UC))
2471 		return false;
2472 
2473 	/* Clear UC indication and reset protection */
2474 	set_pmd(pmdp, clear_pmd_bit(*pmdp, __pgprot(_SEGMENT_ENTRY_GMAP_UC)));
2475 	gmap_protect_pmd(gmap, gaddr, pmdp, PROT_READ, 0);
2476 	return true;
2477 }
2478 
2479 /**
2480  * gmap_sync_dirty_log_pmd - set bitmap based on dirty status of segment
2481  * @gmap: pointer to guest address space
2482  * @bitmap: dirty bitmap for this pmd
2483  * @gaddr: virtual address in the guest address space
2484  * @vmaddr: virtual address in the host address space
2485  *
2486  * This function is assumed to be called with the guest_table_lock
2487  * held.
2488  */
gmap_sync_dirty_log_pmd(struct gmap * gmap,unsigned long bitmap[4],unsigned long gaddr,unsigned long vmaddr)2489 void gmap_sync_dirty_log_pmd(struct gmap *gmap, unsigned long bitmap[4],
2490 			     unsigned long gaddr, unsigned long vmaddr)
2491 {
2492 	int i;
2493 	pmd_t *pmdp;
2494 	pte_t *ptep;
2495 	spinlock_t *ptl;
2496 
2497 	pmdp = gmap_pmd_op_walk(gmap, gaddr);
2498 	if (!pmdp)
2499 		return;
2500 
2501 	if (pmd_large(*pmdp)) {
2502 		if (gmap_test_and_clear_dirty_pmd(gmap, pmdp, gaddr))
2503 			bitmap_fill(bitmap, _PAGE_ENTRIES);
2504 	} else {
2505 		for (i = 0; i < _PAGE_ENTRIES; i++, vmaddr += PAGE_SIZE) {
2506 			ptep = pte_alloc_map_lock(gmap->mm, pmdp, vmaddr, &ptl);
2507 			if (!ptep)
2508 				continue;
2509 			if (ptep_test_and_clear_uc(gmap->mm, vmaddr, ptep))
2510 				set_bit(i, bitmap);
2511 			pte_unmap_unlock(ptep, ptl);
2512 		}
2513 	}
2514 	gmap_pmd_op_end(gmap, pmdp);
2515 }
2516 EXPORT_SYMBOL_GPL(gmap_sync_dirty_log_pmd);
2517 
2518 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
thp_split_walk_pmd_entry(pmd_t * pmd,unsigned long addr,unsigned long end,struct mm_walk * walk)2519 static int thp_split_walk_pmd_entry(pmd_t *pmd, unsigned long addr,
2520 				    unsigned long end, struct mm_walk *walk)
2521 {
2522 	struct vm_area_struct *vma = walk->vma;
2523 
2524 	split_huge_pmd(vma, pmd, addr);
2525 	return 0;
2526 }
2527 
2528 static const struct mm_walk_ops thp_split_walk_ops = {
2529 	.pmd_entry	= thp_split_walk_pmd_entry,
2530 	.walk_lock	= PGWALK_WRLOCK_VERIFY,
2531 };
2532 
thp_split_mm(struct mm_struct * mm)2533 static inline void thp_split_mm(struct mm_struct *mm)
2534 {
2535 	struct vm_area_struct *vma;
2536 	VMA_ITERATOR(vmi, mm, 0);
2537 
2538 	for_each_vma(vmi, vma) {
2539 		vm_flags_mod(vma, VM_NOHUGEPAGE, VM_HUGEPAGE);
2540 		walk_page_vma(vma, &thp_split_walk_ops, NULL);
2541 	}
2542 	mm->def_flags |= VM_NOHUGEPAGE;
2543 }
2544 #else
thp_split_mm(struct mm_struct * mm)2545 static inline void thp_split_mm(struct mm_struct *mm)
2546 {
2547 }
2548 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
2549 
2550 /*
2551  * switch on pgstes for its userspace process (for kvm)
2552  */
s390_enable_sie(void)2553 int s390_enable_sie(void)
2554 {
2555 	struct mm_struct *mm = current->mm;
2556 
2557 	/* Do we have pgstes? if yes, we are done */
2558 	if (mm_has_pgste(mm))
2559 		return 0;
2560 	/* Fail if the page tables are 2K */
2561 	if (!mm_alloc_pgste(mm))
2562 		return -EINVAL;
2563 	mmap_write_lock(mm);
2564 	mm->context.has_pgste = 1;
2565 	/* split thp mappings and disable thp for future mappings */
2566 	thp_split_mm(mm);
2567 	mmap_write_unlock(mm);
2568 	return 0;
2569 }
2570 EXPORT_SYMBOL_GPL(s390_enable_sie);
2571 
find_zeropage_pte_entry(pte_t * pte,unsigned long addr,unsigned long end,struct mm_walk * walk)2572 static int find_zeropage_pte_entry(pte_t *pte, unsigned long addr,
2573 				   unsigned long end, struct mm_walk *walk)
2574 {
2575 	unsigned long *found_addr = walk->private;
2576 
2577 	/* Return 1 of the page is a zeropage. */
2578 	if (is_zero_pfn(pte_pfn(*pte))) {
2579 		/*
2580 		 * Shared zeropage in e.g., a FS DAX mapping? We cannot do the
2581 		 * right thing and likely don't care: FAULT_FLAG_UNSHARE
2582 		 * currently only works in COW mappings, which is also where
2583 		 * mm_forbids_zeropage() is checked.
2584 		 */
2585 		if (!is_cow_mapping(walk->vma->vm_flags))
2586 			return -EFAULT;
2587 
2588 		*found_addr = addr;
2589 		return 1;
2590 	}
2591 	return 0;
2592 }
2593 
2594 static const struct mm_walk_ops find_zeropage_ops = {
2595 	.pte_entry	= find_zeropage_pte_entry,
2596 	.walk_lock	= PGWALK_WRLOCK,
2597 };
2598 
2599 /*
2600  * Unshare all shared zeropages, replacing them by anonymous pages. Note that
2601  * we cannot simply zap all shared zeropages, because this could later
2602  * trigger unexpected userfaultfd missing events.
2603  *
2604  * This must be called after mm->context.allow_cow_sharing was
2605  * set to 0, to avoid future mappings of shared zeropages.
2606  *
2607  * mm contracts with s390, that even if mm were to remove a page table,
2608  * and racing with walk_page_range_vma() calling pte_offset_map_lock()
2609  * would fail, it will never insert a page table containing empty zero
2610  * pages once mm_forbids_zeropage(mm) i.e.
2611  * mm->context.allow_cow_sharing is set to 0.
2612  */
__s390_unshare_zeropages(struct mm_struct * mm)2613 static int __s390_unshare_zeropages(struct mm_struct *mm)
2614 {
2615 	struct vm_area_struct *vma;
2616 	VMA_ITERATOR(vmi, mm, 0);
2617 	unsigned long addr;
2618 	vm_fault_t fault;
2619 	int rc;
2620 
2621 	for_each_vma(vmi, vma) {
2622 		/*
2623 		 * We could only look at COW mappings, but it's more future
2624 		 * proof to catch unexpected zeropages in other mappings and
2625 		 * fail.
2626 		 */
2627 		if ((vma->vm_flags & VM_PFNMAP) || is_vm_hugetlb_page(vma))
2628 			continue;
2629 		addr = vma->vm_start;
2630 
2631 retry:
2632 		rc = walk_page_range_vma(vma, addr, vma->vm_end,
2633 					 &find_zeropage_ops, &addr);
2634 		if (rc < 0)
2635 			return rc;
2636 		else if (!rc)
2637 			continue;
2638 
2639 		/* addr was updated by find_zeropage_pte_entry() */
2640 		fault = handle_mm_fault(vma, addr,
2641 					FAULT_FLAG_UNSHARE | FAULT_FLAG_REMOTE,
2642 					NULL);
2643 		if (fault & VM_FAULT_OOM)
2644 			return -ENOMEM;
2645 		/*
2646 		 * See break_ksm(): even after handle_mm_fault() returned 0, we
2647 		 * must start the lookup from the current address, because
2648 		 * handle_mm_fault() may back out if there's any difficulty.
2649 		 *
2650 		 * VM_FAULT_SIGBUS and VM_FAULT_SIGSEGV are unexpected but
2651 		 * maybe they could trigger in the future on concurrent
2652 		 * truncation. In that case, the shared zeropage would be gone
2653 		 * and we can simply retry and make progress.
2654 		 */
2655 		cond_resched();
2656 		goto retry;
2657 	}
2658 
2659 	return 0;
2660 }
2661 
__s390_disable_cow_sharing(struct mm_struct * mm)2662 static int __s390_disable_cow_sharing(struct mm_struct *mm)
2663 {
2664 	int rc;
2665 
2666 	if (!mm->context.allow_cow_sharing)
2667 		return 0;
2668 
2669 	mm->context.allow_cow_sharing = 0;
2670 
2671 	/* Replace all shared zeropages by anonymous pages. */
2672 	rc = __s390_unshare_zeropages(mm);
2673 	/*
2674 	 * Make sure to disable KSM (if enabled for the whole process or
2675 	 * individual VMAs). Note that nothing currently hinders user space
2676 	 * from re-enabling it.
2677 	 */
2678 	if (!rc)
2679 		rc = ksm_disable(mm);
2680 	if (rc)
2681 		mm->context.allow_cow_sharing = 1;
2682 	return rc;
2683 }
2684 
2685 /*
2686  * Disable most COW-sharing of memory pages for the whole process:
2687  * (1) Disable KSM and unmerge/unshare any KSM pages.
2688  * (2) Disallow shared zeropages and unshare any zerpages that are mapped.
2689  *
2690  * Not that we currently don't bother with COW-shared pages that are shared
2691  * with parent/child processes due to fork().
2692  */
s390_disable_cow_sharing(void)2693 int s390_disable_cow_sharing(void)
2694 {
2695 	int rc;
2696 
2697 	mmap_write_lock(current->mm);
2698 	rc = __s390_disable_cow_sharing(current->mm);
2699 	mmap_write_unlock(current->mm);
2700 	return rc;
2701 }
2702 EXPORT_SYMBOL_GPL(s390_disable_cow_sharing);
2703 
2704 /*
2705  * Enable storage key handling from now on and initialize the storage
2706  * keys with the default key.
2707  */
__s390_enable_skey_pte(pte_t * pte,unsigned long addr,unsigned long next,struct mm_walk * walk)2708 static int __s390_enable_skey_pte(pte_t *pte, unsigned long addr,
2709 				  unsigned long next, struct mm_walk *walk)
2710 {
2711 	/* Clear storage key */
2712 	ptep_zap_key(walk->mm, addr, pte);
2713 	return 0;
2714 }
2715 
2716 /*
2717  * Give a chance to schedule after setting a key to 256 pages.
2718  * We only hold the mm lock, which is a rwsem and the kvm srcu.
2719  * Both can sleep.
2720  */
__s390_enable_skey_pmd(pmd_t * pmd,unsigned long addr,unsigned long next,struct mm_walk * walk)2721 static int __s390_enable_skey_pmd(pmd_t *pmd, unsigned long addr,
2722 				  unsigned long next, struct mm_walk *walk)
2723 {
2724 	cond_resched();
2725 	return 0;
2726 }
2727 
__s390_enable_skey_hugetlb(pte_t * pte,unsigned long addr,unsigned long hmask,unsigned long next,struct mm_walk * walk)2728 static int __s390_enable_skey_hugetlb(pte_t *pte, unsigned long addr,
2729 				      unsigned long hmask, unsigned long next,
2730 				      struct mm_walk *walk)
2731 {
2732 	pmd_t *pmd = (pmd_t *)pte;
2733 	unsigned long start, end;
2734 	struct page *page = pmd_page(*pmd);
2735 
2736 	/*
2737 	 * The write check makes sure we do not set a key on shared
2738 	 * memory. This is needed as the walker does not differentiate
2739 	 * between actual guest memory and the process executable or
2740 	 * shared libraries.
2741 	 */
2742 	if (pmd_val(*pmd) & _SEGMENT_ENTRY_INVALID ||
2743 	    !(pmd_val(*pmd) & _SEGMENT_ENTRY_WRITE))
2744 		return 0;
2745 
2746 	start = pmd_val(*pmd) & HPAGE_MASK;
2747 	end = start + HPAGE_SIZE;
2748 	__storage_key_init_range(start, end);
2749 	set_bit(PG_arch_1, &page->flags);
2750 	cond_resched();
2751 	return 0;
2752 }
2753 
2754 static const struct mm_walk_ops enable_skey_walk_ops = {
2755 	.hugetlb_entry		= __s390_enable_skey_hugetlb,
2756 	.pte_entry		= __s390_enable_skey_pte,
2757 	.pmd_entry		= __s390_enable_skey_pmd,
2758 	.walk_lock		= PGWALK_WRLOCK,
2759 };
2760 
s390_enable_skey(void)2761 int s390_enable_skey(void)
2762 {
2763 	struct mm_struct *mm = current->mm;
2764 	int rc = 0;
2765 
2766 	mmap_write_lock(mm);
2767 	if (mm_uses_skeys(mm))
2768 		goto out_up;
2769 
2770 	mm->context.uses_skeys = 1;
2771 	rc = __s390_disable_cow_sharing(mm);
2772 	if (rc) {
2773 		mm->context.uses_skeys = 0;
2774 		goto out_up;
2775 	}
2776 	walk_page_range(mm, 0, TASK_SIZE, &enable_skey_walk_ops, NULL);
2777 
2778 out_up:
2779 	mmap_write_unlock(mm);
2780 	return rc;
2781 }
2782 EXPORT_SYMBOL_GPL(s390_enable_skey);
2783 
2784 /*
2785  * Reset CMMA state, make all pages stable again.
2786  */
__s390_reset_cmma(pte_t * pte,unsigned long addr,unsigned long next,struct mm_walk * walk)2787 static int __s390_reset_cmma(pte_t *pte, unsigned long addr,
2788 			     unsigned long next, struct mm_walk *walk)
2789 {
2790 	ptep_zap_unused(walk->mm, addr, pte, 1);
2791 	return 0;
2792 }
2793 
2794 static const struct mm_walk_ops reset_cmma_walk_ops = {
2795 	.pte_entry		= __s390_reset_cmma,
2796 	.walk_lock		= PGWALK_WRLOCK,
2797 };
2798 
s390_reset_cmma(struct mm_struct * mm)2799 void s390_reset_cmma(struct mm_struct *mm)
2800 {
2801 	mmap_write_lock(mm);
2802 	walk_page_range(mm, 0, TASK_SIZE, &reset_cmma_walk_ops, NULL);
2803 	mmap_write_unlock(mm);
2804 }
2805 EXPORT_SYMBOL_GPL(s390_reset_cmma);
2806 
2807 #define GATHER_GET_PAGES 32
2808 
2809 struct reset_walk_state {
2810 	unsigned long next;
2811 	unsigned long count;
2812 	unsigned long pfns[GATHER_GET_PAGES];
2813 };
2814 
s390_gather_pages(pte_t * ptep,unsigned long addr,unsigned long next,struct mm_walk * walk)2815 static int s390_gather_pages(pte_t *ptep, unsigned long addr,
2816 			     unsigned long next, struct mm_walk *walk)
2817 {
2818 	struct reset_walk_state *p = walk->private;
2819 	pte_t pte = READ_ONCE(*ptep);
2820 
2821 	if (pte_present(pte)) {
2822 		/* we have a reference from the mapping, take an extra one */
2823 		get_page(phys_to_page(pte_val(pte)));
2824 		p->pfns[p->count] = phys_to_pfn(pte_val(pte));
2825 		p->next = next;
2826 		p->count++;
2827 	}
2828 	return p->count >= GATHER_GET_PAGES;
2829 }
2830 
2831 static const struct mm_walk_ops gather_pages_ops = {
2832 	.pte_entry = s390_gather_pages,
2833 	.walk_lock = PGWALK_RDLOCK,
2834 };
2835 
2836 /*
2837  * Call the Destroy secure page UVC on each page in the given array of PFNs.
2838  * Each page needs to have an extra reference, which will be released here.
2839  */
s390_uv_destroy_pfns(unsigned long count,unsigned long * pfns)2840 void s390_uv_destroy_pfns(unsigned long count, unsigned long *pfns)
2841 {
2842 	unsigned long i;
2843 
2844 	for (i = 0; i < count; i++) {
2845 		/* we always have an extra reference */
2846 		uv_destroy_owned_page(pfn_to_phys(pfns[i]));
2847 		/* get rid of the extra reference */
2848 		put_page(pfn_to_page(pfns[i]));
2849 		cond_resched();
2850 	}
2851 }
2852 EXPORT_SYMBOL_GPL(s390_uv_destroy_pfns);
2853 
2854 /**
2855  * __s390_uv_destroy_range - Call the destroy secure page UVC on each page
2856  * in the given range of the given address space.
2857  * @mm: the mm to operate on
2858  * @start: the start of the range
2859  * @end: the end of the range
2860  * @interruptible: if not 0, stop when a fatal signal is received
2861  *
2862  * Walk the given range of the given address space and call the destroy
2863  * secure page UVC on each page. Optionally exit early if a fatal signal is
2864  * pending.
2865  *
2866  * Return: 0 on success, -EINTR if the function stopped before completing
2867  */
__s390_uv_destroy_range(struct mm_struct * mm,unsigned long start,unsigned long end,bool interruptible)2868 int __s390_uv_destroy_range(struct mm_struct *mm, unsigned long start,
2869 			    unsigned long end, bool interruptible)
2870 {
2871 	struct reset_walk_state state = { .next = start };
2872 	int r = 1;
2873 
2874 	while (r > 0) {
2875 		state.count = 0;
2876 		mmap_read_lock(mm);
2877 		r = walk_page_range(mm, state.next, end, &gather_pages_ops, &state);
2878 		mmap_read_unlock(mm);
2879 		cond_resched();
2880 		s390_uv_destroy_pfns(state.count, state.pfns);
2881 		if (interruptible && fatal_signal_pending(current))
2882 			return -EINTR;
2883 	}
2884 	return 0;
2885 }
2886 EXPORT_SYMBOL_GPL(__s390_uv_destroy_range);
2887 
2888 /**
2889  * s390_unlist_old_asce - Remove the topmost level of page tables from the
2890  * list of page tables of the gmap.
2891  * @gmap: the gmap whose table is to be removed
2892  *
2893  * On s390x, KVM keeps a list of all pages containing the page tables of the
2894  * gmap (the CRST list). This list is used at tear down time to free all
2895  * pages that are now not needed anymore.
2896  *
2897  * This function removes the topmost page of the tree (the one pointed to by
2898  * the ASCE) from the CRST list.
2899  *
2900  * This means that it will not be freed when the VM is torn down, and needs
2901  * to be handled separately by the caller, unless a leak is actually
2902  * intended. Notice that this function will only remove the page from the
2903  * list, the page will still be used as a top level page table (and ASCE).
2904  */
s390_unlist_old_asce(struct gmap * gmap)2905 void s390_unlist_old_asce(struct gmap *gmap)
2906 {
2907 	struct page *old;
2908 
2909 	old = virt_to_page(gmap->table);
2910 	spin_lock(&gmap->guest_table_lock);
2911 	list_del(&old->lru);
2912 	/*
2913 	 * Sometimes the topmost page might need to be "removed" multiple
2914 	 * times, for example if the VM is rebooted into secure mode several
2915 	 * times concurrently, or if s390_replace_asce fails after calling
2916 	 * s390_remove_old_asce and is attempted again later. In that case
2917 	 * the old asce has been removed from the list, and therefore it
2918 	 * will not be freed when the VM terminates, but the ASCE is still
2919 	 * in use and still pointed to.
2920 	 * A subsequent call to replace_asce will follow the pointer and try
2921 	 * to remove the same page from the list again.
2922 	 * Therefore it's necessary that the page of the ASCE has valid
2923 	 * pointers, so list_del can work (and do nothing) without
2924 	 * dereferencing stale or invalid pointers.
2925 	 */
2926 	INIT_LIST_HEAD(&old->lru);
2927 	spin_unlock(&gmap->guest_table_lock);
2928 }
2929 EXPORT_SYMBOL_GPL(s390_unlist_old_asce);
2930 
2931 /**
2932  * s390_replace_asce - Try to replace the current ASCE of a gmap with a copy
2933  * @gmap: the gmap whose ASCE needs to be replaced
2934  *
2935  * If the ASCE is a SEGMENT type then this function will return -EINVAL,
2936  * otherwise the pointers in the host_to_guest radix tree will keep pointing
2937  * to the wrong pages, causing use-after-free and memory corruption.
2938  * If the allocation of the new top level page table fails, the ASCE is not
2939  * replaced.
2940  * In any case, the old ASCE is always removed from the gmap CRST list.
2941  * Therefore the caller has to make sure to save a pointer to it
2942  * beforehand, unless a leak is actually intended.
2943  */
s390_replace_asce(struct gmap * gmap)2944 int s390_replace_asce(struct gmap *gmap)
2945 {
2946 	unsigned long asce;
2947 	struct page *page;
2948 	void *table;
2949 
2950 	s390_unlist_old_asce(gmap);
2951 
2952 	/* Replacing segment type ASCEs would cause serious issues */
2953 	if ((gmap->asce & _ASCE_TYPE_MASK) == _ASCE_TYPE_SEGMENT)
2954 		return -EINVAL;
2955 
2956 	page = gmap_alloc_crst();
2957 	if (!page)
2958 		return -ENOMEM;
2959 	page->index = 0;
2960 	table = page_to_virt(page);
2961 	memcpy(table, gmap->table, 1UL << (CRST_ALLOC_ORDER + PAGE_SHIFT));
2962 
2963 	/*
2964 	 * The caller has to deal with the old ASCE, but here we make sure
2965 	 * the new one is properly added to the CRST list, so that
2966 	 * it will be freed when the VM is torn down.
2967 	 */
2968 	spin_lock(&gmap->guest_table_lock);
2969 	list_add(&page->lru, &gmap->crst_list);
2970 	spin_unlock(&gmap->guest_table_lock);
2971 
2972 	/* Set new table origin while preserving existing ASCE control bits */
2973 	asce = (gmap->asce & ~_ASCE_ORIGIN) | __pa(table);
2974 	WRITE_ONCE(gmap->asce, asce);
2975 	WRITE_ONCE(gmap->mm->context.gmap_asce, asce);
2976 	WRITE_ONCE(gmap->table, table);
2977 
2978 	return 0;
2979 }
2980 EXPORT_SYMBOL_GPL(s390_replace_asce);
2981