]> git.proxmox.com Git - mirror_ubuntu-bionic-kernel.git/blob - arch/s390/mm/gmap.c
Merge tag 'mmc-v4.15-2' of git://git.kernel.org/pub/scm/linux/kernel/git/ulfh/mmc
[mirror_ubuntu-bionic-kernel.git] / arch / s390 / mm / gmap.c
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3 * KVM guest address space mapping code
4 *
5 * Copyright IBM Corp. 2007, 2016
6 * Author(s): Martin Schwidefsky <schwidefsky@de.ibm.com>
7 */
8
9 #include <linux/kernel.h>
10 #include <linux/mm.h>
11 #include <linux/swap.h>
12 #include <linux/smp.h>
13 #include <linux/spinlock.h>
14 #include <linux/slab.h>
15 #include <linux/swapops.h>
16 #include <linux/ksm.h>
17 #include <linux/mman.h>
18
19 #include <asm/pgtable.h>
20 #include <asm/pgalloc.h>
21 #include <asm/gmap.h>
22 #include <asm/tlb.h>
23
24 #define GMAP_SHADOW_FAKE_TABLE 1ULL
25
26 /**
27 * gmap_alloc - allocate and initialize a guest address space
28 * @mm: pointer to the parent mm_struct
29 * @limit: maximum address of the gmap address space
30 *
31 * Returns a guest address space structure.
32 */
33 static struct gmap *gmap_alloc(unsigned long limit)
34 {
35 struct gmap *gmap;
36 struct page *page;
37 unsigned long *table;
38 unsigned long etype, atype;
39
40 if (limit < _REGION3_SIZE) {
41 limit = _REGION3_SIZE - 1;
42 atype = _ASCE_TYPE_SEGMENT;
43 etype = _SEGMENT_ENTRY_EMPTY;
44 } else if (limit < _REGION2_SIZE) {
45 limit = _REGION2_SIZE - 1;
46 atype = _ASCE_TYPE_REGION3;
47 etype = _REGION3_ENTRY_EMPTY;
48 } else if (limit < _REGION1_SIZE) {
49 limit = _REGION1_SIZE - 1;
50 atype = _ASCE_TYPE_REGION2;
51 etype = _REGION2_ENTRY_EMPTY;
52 } else {
53 limit = -1UL;
54 atype = _ASCE_TYPE_REGION1;
55 etype = _REGION1_ENTRY_EMPTY;
56 }
57 gmap = kzalloc(sizeof(struct gmap), GFP_KERNEL);
58 if (!gmap)
59 goto out;
60 INIT_LIST_HEAD(&gmap->crst_list);
61 INIT_LIST_HEAD(&gmap->children);
62 INIT_LIST_HEAD(&gmap->pt_list);
63 INIT_RADIX_TREE(&gmap->guest_to_host, GFP_KERNEL);
64 INIT_RADIX_TREE(&gmap->host_to_guest, GFP_ATOMIC);
65 INIT_RADIX_TREE(&gmap->host_to_rmap, GFP_ATOMIC);
66 spin_lock_init(&gmap->guest_table_lock);
67 spin_lock_init(&gmap->shadow_lock);
68 atomic_set(&gmap->ref_count, 1);
69 page = alloc_pages(GFP_KERNEL, CRST_ALLOC_ORDER);
70 if (!page)
71 goto out_free;
72 page->index = 0;
73 list_add(&page->lru, &gmap->crst_list);
74 table = (unsigned long *) page_to_phys(page);
75 crst_table_init(table, etype);
76 gmap->table = table;
77 gmap->asce = atype | _ASCE_TABLE_LENGTH |
78 _ASCE_USER_BITS | __pa(table);
79 gmap->asce_end = limit;
80 return gmap;
81
82 out_free:
83 kfree(gmap);
84 out:
85 return NULL;
86 }
87
88 /**
89 * gmap_create - create a guest address space
90 * @mm: pointer to the parent mm_struct
91 * @limit: maximum size of the gmap address space
92 *
93 * Returns a guest address space structure.
94 */
95 struct gmap *gmap_create(struct mm_struct *mm, unsigned long limit)
96 {
97 struct gmap *gmap;
98 unsigned long gmap_asce;
99
100 gmap = gmap_alloc(limit);
101 if (!gmap)
102 return NULL;
103 gmap->mm = mm;
104 spin_lock(&mm->context.lock);
105 list_add_rcu(&gmap->list, &mm->context.gmap_list);
106 if (list_is_singular(&mm->context.gmap_list))
107 gmap_asce = gmap->asce;
108 else
109 gmap_asce = -1UL;
110 WRITE_ONCE(mm->context.gmap_asce, gmap_asce);
111 spin_unlock(&mm->context.lock);
112 return gmap;
113 }
114 EXPORT_SYMBOL_GPL(gmap_create);
115
116 static void gmap_flush_tlb(struct gmap *gmap)
117 {
118 if (MACHINE_HAS_IDTE)
119 __tlb_flush_idte(gmap->asce);
120 else
121 __tlb_flush_global();
122 }
123
124 static void gmap_radix_tree_free(struct radix_tree_root *root)
125 {
126 struct radix_tree_iter iter;
127 unsigned long indices[16];
128 unsigned long index;
129 void __rcu **slot;
130 int i, nr;
131
132 /* A radix tree is freed by deleting all of its entries */
133 index = 0;
134 do {
135 nr = 0;
136 radix_tree_for_each_slot(slot, root, &iter, index) {
137 indices[nr] = iter.index;
138 if (++nr == 16)
139 break;
140 }
141 for (i = 0; i < nr; i++) {
142 index = indices[i];
143 radix_tree_delete(root, index);
144 }
145 } while (nr > 0);
146 }
147
148 static void gmap_rmap_radix_tree_free(struct radix_tree_root *root)
149 {
150 struct gmap_rmap *rmap, *rnext, *head;
151 struct radix_tree_iter iter;
152 unsigned long indices[16];
153 unsigned long index;
154 void __rcu **slot;
155 int i, nr;
156
157 /* A radix tree is freed by deleting all of its entries */
158 index = 0;
159 do {
160 nr = 0;
161 radix_tree_for_each_slot(slot, root, &iter, index) {
162 indices[nr] = iter.index;
163 if (++nr == 16)
164 break;
165 }
166 for (i = 0; i < nr; i++) {
167 index = indices[i];
168 head = radix_tree_delete(root, index);
169 gmap_for_each_rmap_safe(rmap, rnext, head)
170 kfree(rmap);
171 }
172 } while (nr > 0);
173 }
174
175 /**
176 * gmap_free - free a guest address space
177 * @gmap: pointer to the guest address space structure
178 *
179 * No locks required. There are no references to this gmap anymore.
180 */
181 static void gmap_free(struct gmap *gmap)
182 {
183 struct page *page, *next;
184
185 /* Flush tlb of all gmaps (if not already done for shadows) */
186 if (!(gmap_is_shadow(gmap) && gmap->removed))
187 gmap_flush_tlb(gmap);
188 /* Free all segment & region tables. */
189 list_for_each_entry_safe(page, next, &gmap->crst_list, lru)
190 __free_pages(page, CRST_ALLOC_ORDER);
191 gmap_radix_tree_free(&gmap->guest_to_host);
192 gmap_radix_tree_free(&gmap->host_to_guest);
193
194 /* Free additional data for a shadow gmap */
195 if (gmap_is_shadow(gmap)) {
196 /* Free all page tables. */
197 list_for_each_entry_safe(page, next, &gmap->pt_list, lru)
198 page_table_free_pgste(page);
199 gmap_rmap_radix_tree_free(&gmap->host_to_rmap);
200 /* Release reference to the parent */
201 gmap_put(gmap->parent);
202 }
203
204 kfree(gmap);
205 }
206
207 /**
208 * gmap_get - increase reference counter for guest address space
209 * @gmap: pointer to the guest address space structure
210 *
211 * Returns the gmap pointer
212 */
213 struct gmap *gmap_get(struct gmap *gmap)
214 {
215 atomic_inc(&gmap->ref_count);
216 return gmap;
217 }
218 EXPORT_SYMBOL_GPL(gmap_get);
219
220 /**
221 * gmap_put - decrease reference counter for guest address space
222 * @gmap: pointer to the guest address space structure
223 *
224 * If the reference counter reaches zero the guest address space is freed.
225 */
226 void gmap_put(struct gmap *gmap)
227 {
228 if (atomic_dec_return(&gmap->ref_count) == 0)
229 gmap_free(gmap);
230 }
231 EXPORT_SYMBOL_GPL(gmap_put);
232
233 /**
234 * gmap_remove - remove a guest address space but do not free it yet
235 * @gmap: pointer to the guest address space structure
236 */
237 void gmap_remove(struct gmap *gmap)
238 {
239 struct gmap *sg, *next;
240 unsigned long gmap_asce;
241
242 /* Remove all shadow gmaps linked to this gmap */
243 if (!list_empty(&gmap->children)) {
244 spin_lock(&gmap->shadow_lock);
245 list_for_each_entry_safe(sg, next, &gmap->children, list) {
246 list_del(&sg->list);
247 gmap_put(sg);
248 }
249 spin_unlock(&gmap->shadow_lock);
250 }
251 /* Remove gmap from the pre-mm list */
252 spin_lock(&gmap->mm->context.lock);
253 list_del_rcu(&gmap->list);
254 if (list_empty(&gmap->mm->context.gmap_list))
255 gmap_asce = 0;
256 else if (list_is_singular(&gmap->mm->context.gmap_list))
257 gmap_asce = list_first_entry(&gmap->mm->context.gmap_list,
258 struct gmap, list)->asce;
259 else
260 gmap_asce = -1UL;
261 WRITE_ONCE(gmap->mm->context.gmap_asce, gmap_asce);
262 spin_unlock(&gmap->mm->context.lock);
263 synchronize_rcu();
264 /* Put reference */
265 gmap_put(gmap);
266 }
267 EXPORT_SYMBOL_GPL(gmap_remove);
268
269 /**
270 * gmap_enable - switch primary space to the guest address space
271 * @gmap: pointer to the guest address space structure
272 */
273 void gmap_enable(struct gmap *gmap)
274 {
275 S390_lowcore.gmap = (unsigned long) gmap;
276 }
277 EXPORT_SYMBOL_GPL(gmap_enable);
278
279 /**
280 * gmap_disable - switch back to the standard primary address space
281 * @gmap: pointer to the guest address space structure
282 */
283 void gmap_disable(struct gmap *gmap)
284 {
285 S390_lowcore.gmap = 0UL;
286 }
287 EXPORT_SYMBOL_GPL(gmap_disable);
288
289 /**
290 * gmap_get_enabled - get a pointer to the currently enabled gmap
291 *
292 * Returns a pointer to the currently enabled gmap. 0 if none is enabled.
293 */
294 struct gmap *gmap_get_enabled(void)
295 {
296 return (struct gmap *) S390_lowcore.gmap;
297 }
298 EXPORT_SYMBOL_GPL(gmap_get_enabled);
299
300 /*
301 * gmap_alloc_table is assumed to be called with mmap_sem held
302 */
303 static int gmap_alloc_table(struct gmap *gmap, unsigned long *table,
304 unsigned long init, unsigned long gaddr)
305 {
306 struct page *page;
307 unsigned long *new;
308
309 /* since we dont free the gmap table until gmap_free we can unlock */
310 page = alloc_pages(GFP_KERNEL, CRST_ALLOC_ORDER);
311 if (!page)
312 return -ENOMEM;
313 new = (unsigned long *) page_to_phys(page);
314 crst_table_init(new, init);
315 spin_lock(&gmap->guest_table_lock);
316 if (*table & _REGION_ENTRY_INVALID) {
317 list_add(&page->lru, &gmap->crst_list);
318 *table = (unsigned long) new | _REGION_ENTRY_LENGTH |
319 (*table & _REGION_ENTRY_TYPE_MASK);
320 page->index = gaddr;
321 page = NULL;
322 }
323 spin_unlock(&gmap->guest_table_lock);
324 if (page)
325 __free_pages(page, CRST_ALLOC_ORDER);
326 return 0;
327 }
328
329 /**
330 * __gmap_segment_gaddr - find virtual address from segment pointer
331 * @entry: pointer to a segment table entry in the guest address space
332 *
333 * Returns the virtual address in the guest address space for the segment
334 */
335 static unsigned long __gmap_segment_gaddr(unsigned long *entry)
336 {
337 struct page *page;
338 unsigned long offset, mask;
339
340 offset = (unsigned long) entry / sizeof(unsigned long);
341 offset = (offset & (PTRS_PER_PMD - 1)) * PMD_SIZE;
342 mask = ~(PTRS_PER_PMD * sizeof(pmd_t) - 1);
343 page = virt_to_page((void *)((unsigned long) entry & mask));
344 return page->index + offset;
345 }
346
347 /**
348 * __gmap_unlink_by_vmaddr - unlink a single segment via a host address
349 * @gmap: pointer to the guest address space structure
350 * @vmaddr: address in the host process address space
351 *
352 * Returns 1 if a TLB flush is required
353 */
354 static int __gmap_unlink_by_vmaddr(struct gmap *gmap, unsigned long vmaddr)
355 {
356 unsigned long *entry;
357 int flush = 0;
358
359 BUG_ON(gmap_is_shadow(gmap));
360 spin_lock(&gmap->guest_table_lock);
361 entry = radix_tree_delete(&gmap->host_to_guest, vmaddr >> PMD_SHIFT);
362 if (entry) {
363 flush = (*entry != _SEGMENT_ENTRY_EMPTY);
364 *entry = _SEGMENT_ENTRY_EMPTY;
365 }
366 spin_unlock(&gmap->guest_table_lock);
367 return flush;
368 }
369
370 /**
371 * __gmap_unmap_by_gaddr - unmap a single segment via a guest address
372 * @gmap: pointer to the guest address space structure
373 * @gaddr: address in the guest address space
374 *
375 * Returns 1 if a TLB flush is required
376 */
377 static int __gmap_unmap_by_gaddr(struct gmap *gmap, unsigned long gaddr)
378 {
379 unsigned long vmaddr;
380
381 vmaddr = (unsigned long) radix_tree_delete(&gmap->guest_to_host,
382 gaddr >> PMD_SHIFT);
383 return vmaddr ? __gmap_unlink_by_vmaddr(gmap, vmaddr) : 0;
384 }
385
386 /**
387 * gmap_unmap_segment - unmap segment from the guest address space
388 * @gmap: pointer to the guest address space structure
389 * @to: address in the guest address space
390 * @len: length of the memory area to unmap
391 *
392 * Returns 0 if the unmap succeeded, -EINVAL if not.
393 */
394 int gmap_unmap_segment(struct gmap *gmap, unsigned long to, unsigned long len)
395 {
396 unsigned long off;
397 int flush;
398
399 BUG_ON(gmap_is_shadow(gmap));
400 if ((to | len) & (PMD_SIZE - 1))
401 return -EINVAL;
402 if (len == 0 || to + len < to)
403 return -EINVAL;
404
405 flush = 0;
406 down_write(&gmap->mm->mmap_sem);
407 for (off = 0; off < len; off += PMD_SIZE)
408 flush |= __gmap_unmap_by_gaddr(gmap, to + off);
409 up_write(&gmap->mm->mmap_sem);
410 if (flush)
411 gmap_flush_tlb(gmap);
412 return 0;
413 }
414 EXPORT_SYMBOL_GPL(gmap_unmap_segment);
415
416 /**
417 * gmap_map_segment - map a segment to the guest address space
418 * @gmap: pointer to the guest address space structure
419 * @from: source address in the parent address space
420 * @to: target address in the guest address space
421 * @len: length of the memory area to map
422 *
423 * Returns 0 if the mmap succeeded, -EINVAL or -ENOMEM if not.
424 */
425 int gmap_map_segment(struct gmap *gmap, unsigned long from,
426 unsigned long to, unsigned long len)
427 {
428 unsigned long off;
429 int flush;
430
431 BUG_ON(gmap_is_shadow(gmap));
432 if ((from | to | len) & (PMD_SIZE - 1))
433 return -EINVAL;
434 if (len == 0 || from + len < from || to + len < to ||
435 from + len - 1 > TASK_SIZE_MAX || to + len - 1 > gmap->asce_end)
436 return -EINVAL;
437
438 flush = 0;
439 down_write(&gmap->mm->mmap_sem);
440 for (off = 0; off < len; off += PMD_SIZE) {
441 /* Remove old translation */
442 flush |= __gmap_unmap_by_gaddr(gmap, to + off);
443 /* Store new translation */
444 if (radix_tree_insert(&gmap->guest_to_host,
445 (to + off) >> PMD_SHIFT,
446 (void *) from + off))
447 break;
448 }
449 up_write(&gmap->mm->mmap_sem);
450 if (flush)
451 gmap_flush_tlb(gmap);
452 if (off >= len)
453 return 0;
454 gmap_unmap_segment(gmap, to, len);
455 return -ENOMEM;
456 }
457 EXPORT_SYMBOL_GPL(gmap_map_segment);
458
459 /**
460 * __gmap_translate - translate a guest address to a user space address
461 * @gmap: pointer to guest mapping meta data structure
462 * @gaddr: guest address
463 *
464 * Returns user space address which corresponds to the guest address or
465 * -EFAULT if no such mapping exists.
466 * This function does not establish potentially missing page table entries.
467 * The mmap_sem of the mm that belongs to the address space must be held
468 * when this function gets called.
469 *
470 * Note: Can also be called for shadow gmaps.
471 */
472 unsigned long __gmap_translate(struct gmap *gmap, unsigned long gaddr)
473 {
474 unsigned long vmaddr;
475
476 vmaddr = (unsigned long)
477 radix_tree_lookup(&gmap->guest_to_host, gaddr >> PMD_SHIFT);
478 /* Note: guest_to_host is empty for a shadow gmap */
479 return vmaddr ? (vmaddr | (gaddr & ~PMD_MASK)) : -EFAULT;
480 }
481 EXPORT_SYMBOL_GPL(__gmap_translate);
482
483 /**
484 * gmap_translate - translate a guest address to a user space address
485 * @gmap: pointer to guest mapping meta data structure
486 * @gaddr: guest address
487 *
488 * Returns user space address which corresponds to the guest address or
489 * -EFAULT if no such mapping exists.
490 * This function does not establish potentially missing page table entries.
491 */
492 unsigned long gmap_translate(struct gmap *gmap, unsigned long gaddr)
493 {
494 unsigned long rc;
495
496 down_read(&gmap->mm->mmap_sem);
497 rc = __gmap_translate(gmap, gaddr);
498 up_read(&gmap->mm->mmap_sem);
499 return rc;
500 }
501 EXPORT_SYMBOL_GPL(gmap_translate);
502
503 /**
504 * gmap_unlink - disconnect a page table from the gmap shadow tables
505 * @gmap: pointer to guest mapping meta data structure
506 * @table: pointer to the host page table
507 * @vmaddr: vm address associated with the host page table
508 */
509 void gmap_unlink(struct mm_struct *mm, unsigned long *table,
510 unsigned long vmaddr)
511 {
512 struct gmap *gmap;
513 int flush;
514
515 rcu_read_lock();
516 list_for_each_entry_rcu(gmap, &mm->context.gmap_list, list) {
517 flush = __gmap_unlink_by_vmaddr(gmap, vmaddr);
518 if (flush)
519 gmap_flush_tlb(gmap);
520 }
521 rcu_read_unlock();
522 }
523
524 /**
525 * gmap_link - set up shadow page tables to connect a host to a guest address
526 * @gmap: pointer to guest mapping meta data structure
527 * @gaddr: guest address
528 * @vmaddr: vm address
529 *
530 * Returns 0 on success, -ENOMEM for out of memory conditions, and -EFAULT
531 * if the vm address is already mapped to a different guest segment.
532 * The mmap_sem of the mm that belongs to the address space must be held
533 * when this function gets called.
534 */
535 int __gmap_link(struct gmap *gmap, unsigned long gaddr, unsigned long vmaddr)
536 {
537 struct mm_struct *mm;
538 unsigned long *table;
539 spinlock_t *ptl;
540 pgd_t *pgd;
541 p4d_t *p4d;
542 pud_t *pud;
543 pmd_t *pmd;
544 int rc;
545
546 BUG_ON(gmap_is_shadow(gmap));
547 /* Create higher level tables in the gmap page table */
548 table = gmap->table;
549 if ((gmap->asce & _ASCE_TYPE_MASK) >= _ASCE_TYPE_REGION1) {
550 table += (gaddr & _REGION1_INDEX) >> _REGION1_SHIFT;
551 if ((*table & _REGION_ENTRY_INVALID) &&
552 gmap_alloc_table(gmap, table, _REGION2_ENTRY_EMPTY,
553 gaddr & _REGION1_MASK))
554 return -ENOMEM;
555 table = (unsigned long *)(*table & _REGION_ENTRY_ORIGIN);
556 }
557 if ((gmap->asce & _ASCE_TYPE_MASK) >= _ASCE_TYPE_REGION2) {
558 table += (gaddr & _REGION2_INDEX) >> _REGION2_SHIFT;
559 if ((*table & _REGION_ENTRY_INVALID) &&
560 gmap_alloc_table(gmap, table, _REGION3_ENTRY_EMPTY,
561 gaddr & _REGION2_MASK))
562 return -ENOMEM;
563 table = (unsigned long *)(*table & _REGION_ENTRY_ORIGIN);
564 }
565 if ((gmap->asce & _ASCE_TYPE_MASK) >= _ASCE_TYPE_REGION3) {
566 table += (gaddr & _REGION3_INDEX) >> _REGION3_SHIFT;
567 if ((*table & _REGION_ENTRY_INVALID) &&
568 gmap_alloc_table(gmap, table, _SEGMENT_ENTRY_EMPTY,
569 gaddr & _REGION3_MASK))
570 return -ENOMEM;
571 table = (unsigned long *)(*table & _REGION_ENTRY_ORIGIN);
572 }
573 table += (gaddr & _SEGMENT_INDEX) >> _SEGMENT_SHIFT;
574 /* Walk the parent mm page table */
575 mm = gmap->mm;
576 pgd = pgd_offset(mm, vmaddr);
577 VM_BUG_ON(pgd_none(*pgd));
578 p4d = p4d_offset(pgd, vmaddr);
579 VM_BUG_ON(p4d_none(*p4d));
580 pud = pud_offset(p4d, vmaddr);
581 VM_BUG_ON(pud_none(*pud));
582 /* large puds cannot yet be handled */
583 if (pud_large(*pud))
584 return -EFAULT;
585 pmd = pmd_offset(pud, vmaddr);
586 VM_BUG_ON(pmd_none(*pmd));
587 /* large pmds cannot yet be handled */
588 if (pmd_large(*pmd))
589 return -EFAULT;
590 /* Link gmap segment table entry location to page table. */
591 rc = radix_tree_preload(GFP_KERNEL);
592 if (rc)
593 return rc;
594 ptl = pmd_lock(mm, pmd);
595 spin_lock(&gmap->guest_table_lock);
596 if (*table == _SEGMENT_ENTRY_EMPTY) {
597 rc = radix_tree_insert(&gmap->host_to_guest,
598 vmaddr >> PMD_SHIFT, table);
599 if (!rc)
600 *table = pmd_val(*pmd);
601 } else
602 rc = 0;
603 spin_unlock(&gmap->guest_table_lock);
604 spin_unlock(ptl);
605 radix_tree_preload_end();
606 return rc;
607 }
608
609 /**
610 * gmap_fault - resolve a fault on a guest address
611 * @gmap: pointer to guest mapping meta data structure
612 * @gaddr: guest address
613 * @fault_flags: flags to pass down to handle_mm_fault()
614 *
615 * Returns 0 on success, -ENOMEM for out of memory conditions, and -EFAULT
616 * if the vm address is already mapped to a different guest segment.
617 */
618 int gmap_fault(struct gmap *gmap, unsigned long gaddr,
619 unsigned int fault_flags)
620 {
621 unsigned long vmaddr;
622 int rc;
623 bool unlocked;
624
625 down_read(&gmap->mm->mmap_sem);
626
627 retry:
628 unlocked = false;
629 vmaddr = __gmap_translate(gmap, gaddr);
630 if (IS_ERR_VALUE(vmaddr)) {
631 rc = vmaddr;
632 goto out_up;
633 }
634 if (fixup_user_fault(current, gmap->mm, vmaddr, fault_flags,
635 &unlocked)) {
636 rc = -EFAULT;
637 goto out_up;
638 }
639 /*
640 * In the case that fixup_user_fault unlocked the mmap_sem during
641 * faultin redo __gmap_translate to not race with a map/unmap_segment.
642 */
643 if (unlocked)
644 goto retry;
645
646 rc = __gmap_link(gmap, gaddr, vmaddr);
647 out_up:
648 up_read(&gmap->mm->mmap_sem);
649 return rc;
650 }
651 EXPORT_SYMBOL_GPL(gmap_fault);
652
653 /*
654 * this function is assumed to be called with mmap_sem held
655 */
656 void __gmap_zap(struct gmap *gmap, unsigned long gaddr)
657 {
658 unsigned long vmaddr;
659 spinlock_t *ptl;
660 pte_t *ptep;
661
662 /* Find the vm address for the guest address */
663 vmaddr = (unsigned long) radix_tree_lookup(&gmap->guest_to_host,
664 gaddr >> PMD_SHIFT);
665 if (vmaddr) {
666 vmaddr |= gaddr & ~PMD_MASK;
667 /* Get pointer to the page table entry */
668 ptep = get_locked_pte(gmap->mm, vmaddr, &ptl);
669 if (likely(ptep))
670 ptep_zap_unused(gmap->mm, vmaddr, ptep, 0);
671 pte_unmap_unlock(ptep, ptl);
672 }
673 }
674 EXPORT_SYMBOL_GPL(__gmap_zap);
675
676 void gmap_discard(struct gmap *gmap, unsigned long from, unsigned long to)
677 {
678 unsigned long gaddr, vmaddr, size;
679 struct vm_area_struct *vma;
680
681 down_read(&gmap->mm->mmap_sem);
682 for (gaddr = from; gaddr < to;
683 gaddr = (gaddr + PMD_SIZE) & PMD_MASK) {
684 /* Find the vm address for the guest address */
685 vmaddr = (unsigned long)
686 radix_tree_lookup(&gmap->guest_to_host,
687 gaddr >> PMD_SHIFT);
688 if (!vmaddr)
689 continue;
690 vmaddr |= gaddr & ~PMD_MASK;
691 /* Find vma in the parent mm */
692 vma = find_vma(gmap->mm, vmaddr);
693 size = min(to - gaddr, PMD_SIZE - (gaddr & ~PMD_MASK));
694 zap_page_range(vma, vmaddr, size);
695 }
696 up_read(&gmap->mm->mmap_sem);
697 }
698 EXPORT_SYMBOL_GPL(gmap_discard);
699
700 static LIST_HEAD(gmap_notifier_list);
701 static DEFINE_SPINLOCK(gmap_notifier_lock);
702
703 /**
704 * gmap_register_pte_notifier - register a pte invalidation callback
705 * @nb: pointer to the gmap notifier block
706 */
707 void gmap_register_pte_notifier(struct gmap_notifier *nb)
708 {
709 spin_lock(&gmap_notifier_lock);
710 list_add_rcu(&nb->list, &gmap_notifier_list);
711 spin_unlock(&gmap_notifier_lock);
712 }
713 EXPORT_SYMBOL_GPL(gmap_register_pte_notifier);
714
715 /**
716 * gmap_unregister_pte_notifier - remove a pte invalidation callback
717 * @nb: pointer to the gmap notifier block
718 */
719 void gmap_unregister_pte_notifier(struct gmap_notifier *nb)
720 {
721 spin_lock(&gmap_notifier_lock);
722 list_del_rcu(&nb->list);
723 spin_unlock(&gmap_notifier_lock);
724 synchronize_rcu();
725 }
726 EXPORT_SYMBOL_GPL(gmap_unregister_pte_notifier);
727
728 /**
729 * gmap_call_notifier - call all registered invalidation callbacks
730 * @gmap: pointer to guest mapping meta data structure
731 * @start: start virtual address in the guest address space
732 * @end: end virtual address in the guest address space
733 */
734 static void gmap_call_notifier(struct gmap *gmap, unsigned long start,
735 unsigned long end)
736 {
737 struct gmap_notifier *nb;
738
739 list_for_each_entry(nb, &gmap_notifier_list, list)
740 nb->notifier_call(gmap, start, end);
741 }
742
743 /**
744 * gmap_table_walk - walk the gmap page tables
745 * @gmap: pointer to guest mapping meta data structure
746 * @gaddr: virtual address in the guest address space
747 * @level: page table level to stop at
748 *
749 * Returns a table entry pointer for the given guest address and @level
750 * @level=0 : returns a pointer to a page table table entry (or NULL)
751 * @level=1 : returns a pointer to a segment table entry (or NULL)
752 * @level=2 : returns a pointer to a region-3 table entry (or NULL)
753 * @level=3 : returns a pointer to a region-2 table entry (or NULL)
754 * @level=4 : returns a pointer to a region-1 table entry (or NULL)
755 *
756 * Returns NULL if the gmap page tables could not be walked to the
757 * requested level.
758 *
759 * Note: Can also be called for shadow gmaps.
760 */
761 static inline unsigned long *gmap_table_walk(struct gmap *gmap,
762 unsigned long gaddr, int level)
763 {
764 unsigned long *table;
765
766 if ((gmap->asce & _ASCE_TYPE_MASK) + 4 < (level * 4))
767 return NULL;
768 if (gmap_is_shadow(gmap) && gmap->removed)
769 return NULL;
770 if (gaddr & (-1UL << (31 + ((gmap->asce & _ASCE_TYPE_MASK) >> 2)*11)))
771 return NULL;
772 table = gmap->table;
773 switch (gmap->asce & _ASCE_TYPE_MASK) {
774 case _ASCE_TYPE_REGION1:
775 table += (gaddr & _REGION1_INDEX) >> _REGION1_SHIFT;
776 if (level == 4)
777 break;
778 if (*table & _REGION_ENTRY_INVALID)
779 return NULL;
780 table = (unsigned long *)(*table & _REGION_ENTRY_ORIGIN);
781 /* Fallthrough */
782 case _ASCE_TYPE_REGION2:
783 table += (gaddr & _REGION2_INDEX) >> _REGION2_SHIFT;
784 if (level == 3)
785 break;
786 if (*table & _REGION_ENTRY_INVALID)
787 return NULL;
788 table = (unsigned long *)(*table & _REGION_ENTRY_ORIGIN);
789 /* Fallthrough */
790 case _ASCE_TYPE_REGION3:
791 table += (gaddr & _REGION3_INDEX) >> _REGION3_SHIFT;
792 if (level == 2)
793 break;
794 if (*table & _REGION_ENTRY_INVALID)
795 return NULL;
796 table = (unsigned long *)(*table & _REGION_ENTRY_ORIGIN);
797 /* Fallthrough */
798 case _ASCE_TYPE_SEGMENT:
799 table += (gaddr & _SEGMENT_INDEX) >> _SEGMENT_SHIFT;
800 if (level == 1)
801 break;
802 if (*table & _REGION_ENTRY_INVALID)
803 return NULL;
804 table = (unsigned long *)(*table & _SEGMENT_ENTRY_ORIGIN);
805 table += (gaddr & _PAGE_INDEX) >> _PAGE_SHIFT;
806 }
807 return table;
808 }
809
810 /**
811 * gmap_pte_op_walk - walk the gmap page table, get the page table lock
812 * and return the pte pointer
813 * @gmap: pointer to guest mapping meta data structure
814 * @gaddr: virtual address in the guest address space
815 * @ptl: pointer to the spinlock pointer
816 *
817 * Returns a pointer to the locked pte for a guest address, or NULL
818 *
819 * Note: Can also be called for shadow gmaps.
820 */
821 static pte_t *gmap_pte_op_walk(struct gmap *gmap, unsigned long gaddr,
822 spinlock_t **ptl)
823 {
824 unsigned long *table;
825
826 if (gmap_is_shadow(gmap))
827 spin_lock(&gmap->guest_table_lock);
828 /* Walk the gmap page table, lock and get pte pointer */
829 table = gmap_table_walk(gmap, gaddr, 1); /* get segment pointer */
830 if (!table || *table & _SEGMENT_ENTRY_INVALID) {
831 if (gmap_is_shadow(gmap))
832 spin_unlock(&gmap->guest_table_lock);
833 return NULL;
834 }
835 if (gmap_is_shadow(gmap)) {
836 *ptl = &gmap->guest_table_lock;
837 return pte_offset_map((pmd_t *) table, gaddr);
838 }
839 return pte_alloc_map_lock(gmap->mm, (pmd_t *) table, gaddr, ptl);
840 }
841
842 /**
843 * gmap_pte_op_fixup - force a page in and connect the gmap page table
844 * @gmap: pointer to guest mapping meta data structure
845 * @gaddr: virtual address in the guest address space
846 * @vmaddr: address in the host process address space
847 * @prot: indicates access rights: PROT_NONE, PROT_READ or PROT_WRITE
848 *
849 * Returns 0 if the caller can retry __gmap_translate (might fail again),
850 * -ENOMEM if out of memory and -EFAULT if anything goes wrong while fixing
851 * up or connecting the gmap page table.
852 */
853 static int gmap_pte_op_fixup(struct gmap *gmap, unsigned long gaddr,
854 unsigned long vmaddr, int prot)
855 {
856 struct mm_struct *mm = gmap->mm;
857 unsigned int fault_flags;
858 bool unlocked = false;
859
860 BUG_ON(gmap_is_shadow(gmap));
861 fault_flags = (prot == PROT_WRITE) ? FAULT_FLAG_WRITE : 0;
862 if (fixup_user_fault(current, mm, vmaddr, fault_flags, &unlocked))
863 return -EFAULT;
864 if (unlocked)
865 /* lost mmap_sem, caller has to retry __gmap_translate */
866 return 0;
867 /* Connect the page tables */
868 return __gmap_link(gmap, gaddr, vmaddr);
869 }
870
871 /**
872 * gmap_pte_op_end - release the page table lock
873 * @ptl: pointer to the spinlock pointer
874 */
875 static void gmap_pte_op_end(spinlock_t *ptl)
876 {
877 spin_unlock(ptl);
878 }
879
880 /*
881 * gmap_protect_range - remove access rights to memory and set pgste bits
882 * @gmap: pointer to guest mapping meta data structure
883 * @gaddr: virtual address in the guest address space
884 * @len: size of area
885 * @prot: indicates access rights: PROT_NONE, PROT_READ or PROT_WRITE
886 * @bits: pgste notification bits to set
887 *
888 * Returns 0 if successfully protected, -ENOMEM if out of memory and
889 * -EFAULT if gaddr is invalid (or mapping for shadows is missing).
890 *
891 * Called with sg->mm->mmap_sem in read.
892 *
893 * Note: Can also be called for shadow gmaps.
894 */
895 static int gmap_protect_range(struct gmap *gmap, unsigned long gaddr,
896 unsigned long len, int prot, unsigned long bits)
897 {
898 unsigned long vmaddr;
899 spinlock_t *ptl;
900 pte_t *ptep;
901 int rc;
902
903 while (len) {
904 rc = -EAGAIN;
905 ptep = gmap_pte_op_walk(gmap, gaddr, &ptl);
906 if (ptep) {
907 rc = ptep_force_prot(gmap->mm, gaddr, ptep, prot, bits);
908 gmap_pte_op_end(ptl);
909 }
910 if (rc) {
911 vmaddr = __gmap_translate(gmap, gaddr);
912 if (IS_ERR_VALUE(vmaddr))
913 return vmaddr;
914 rc = gmap_pte_op_fixup(gmap, gaddr, vmaddr, prot);
915 if (rc)
916 return rc;
917 continue;
918 }
919 gaddr += PAGE_SIZE;
920 len -= PAGE_SIZE;
921 }
922 return 0;
923 }
924
925 /**
926 * gmap_mprotect_notify - change access rights for a range of ptes and
927 * call the notifier if any pte changes again
928 * @gmap: pointer to guest mapping meta data structure
929 * @gaddr: virtual address in the guest address space
930 * @len: size of area
931 * @prot: indicates access rights: PROT_NONE, PROT_READ or PROT_WRITE
932 *
933 * Returns 0 if for each page in the given range a gmap mapping exists,
934 * the new access rights could be set and the notifier could be armed.
935 * If the gmap mapping is missing for one or more pages -EFAULT is
936 * returned. If no memory could be allocated -ENOMEM is returned.
937 * This function establishes missing page table entries.
938 */
939 int gmap_mprotect_notify(struct gmap *gmap, unsigned long gaddr,
940 unsigned long len, int prot)
941 {
942 int rc;
943
944 if ((gaddr & ~PAGE_MASK) || (len & ~PAGE_MASK) || gmap_is_shadow(gmap))
945 return -EINVAL;
946 if (!MACHINE_HAS_ESOP && prot == PROT_READ)
947 return -EINVAL;
948 down_read(&gmap->mm->mmap_sem);
949 rc = gmap_protect_range(gmap, gaddr, len, prot, PGSTE_IN_BIT);
950 up_read(&gmap->mm->mmap_sem);
951 return rc;
952 }
953 EXPORT_SYMBOL_GPL(gmap_mprotect_notify);
954
955 /**
956 * gmap_read_table - get an unsigned long value from a guest page table using
957 * absolute addressing, without marking the page referenced.
958 * @gmap: pointer to guest mapping meta data structure
959 * @gaddr: virtual address in the guest address space
960 * @val: pointer to the unsigned long value to return
961 *
962 * Returns 0 if the value was read, -ENOMEM if out of memory and -EFAULT
963 * if reading using the virtual address failed.
964 *
965 * Called with gmap->mm->mmap_sem in read.
966 */
967 int gmap_read_table(struct gmap *gmap, unsigned long gaddr, unsigned long *val)
968 {
969 unsigned long address, vmaddr;
970 spinlock_t *ptl;
971 pte_t *ptep, pte;
972 int rc;
973
974 while (1) {
975 rc = -EAGAIN;
976 ptep = gmap_pte_op_walk(gmap, gaddr, &ptl);
977 if (ptep) {
978 pte = *ptep;
979 if (pte_present(pte) && (pte_val(pte) & _PAGE_READ)) {
980 address = pte_val(pte) & PAGE_MASK;
981 address += gaddr & ~PAGE_MASK;
982 *val = *(unsigned long *) address;
983 pte_val(*ptep) |= _PAGE_YOUNG;
984 /* Do *NOT* clear the _PAGE_INVALID bit! */
985 rc = 0;
986 }
987 gmap_pte_op_end(ptl);
988 }
989 if (!rc)
990 break;
991 vmaddr = __gmap_translate(gmap, gaddr);
992 if (IS_ERR_VALUE(vmaddr)) {
993 rc = vmaddr;
994 break;
995 }
996 rc = gmap_pte_op_fixup(gmap, gaddr, vmaddr, PROT_READ);
997 if (rc)
998 break;
999 }
1000 return rc;
1001 }
1002 EXPORT_SYMBOL_GPL(gmap_read_table);
1003
1004 /**
1005 * gmap_insert_rmap - add a rmap to the host_to_rmap radix tree
1006 * @sg: pointer to the shadow guest address space structure
1007 * @vmaddr: vm address associated with the rmap
1008 * @rmap: pointer to the rmap structure
1009 *
1010 * Called with the sg->guest_table_lock
1011 */
1012 static inline void gmap_insert_rmap(struct gmap *sg, unsigned long vmaddr,
1013 struct gmap_rmap *rmap)
1014 {
1015 void __rcu **slot;
1016
1017 BUG_ON(!gmap_is_shadow(sg));
1018 slot = radix_tree_lookup_slot(&sg->host_to_rmap, vmaddr >> PAGE_SHIFT);
1019 if (slot) {
1020 rmap->next = radix_tree_deref_slot_protected(slot,
1021 &sg->guest_table_lock);
1022 radix_tree_replace_slot(&sg->host_to_rmap, slot, rmap);
1023 } else {
1024 rmap->next = NULL;
1025 radix_tree_insert(&sg->host_to_rmap, vmaddr >> PAGE_SHIFT,
1026 rmap);
1027 }
1028 }
1029
1030 /**
1031 * gmap_protect_rmap - modify access rights to memory and create an rmap
1032 * @sg: pointer to the shadow guest address space structure
1033 * @raddr: rmap address in the shadow gmap
1034 * @paddr: address in the parent guest address space
1035 * @len: length of the memory area to protect
1036 * @prot: indicates access rights: none, read-only or read-write
1037 *
1038 * Returns 0 if successfully protected and the rmap was created, -ENOMEM
1039 * if out of memory and -EFAULT if paddr is invalid.
1040 */
1041 static int gmap_protect_rmap(struct gmap *sg, unsigned long raddr,
1042 unsigned long paddr, unsigned long len, int prot)
1043 {
1044 struct gmap *parent;
1045 struct gmap_rmap *rmap;
1046 unsigned long vmaddr;
1047 spinlock_t *ptl;
1048 pte_t *ptep;
1049 int rc;
1050
1051 BUG_ON(!gmap_is_shadow(sg));
1052 parent = sg->parent;
1053 while (len) {
1054 vmaddr = __gmap_translate(parent, paddr);
1055 if (IS_ERR_VALUE(vmaddr))
1056 return vmaddr;
1057 rmap = kzalloc(sizeof(*rmap), GFP_KERNEL);
1058 if (!rmap)
1059 return -ENOMEM;
1060 rmap->raddr = raddr;
1061 rc = radix_tree_preload(GFP_KERNEL);
1062 if (rc) {
1063 kfree(rmap);
1064 return rc;
1065 }
1066 rc = -EAGAIN;
1067 ptep = gmap_pte_op_walk(parent, paddr, &ptl);
1068 if (ptep) {
1069 spin_lock(&sg->guest_table_lock);
1070 rc = ptep_force_prot(parent->mm, paddr, ptep, prot,
1071 PGSTE_VSIE_BIT);
1072 if (!rc)
1073 gmap_insert_rmap(sg, vmaddr, rmap);
1074 spin_unlock(&sg->guest_table_lock);
1075 gmap_pte_op_end(ptl);
1076 }
1077 radix_tree_preload_end();
1078 if (rc) {
1079 kfree(rmap);
1080 rc = gmap_pte_op_fixup(parent, paddr, vmaddr, prot);
1081 if (rc)
1082 return rc;
1083 continue;
1084 }
1085 paddr += PAGE_SIZE;
1086 len -= PAGE_SIZE;
1087 }
1088 return 0;
1089 }
1090
1091 #define _SHADOW_RMAP_MASK 0x7
1092 #define _SHADOW_RMAP_REGION1 0x5
1093 #define _SHADOW_RMAP_REGION2 0x4
1094 #define _SHADOW_RMAP_REGION3 0x3
1095 #define _SHADOW_RMAP_SEGMENT 0x2
1096 #define _SHADOW_RMAP_PGTABLE 0x1
1097
1098 /**
1099 * gmap_idte_one - invalidate a single region or segment table entry
1100 * @asce: region or segment table *origin* + table-type bits
1101 * @vaddr: virtual address to identify the table entry to flush
1102 *
1103 * The invalid bit of a single region or segment table entry is set
1104 * and the associated TLB entries depending on the entry are flushed.
1105 * The table-type of the @asce identifies the portion of the @vaddr
1106 * that is used as the invalidation index.
1107 */
1108 static inline void gmap_idte_one(unsigned long asce, unsigned long vaddr)
1109 {
1110 asm volatile(
1111 " .insn rrf,0xb98e0000,%0,%1,0,0"
1112 : : "a" (asce), "a" (vaddr) : "cc", "memory");
1113 }
1114
1115 /**
1116 * gmap_unshadow_page - remove a page from a shadow page table
1117 * @sg: pointer to the shadow guest address space structure
1118 * @raddr: rmap address in the shadow guest address space
1119 *
1120 * Called with the sg->guest_table_lock
1121 */
1122 static void gmap_unshadow_page(struct gmap *sg, unsigned long raddr)
1123 {
1124 unsigned long *table;
1125
1126 BUG_ON(!gmap_is_shadow(sg));
1127 table = gmap_table_walk(sg, raddr, 0); /* get page table pointer */
1128 if (!table || *table & _PAGE_INVALID)
1129 return;
1130 gmap_call_notifier(sg, raddr, raddr + _PAGE_SIZE - 1);
1131 ptep_unshadow_pte(sg->mm, raddr, (pte_t *) table);
1132 }
1133
1134 /**
1135 * __gmap_unshadow_pgt - remove all entries from a shadow page table
1136 * @sg: pointer to the shadow guest address space structure
1137 * @raddr: rmap address in the shadow guest address space
1138 * @pgt: pointer to the start of a shadow page table
1139 *
1140 * Called with the sg->guest_table_lock
1141 */
1142 static void __gmap_unshadow_pgt(struct gmap *sg, unsigned long raddr,
1143 unsigned long *pgt)
1144 {
1145 int i;
1146
1147 BUG_ON(!gmap_is_shadow(sg));
1148 for (i = 0; i < _PAGE_ENTRIES; i++, raddr += _PAGE_SIZE)
1149 pgt[i] = _PAGE_INVALID;
1150 }
1151
1152 /**
1153 * gmap_unshadow_pgt - remove a shadow page table from a segment entry
1154 * @sg: pointer to the shadow guest address space structure
1155 * @raddr: address in the shadow guest address space
1156 *
1157 * Called with the sg->guest_table_lock
1158 */
1159 static void gmap_unshadow_pgt(struct gmap *sg, unsigned long raddr)
1160 {
1161 unsigned long sto, *ste, *pgt;
1162 struct page *page;
1163
1164 BUG_ON(!gmap_is_shadow(sg));
1165 ste = gmap_table_walk(sg, raddr, 1); /* get segment pointer */
1166 if (!ste || !(*ste & _SEGMENT_ENTRY_ORIGIN))
1167 return;
1168 gmap_call_notifier(sg, raddr, raddr + _SEGMENT_SIZE - 1);
1169 sto = (unsigned long) (ste - ((raddr & _SEGMENT_INDEX) >> _SEGMENT_SHIFT));
1170 gmap_idte_one(sto | _ASCE_TYPE_SEGMENT, raddr);
1171 pgt = (unsigned long *)(*ste & _SEGMENT_ENTRY_ORIGIN);
1172 *ste = _SEGMENT_ENTRY_EMPTY;
1173 __gmap_unshadow_pgt(sg, raddr, pgt);
1174 /* Free page table */
1175 page = pfn_to_page(__pa(pgt) >> PAGE_SHIFT);
1176 list_del(&page->lru);
1177 page_table_free_pgste(page);
1178 }
1179
1180 /**
1181 * __gmap_unshadow_sgt - remove all entries from a shadow segment table
1182 * @sg: pointer to the shadow guest address space structure
1183 * @raddr: rmap address in the shadow guest address space
1184 * @sgt: pointer to the start of a shadow segment table
1185 *
1186 * Called with the sg->guest_table_lock
1187 */
1188 static void __gmap_unshadow_sgt(struct gmap *sg, unsigned long raddr,
1189 unsigned long *sgt)
1190 {
1191 unsigned long *pgt;
1192 struct page *page;
1193 int i;
1194
1195 BUG_ON(!gmap_is_shadow(sg));
1196 for (i = 0; i < _CRST_ENTRIES; i++, raddr += _SEGMENT_SIZE) {
1197 if (!(sgt[i] & _SEGMENT_ENTRY_ORIGIN))
1198 continue;
1199 pgt = (unsigned long *)(sgt[i] & _REGION_ENTRY_ORIGIN);
1200 sgt[i] = _SEGMENT_ENTRY_EMPTY;
1201 __gmap_unshadow_pgt(sg, raddr, pgt);
1202 /* Free page table */
1203 page = pfn_to_page(__pa(pgt) >> PAGE_SHIFT);
1204 list_del(&page->lru);
1205 page_table_free_pgste(page);
1206 }
1207 }
1208
1209 /**
1210 * gmap_unshadow_sgt - remove a shadow segment table from a region-3 entry
1211 * @sg: pointer to the shadow guest address space structure
1212 * @raddr: rmap address in the shadow guest address space
1213 *
1214 * Called with the shadow->guest_table_lock
1215 */
1216 static void gmap_unshadow_sgt(struct gmap *sg, unsigned long raddr)
1217 {
1218 unsigned long r3o, *r3e, *sgt;
1219 struct page *page;
1220
1221 BUG_ON(!gmap_is_shadow(sg));
1222 r3e = gmap_table_walk(sg, raddr, 2); /* get region-3 pointer */
1223 if (!r3e || !(*r3e & _REGION_ENTRY_ORIGIN))
1224 return;
1225 gmap_call_notifier(sg, raddr, raddr + _REGION3_SIZE - 1);
1226 r3o = (unsigned long) (r3e - ((raddr & _REGION3_INDEX) >> _REGION3_SHIFT));
1227 gmap_idte_one(r3o | _ASCE_TYPE_REGION3, raddr);
1228 sgt = (unsigned long *)(*r3e & _REGION_ENTRY_ORIGIN);
1229 *r3e = _REGION3_ENTRY_EMPTY;
1230 __gmap_unshadow_sgt(sg, raddr, sgt);
1231 /* Free segment table */
1232 page = pfn_to_page(__pa(sgt) >> PAGE_SHIFT);
1233 list_del(&page->lru);
1234 __free_pages(page, CRST_ALLOC_ORDER);
1235 }
1236
1237 /**
1238 * __gmap_unshadow_r3t - remove all entries from a shadow region-3 table
1239 * @sg: pointer to the shadow guest address space structure
1240 * @raddr: address in the shadow guest address space
1241 * @r3t: pointer to the start of a shadow region-3 table
1242 *
1243 * Called with the sg->guest_table_lock
1244 */
1245 static void __gmap_unshadow_r3t(struct gmap *sg, unsigned long raddr,
1246 unsigned long *r3t)
1247 {
1248 unsigned long *sgt;
1249 struct page *page;
1250 int i;
1251
1252 BUG_ON(!gmap_is_shadow(sg));
1253 for (i = 0; i < _CRST_ENTRIES; i++, raddr += _REGION3_SIZE) {
1254 if (!(r3t[i] & _REGION_ENTRY_ORIGIN))
1255 continue;
1256 sgt = (unsigned long *)(r3t[i] & _REGION_ENTRY_ORIGIN);
1257 r3t[i] = _REGION3_ENTRY_EMPTY;
1258 __gmap_unshadow_sgt(sg, raddr, sgt);
1259 /* Free segment table */
1260 page = pfn_to_page(__pa(sgt) >> PAGE_SHIFT);
1261 list_del(&page->lru);
1262 __free_pages(page, CRST_ALLOC_ORDER);
1263 }
1264 }
1265
1266 /**
1267 * gmap_unshadow_r3t - remove a shadow region-3 table from a region-2 entry
1268 * @sg: pointer to the shadow guest address space structure
1269 * @raddr: rmap address in the shadow guest address space
1270 *
1271 * Called with the sg->guest_table_lock
1272 */
1273 static void gmap_unshadow_r3t(struct gmap *sg, unsigned long raddr)
1274 {
1275 unsigned long r2o, *r2e, *r3t;
1276 struct page *page;
1277
1278 BUG_ON(!gmap_is_shadow(sg));
1279 r2e = gmap_table_walk(sg, raddr, 3); /* get region-2 pointer */
1280 if (!r2e || !(*r2e & _REGION_ENTRY_ORIGIN))
1281 return;
1282 gmap_call_notifier(sg, raddr, raddr + _REGION2_SIZE - 1);
1283 r2o = (unsigned long) (r2e - ((raddr & _REGION2_INDEX) >> _REGION2_SHIFT));
1284 gmap_idte_one(r2o | _ASCE_TYPE_REGION2, raddr);
1285 r3t = (unsigned long *)(*r2e & _REGION_ENTRY_ORIGIN);
1286 *r2e = _REGION2_ENTRY_EMPTY;
1287 __gmap_unshadow_r3t(sg, raddr, r3t);
1288 /* Free region 3 table */
1289 page = pfn_to_page(__pa(r3t) >> PAGE_SHIFT);
1290 list_del(&page->lru);
1291 __free_pages(page, CRST_ALLOC_ORDER);
1292 }
1293
1294 /**
1295 * __gmap_unshadow_r2t - remove all entries from a shadow region-2 table
1296 * @sg: pointer to the shadow guest address space structure
1297 * @raddr: rmap address in the shadow guest address space
1298 * @r2t: pointer to the start of a shadow region-2 table
1299 *
1300 * Called with the sg->guest_table_lock
1301 */
1302 static void __gmap_unshadow_r2t(struct gmap *sg, unsigned long raddr,
1303 unsigned long *r2t)
1304 {
1305 unsigned long *r3t;
1306 struct page *page;
1307 int i;
1308
1309 BUG_ON(!gmap_is_shadow(sg));
1310 for (i = 0; i < _CRST_ENTRIES; i++, raddr += _REGION2_SIZE) {
1311 if (!(r2t[i] & _REGION_ENTRY_ORIGIN))
1312 continue;
1313 r3t = (unsigned long *)(r2t[i] & _REGION_ENTRY_ORIGIN);
1314 r2t[i] = _REGION2_ENTRY_EMPTY;
1315 __gmap_unshadow_r3t(sg, raddr, r3t);
1316 /* Free region 3 table */
1317 page = pfn_to_page(__pa(r3t) >> PAGE_SHIFT);
1318 list_del(&page->lru);
1319 __free_pages(page, CRST_ALLOC_ORDER);
1320 }
1321 }
1322
1323 /**
1324 * gmap_unshadow_r2t - remove a shadow region-2 table from a region-1 entry
1325 * @sg: pointer to the shadow guest address space structure
1326 * @raddr: rmap address in the shadow guest address space
1327 *
1328 * Called with the sg->guest_table_lock
1329 */
1330 static void gmap_unshadow_r2t(struct gmap *sg, unsigned long raddr)
1331 {
1332 unsigned long r1o, *r1e, *r2t;
1333 struct page *page;
1334
1335 BUG_ON(!gmap_is_shadow(sg));
1336 r1e = gmap_table_walk(sg, raddr, 4); /* get region-1 pointer */
1337 if (!r1e || !(*r1e & _REGION_ENTRY_ORIGIN))
1338 return;
1339 gmap_call_notifier(sg, raddr, raddr + _REGION1_SIZE - 1);
1340 r1o = (unsigned long) (r1e - ((raddr & _REGION1_INDEX) >> _REGION1_SHIFT));
1341 gmap_idte_one(r1o | _ASCE_TYPE_REGION1, raddr);
1342 r2t = (unsigned long *)(*r1e & _REGION_ENTRY_ORIGIN);
1343 *r1e = _REGION1_ENTRY_EMPTY;
1344 __gmap_unshadow_r2t(sg, raddr, r2t);
1345 /* Free region 2 table */
1346 page = pfn_to_page(__pa(r2t) >> PAGE_SHIFT);
1347 list_del(&page->lru);
1348 __free_pages(page, CRST_ALLOC_ORDER);
1349 }
1350
1351 /**
1352 * __gmap_unshadow_r1t - remove all entries from a shadow region-1 table
1353 * @sg: pointer to the shadow guest address space structure
1354 * @raddr: rmap address in the shadow guest address space
1355 * @r1t: pointer to the start of a shadow region-1 table
1356 *
1357 * Called with the shadow->guest_table_lock
1358 */
1359 static void __gmap_unshadow_r1t(struct gmap *sg, unsigned long raddr,
1360 unsigned long *r1t)
1361 {
1362 unsigned long asce, *r2t;
1363 struct page *page;
1364 int i;
1365
1366 BUG_ON(!gmap_is_shadow(sg));
1367 asce = (unsigned long) r1t | _ASCE_TYPE_REGION1;
1368 for (i = 0; i < _CRST_ENTRIES; i++, raddr += _REGION1_SIZE) {
1369 if (!(r1t[i] & _REGION_ENTRY_ORIGIN))
1370 continue;
1371 r2t = (unsigned long *)(r1t[i] & _REGION_ENTRY_ORIGIN);
1372 __gmap_unshadow_r2t(sg, raddr, r2t);
1373 /* Clear entry and flush translation r1t -> r2t */
1374 gmap_idte_one(asce, raddr);
1375 r1t[i] = _REGION1_ENTRY_EMPTY;
1376 /* Free region 2 table */
1377 page = pfn_to_page(__pa(r2t) >> PAGE_SHIFT);
1378 list_del(&page->lru);
1379 __free_pages(page, CRST_ALLOC_ORDER);
1380 }
1381 }
1382
1383 /**
1384 * gmap_unshadow - remove a shadow page table completely
1385 * @sg: pointer to the shadow guest address space structure
1386 *
1387 * Called with sg->guest_table_lock
1388 */
1389 static void gmap_unshadow(struct gmap *sg)
1390 {
1391 unsigned long *table;
1392
1393 BUG_ON(!gmap_is_shadow(sg));
1394 if (sg->removed)
1395 return;
1396 sg->removed = 1;
1397 gmap_call_notifier(sg, 0, -1UL);
1398 gmap_flush_tlb(sg);
1399 table = (unsigned long *)(sg->asce & _ASCE_ORIGIN);
1400 switch (sg->asce & _ASCE_TYPE_MASK) {
1401 case _ASCE_TYPE_REGION1:
1402 __gmap_unshadow_r1t(sg, 0, table);
1403 break;
1404 case _ASCE_TYPE_REGION2:
1405 __gmap_unshadow_r2t(sg, 0, table);
1406 break;
1407 case _ASCE_TYPE_REGION3:
1408 __gmap_unshadow_r3t(sg, 0, table);
1409 break;
1410 case _ASCE_TYPE_SEGMENT:
1411 __gmap_unshadow_sgt(sg, 0, table);
1412 break;
1413 }
1414 }
1415
1416 /**
1417 * gmap_find_shadow - find a specific asce in the list of shadow tables
1418 * @parent: pointer to the parent gmap
1419 * @asce: ASCE for which the shadow table is created
1420 * @edat_level: edat level to be used for the shadow translation
1421 *
1422 * Returns the pointer to a gmap if a shadow table with the given asce is
1423 * already available, ERR_PTR(-EAGAIN) if another one is just being created,
1424 * otherwise NULL
1425 */
1426 static struct gmap *gmap_find_shadow(struct gmap *parent, unsigned long asce,
1427 int edat_level)
1428 {
1429 struct gmap *sg;
1430
1431 list_for_each_entry(sg, &parent->children, list) {
1432 if (sg->orig_asce != asce || sg->edat_level != edat_level ||
1433 sg->removed)
1434 continue;
1435 if (!sg->initialized)
1436 return ERR_PTR(-EAGAIN);
1437 atomic_inc(&sg->ref_count);
1438 return sg;
1439 }
1440 return NULL;
1441 }
1442
1443 /**
1444 * gmap_shadow_valid - check if a shadow guest address space matches the
1445 * given properties and is still valid
1446 * @sg: pointer to the shadow guest address space structure
1447 * @asce: ASCE for which the shadow table is requested
1448 * @edat_level: edat level to be used for the shadow translation
1449 *
1450 * Returns 1 if the gmap shadow is still valid and matches the given
1451 * properties, the caller can continue using it. Returns 0 otherwise, the
1452 * caller has to request a new shadow gmap in this case.
1453 *
1454 */
1455 int gmap_shadow_valid(struct gmap *sg, unsigned long asce, int edat_level)
1456 {
1457 if (sg->removed)
1458 return 0;
1459 return sg->orig_asce == asce && sg->edat_level == edat_level;
1460 }
1461 EXPORT_SYMBOL_GPL(gmap_shadow_valid);
1462
1463 /**
1464 * gmap_shadow - create/find a shadow guest address space
1465 * @parent: pointer to the parent gmap
1466 * @asce: ASCE for which the shadow table is created
1467 * @edat_level: edat level to be used for the shadow translation
1468 *
1469 * The pages of the top level page table referred by the asce parameter
1470 * will be set to read-only and marked in the PGSTEs of the kvm process.
1471 * The shadow table will be removed automatically on any change to the
1472 * PTE mapping for the source table.
1473 *
1474 * Returns a guest address space structure, ERR_PTR(-ENOMEM) if out of memory,
1475 * ERR_PTR(-EAGAIN) if the caller has to retry and ERR_PTR(-EFAULT) if the
1476 * parent gmap table could not be protected.
1477 */
1478 struct gmap *gmap_shadow(struct gmap *parent, unsigned long asce,
1479 int edat_level)
1480 {
1481 struct gmap *sg, *new;
1482 unsigned long limit;
1483 int rc;
1484
1485 BUG_ON(gmap_is_shadow(parent));
1486 spin_lock(&parent->shadow_lock);
1487 sg = gmap_find_shadow(parent, asce, edat_level);
1488 spin_unlock(&parent->shadow_lock);
1489 if (sg)
1490 return sg;
1491 /* Create a new shadow gmap */
1492 limit = -1UL >> (33 - (((asce & _ASCE_TYPE_MASK) >> 2) * 11));
1493 if (asce & _ASCE_REAL_SPACE)
1494 limit = -1UL;
1495 new = gmap_alloc(limit);
1496 if (!new)
1497 return ERR_PTR(-ENOMEM);
1498 new->mm = parent->mm;
1499 new->parent = gmap_get(parent);
1500 new->orig_asce = asce;
1501 new->edat_level = edat_level;
1502 new->initialized = false;
1503 spin_lock(&parent->shadow_lock);
1504 /* Recheck if another CPU created the same shadow */
1505 sg = gmap_find_shadow(parent, asce, edat_level);
1506 if (sg) {
1507 spin_unlock(&parent->shadow_lock);
1508 gmap_free(new);
1509 return sg;
1510 }
1511 if (asce & _ASCE_REAL_SPACE) {
1512 /* only allow one real-space gmap shadow */
1513 list_for_each_entry(sg, &parent->children, list) {
1514 if (sg->orig_asce & _ASCE_REAL_SPACE) {
1515 spin_lock(&sg->guest_table_lock);
1516 gmap_unshadow(sg);
1517 spin_unlock(&sg->guest_table_lock);
1518 list_del(&sg->list);
1519 gmap_put(sg);
1520 break;
1521 }
1522 }
1523 }
1524 atomic_set(&new->ref_count, 2);
1525 list_add(&new->list, &parent->children);
1526 if (asce & _ASCE_REAL_SPACE) {
1527 /* nothing to protect, return right away */
1528 new->initialized = true;
1529 spin_unlock(&parent->shadow_lock);
1530 return new;
1531 }
1532 spin_unlock(&parent->shadow_lock);
1533 /* protect after insertion, so it will get properly invalidated */
1534 down_read(&parent->mm->mmap_sem);
1535 rc = gmap_protect_range(parent, asce & _ASCE_ORIGIN,
1536 ((asce & _ASCE_TABLE_LENGTH) + 1) * PAGE_SIZE,
1537 PROT_READ, PGSTE_VSIE_BIT);
1538 up_read(&parent->mm->mmap_sem);
1539 spin_lock(&parent->shadow_lock);
1540 new->initialized = true;
1541 if (rc) {
1542 list_del(&new->list);
1543 gmap_free(new);
1544 new = ERR_PTR(rc);
1545 }
1546 spin_unlock(&parent->shadow_lock);
1547 return new;
1548 }
1549 EXPORT_SYMBOL_GPL(gmap_shadow);
1550
1551 /**
1552 * gmap_shadow_r2t - create an empty shadow region 2 table
1553 * @sg: pointer to the shadow guest address space structure
1554 * @saddr: faulting address in the shadow gmap
1555 * @r2t: parent gmap address of the region 2 table to get shadowed
1556 * @fake: r2t references contiguous guest memory block, not a r2t
1557 *
1558 * The r2t parameter specifies the address of the source table. The
1559 * four pages of the source table are made read-only in the parent gmap
1560 * address space. A write to the source table area @r2t will automatically
1561 * remove the shadow r2 table and all of its decendents.
1562 *
1563 * Returns 0 if successfully shadowed or already shadowed, -EAGAIN if the
1564 * shadow table structure is incomplete, -ENOMEM if out of memory and
1565 * -EFAULT if an address in the parent gmap could not be resolved.
1566 *
1567 * Called with sg->mm->mmap_sem in read.
1568 */
1569 int gmap_shadow_r2t(struct gmap *sg, unsigned long saddr, unsigned long r2t,
1570 int fake)
1571 {
1572 unsigned long raddr, origin, offset, len;
1573 unsigned long *s_r2t, *table;
1574 struct page *page;
1575 int rc;
1576
1577 BUG_ON(!gmap_is_shadow(sg));
1578 /* Allocate a shadow region second table */
1579 page = alloc_pages(GFP_KERNEL, CRST_ALLOC_ORDER);
1580 if (!page)
1581 return -ENOMEM;
1582 page->index = r2t & _REGION_ENTRY_ORIGIN;
1583 if (fake)
1584 page->index |= GMAP_SHADOW_FAKE_TABLE;
1585 s_r2t = (unsigned long *) page_to_phys(page);
1586 /* Install shadow region second table */
1587 spin_lock(&sg->guest_table_lock);
1588 table = gmap_table_walk(sg, saddr, 4); /* get region-1 pointer */
1589 if (!table) {
1590 rc = -EAGAIN; /* Race with unshadow */
1591 goto out_free;
1592 }
1593 if (!(*table & _REGION_ENTRY_INVALID)) {
1594 rc = 0; /* Already established */
1595 goto out_free;
1596 } else if (*table & _REGION_ENTRY_ORIGIN) {
1597 rc = -EAGAIN; /* Race with shadow */
1598 goto out_free;
1599 }
1600 crst_table_init(s_r2t, _REGION2_ENTRY_EMPTY);
1601 /* mark as invalid as long as the parent table is not protected */
1602 *table = (unsigned long) s_r2t | _REGION_ENTRY_LENGTH |
1603 _REGION_ENTRY_TYPE_R1 | _REGION_ENTRY_INVALID;
1604 if (sg->edat_level >= 1)
1605 *table |= (r2t & _REGION_ENTRY_PROTECT);
1606 list_add(&page->lru, &sg->crst_list);
1607 if (fake) {
1608 /* nothing to protect for fake tables */
1609 *table &= ~_REGION_ENTRY_INVALID;
1610 spin_unlock(&sg->guest_table_lock);
1611 return 0;
1612 }
1613 spin_unlock(&sg->guest_table_lock);
1614 /* Make r2t read-only in parent gmap page table */
1615 raddr = (saddr & _REGION1_MASK) | _SHADOW_RMAP_REGION1;
1616 origin = r2t & _REGION_ENTRY_ORIGIN;
1617 offset = ((r2t & _REGION_ENTRY_OFFSET) >> 6) * PAGE_SIZE;
1618 len = ((r2t & _REGION_ENTRY_LENGTH) + 1) * PAGE_SIZE - offset;
1619 rc = gmap_protect_rmap(sg, raddr, origin + offset, len, PROT_READ);
1620 spin_lock(&sg->guest_table_lock);
1621 if (!rc) {
1622 table = gmap_table_walk(sg, saddr, 4);
1623 if (!table || (*table & _REGION_ENTRY_ORIGIN) !=
1624 (unsigned long) s_r2t)
1625 rc = -EAGAIN; /* Race with unshadow */
1626 else
1627 *table &= ~_REGION_ENTRY_INVALID;
1628 } else {
1629 gmap_unshadow_r2t(sg, raddr);
1630 }
1631 spin_unlock(&sg->guest_table_lock);
1632 return rc;
1633 out_free:
1634 spin_unlock(&sg->guest_table_lock);
1635 __free_pages(page, CRST_ALLOC_ORDER);
1636 return rc;
1637 }
1638 EXPORT_SYMBOL_GPL(gmap_shadow_r2t);
1639
1640 /**
1641 * gmap_shadow_r3t - create a shadow region 3 table
1642 * @sg: pointer to the shadow guest address space structure
1643 * @saddr: faulting address in the shadow gmap
1644 * @r3t: parent gmap address of the region 3 table to get shadowed
1645 * @fake: r3t references contiguous guest memory block, not a r3t
1646 *
1647 * Returns 0 if successfully shadowed or already shadowed, -EAGAIN if the
1648 * shadow table structure is incomplete, -ENOMEM if out of memory and
1649 * -EFAULT if an address in the parent gmap could not be resolved.
1650 *
1651 * Called with sg->mm->mmap_sem in read.
1652 */
1653 int gmap_shadow_r3t(struct gmap *sg, unsigned long saddr, unsigned long r3t,
1654 int fake)
1655 {
1656 unsigned long raddr, origin, offset, len;
1657 unsigned long *s_r3t, *table;
1658 struct page *page;
1659 int rc;
1660
1661 BUG_ON(!gmap_is_shadow(sg));
1662 /* Allocate a shadow region second table */
1663 page = alloc_pages(GFP_KERNEL, CRST_ALLOC_ORDER);
1664 if (!page)
1665 return -ENOMEM;
1666 page->index = r3t & _REGION_ENTRY_ORIGIN;
1667 if (fake)
1668 page->index |= GMAP_SHADOW_FAKE_TABLE;
1669 s_r3t = (unsigned long *) page_to_phys(page);
1670 /* Install shadow region second table */
1671 spin_lock(&sg->guest_table_lock);
1672 table = gmap_table_walk(sg, saddr, 3); /* get region-2 pointer */
1673 if (!table) {
1674 rc = -EAGAIN; /* Race with unshadow */
1675 goto out_free;
1676 }
1677 if (!(*table & _REGION_ENTRY_INVALID)) {
1678 rc = 0; /* Already established */
1679 goto out_free;
1680 } else if (*table & _REGION_ENTRY_ORIGIN) {
1681 rc = -EAGAIN; /* Race with shadow */
1682 }
1683 crst_table_init(s_r3t, _REGION3_ENTRY_EMPTY);
1684 /* mark as invalid as long as the parent table is not protected */
1685 *table = (unsigned long) s_r3t | _REGION_ENTRY_LENGTH |
1686 _REGION_ENTRY_TYPE_R2 | _REGION_ENTRY_INVALID;
1687 if (sg->edat_level >= 1)
1688 *table |= (r3t & _REGION_ENTRY_PROTECT);
1689 list_add(&page->lru, &sg->crst_list);
1690 if (fake) {
1691 /* nothing to protect for fake tables */
1692 *table &= ~_REGION_ENTRY_INVALID;
1693 spin_unlock(&sg->guest_table_lock);
1694 return 0;
1695 }
1696 spin_unlock(&sg->guest_table_lock);
1697 /* Make r3t read-only in parent gmap page table */
1698 raddr = (saddr & _REGION2_MASK) | _SHADOW_RMAP_REGION2;
1699 origin = r3t & _REGION_ENTRY_ORIGIN;
1700 offset = ((r3t & _REGION_ENTRY_OFFSET) >> 6) * PAGE_SIZE;
1701 len = ((r3t & _REGION_ENTRY_LENGTH) + 1) * PAGE_SIZE - offset;
1702 rc = gmap_protect_rmap(sg, raddr, origin + offset, len, PROT_READ);
1703 spin_lock(&sg->guest_table_lock);
1704 if (!rc) {
1705 table = gmap_table_walk(sg, saddr, 3);
1706 if (!table || (*table & _REGION_ENTRY_ORIGIN) !=
1707 (unsigned long) s_r3t)
1708 rc = -EAGAIN; /* Race with unshadow */
1709 else
1710 *table &= ~_REGION_ENTRY_INVALID;
1711 } else {
1712 gmap_unshadow_r3t(sg, raddr);
1713 }
1714 spin_unlock(&sg->guest_table_lock);
1715 return rc;
1716 out_free:
1717 spin_unlock(&sg->guest_table_lock);
1718 __free_pages(page, CRST_ALLOC_ORDER);
1719 return rc;
1720 }
1721 EXPORT_SYMBOL_GPL(gmap_shadow_r3t);
1722
1723 /**
1724 * gmap_shadow_sgt - create a shadow segment table
1725 * @sg: pointer to the shadow guest address space structure
1726 * @saddr: faulting address in the shadow gmap
1727 * @sgt: parent gmap address of the segment table to get shadowed
1728 * @fake: sgt references contiguous guest memory block, not a sgt
1729 *
1730 * Returns: 0 if successfully shadowed or already shadowed, -EAGAIN if the
1731 * shadow table structure is incomplete, -ENOMEM if out of memory and
1732 * -EFAULT if an address in the parent gmap could not be resolved.
1733 *
1734 * Called with sg->mm->mmap_sem in read.
1735 */
1736 int gmap_shadow_sgt(struct gmap *sg, unsigned long saddr, unsigned long sgt,
1737 int fake)
1738 {
1739 unsigned long raddr, origin, offset, len;
1740 unsigned long *s_sgt, *table;
1741 struct page *page;
1742 int rc;
1743
1744 BUG_ON(!gmap_is_shadow(sg) || (sgt & _REGION3_ENTRY_LARGE));
1745 /* Allocate a shadow segment table */
1746 page = alloc_pages(GFP_KERNEL, CRST_ALLOC_ORDER);
1747 if (!page)
1748 return -ENOMEM;
1749 page->index = sgt & _REGION_ENTRY_ORIGIN;
1750 if (fake)
1751 page->index |= GMAP_SHADOW_FAKE_TABLE;
1752 s_sgt = (unsigned long *) page_to_phys(page);
1753 /* Install shadow region second table */
1754 spin_lock(&sg->guest_table_lock);
1755 table = gmap_table_walk(sg, saddr, 2); /* get region-3 pointer */
1756 if (!table) {
1757 rc = -EAGAIN; /* Race with unshadow */
1758 goto out_free;
1759 }
1760 if (!(*table & _REGION_ENTRY_INVALID)) {
1761 rc = 0; /* Already established */
1762 goto out_free;
1763 } else if (*table & _REGION_ENTRY_ORIGIN) {
1764 rc = -EAGAIN; /* Race with shadow */
1765 goto out_free;
1766 }
1767 crst_table_init(s_sgt, _SEGMENT_ENTRY_EMPTY);
1768 /* mark as invalid as long as the parent table is not protected */
1769 *table = (unsigned long) s_sgt | _REGION_ENTRY_LENGTH |
1770 _REGION_ENTRY_TYPE_R3 | _REGION_ENTRY_INVALID;
1771 if (sg->edat_level >= 1)
1772 *table |= sgt & _REGION_ENTRY_PROTECT;
1773 list_add(&page->lru, &sg->crst_list);
1774 if (fake) {
1775 /* nothing to protect for fake tables */
1776 *table &= ~_REGION_ENTRY_INVALID;
1777 spin_unlock(&sg->guest_table_lock);
1778 return 0;
1779 }
1780 spin_unlock(&sg->guest_table_lock);
1781 /* Make sgt read-only in parent gmap page table */
1782 raddr = (saddr & _REGION3_MASK) | _SHADOW_RMAP_REGION3;
1783 origin = sgt & _REGION_ENTRY_ORIGIN;
1784 offset = ((sgt & _REGION_ENTRY_OFFSET) >> 6) * PAGE_SIZE;
1785 len = ((sgt & _REGION_ENTRY_LENGTH) + 1) * PAGE_SIZE - offset;
1786 rc = gmap_protect_rmap(sg, raddr, origin + offset, len, PROT_READ);
1787 spin_lock(&sg->guest_table_lock);
1788 if (!rc) {
1789 table = gmap_table_walk(sg, saddr, 2);
1790 if (!table || (*table & _REGION_ENTRY_ORIGIN) !=
1791 (unsigned long) s_sgt)
1792 rc = -EAGAIN; /* Race with unshadow */
1793 else
1794 *table &= ~_REGION_ENTRY_INVALID;
1795 } else {
1796 gmap_unshadow_sgt(sg, raddr);
1797 }
1798 spin_unlock(&sg->guest_table_lock);
1799 return rc;
1800 out_free:
1801 spin_unlock(&sg->guest_table_lock);
1802 __free_pages(page, CRST_ALLOC_ORDER);
1803 return rc;
1804 }
1805 EXPORT_SYMBOL_GPL(gmap_shadow_sgt);
1806
1807 /**
1808 * gmap_shadow_lookup_pgtable - find a shadow page table
1809 * @sg: pointer to the shadow guest address space structure
1810 * @saddr: the address in the shadow aguest address space
1811 * @pgt: parent gmap address of the page table to get shadowed
1812 * @dat_protection: if the pgtable is marked as protected by dat
1813 * @fake: pgt references contiguous guest memory block, not a pgtable
1814 *
1815 * Returns 0 if the shadow page table was found and -EAGAIN if the page
1816 * table was not found.
1817 *
1818 * Called with sg->mm->mmap_sem in read.
1819 */
1820 int gmap_shadow_pgt_lookup(struct gmap *sg, unsigned long saddr,
1821 unsigned long *pgt, int *dat_protection,
1822 int *fake)
1823 {
1824 unsigned long *table;
1825 struct page *page;
1826 int rc;
1827
1828 BUG_ON(!gmap_is_shadow(sg));
1829 spin_lock(&sg->guest_table_lock);
1830 table = gmap_table_walk(sg, saddr, 1); /* get segment pointer */
1831 if (table && !(*table & _SEGMENT_ENTRY_INVALID)) {
1832 /* Shadow page tables are full pages (pte+pgste) */
1833 page = pfn_to_page(*table >> PAGE_SHIFT);
1834 *pgt = page->index & ~GMAP_SHADOW_FAKE_TABLE;
1835 *dat_protection = !!(*table & _SEGMENT_ENTRY_PROTECT);
1836 *fake = !!(page->index & GMAP_SHADOW_FAKE_TABLE);
1837 rc = 0;
1838 } else {
1839 rc = -EAGAIN;
1840 }
1841 spin_unlock(&sg->guest_table_lock);
1842 return rc;
1843
1844 }
1845 EXPORT_SYMBOL_GPL(gmap_shadow_pgt_lookup);
1846
1847 /**
1848 * gmap_shadow_pgt - instantiate a shadow page table
1849 * @sg: pointer to the shadow guest address space structure
1850 * @saddr: faulting address in the shadow gmap
1851 * @pgt: parent gmap address of the page table to get shadowed
1852 * @fake: pgt references contiguous guest memory block, not a pgtable
1853 *
1854 * Returns 0 if successfully shadowed or already shadowed, -EAGAIN if the
1855 * shadow table structure is incomplete, -ENOMEM if out of memory,
1856 * -EFAULT if an address in the parent gmap could not be resolved and
1857 *
1858 * Called with gmap->mm->mmap_sem in read
1859 */
1860 int gmap_shadow_pgt(struct gmap *sg, unsigned long saddr, unsigned long pgt,
1861 int fake)
1862 {
1863 unsigned long raddr, origin;
1864 unsigned long *s_pgt, *table;
1865 struct page *page;
1866 int rc;
1867
1868 BUG_ON(!gmap_is_shadow(sg) || (pgt & _SEGMENT_ENTRY_LARGE));
1869 /* Allocate a shadow page table */
1870 page = page_table_alloc_pgste(sg->mm);
1871 if (!page)
1872 return -ENOMEM;
1873 page->index = pgt & _SEGMENT_ENTRY_ORIGIN;
1874 if (fake)
1875 page->index |= GMAP_SHADOW_FAKE_TABLE;
1876 s_pgt = (unsigned long *) page_to_phys(page);
1877 /* Install shadow page table */
1878 spin_lock(&sg->guest_table_lock);
1879 table = gmap_table_walk(sg, saddr, 1); /* get segment pointer */
1880 if (!table) {
1881 rc = -EAGAIN; /* Race with unshadow */
1882 goto out_free;
1883 }
1884 if (!(*table & _SEGMENT_ENTRY_INVALID)) {
1885 rc = 0; /* Already established */
1886 goto out_free;
1887 } else if (*table & _SEGMENT_ENTRY_ORIGIN) {
1888 rc = -EAGAIN; /* Race with shadow */
1889 goto out_free;
1890 }
1891 /* mark as invalid as long as the parent table is not protected */
1892 *table = (unsigned long) s_pgt | _SEGMENT_ENTRY |
1893 (pgt & _SEGMENT_ENTRY_PROTECT) | _SEGMENT_ENTRY_INVALID;
1894 list_add(&page->lru, &sg->pt_list);
1895 if (fake) {
1896 /* nothing to protect for fake tables */
1897 *table &= ~_SEGMENT_ENTRY_INVALID;
1898 spin_unlock(&sg->guest_table_lock);
1899 return 0;
1900 }
1901 spin_unlock(&sg->guest_table_lock);
1902 /* Make pgt read-only in parent gmap page table (not the pgste) */
1903 raddr = (saddr & _SEGMENT_MASK) | _SHADOW_RMAP_SEGMENT;
1904 origin = pgt & _SEGMENT_ENTRY_ORIGIN & PAGE_MASK;
1905 rc = gmap_protect_rmap(sg, raddr, origin, PAGE_SIZE, PROT_READ);
1906 spin_lock(&sg->guest_table_lock);
1907 if (!rc) {
1908 table = gmap_table_walk(sg, saddr, 1);
1909 if (!table || (*table & _SEGMENT_ENTRY_ORIGIN) !=
1910 (unsigned long) s_pgt)
1911 rc = -EAGAIN; /* Race with unshadow */
1912 else
1913 *table &= ~_SEGMENT_ENTRY_INVALID;
1914 } else {
1915 gmap_unshadow_pgt(sg, raddr);
1916 }
1917 spin_unlock(&sg->guest_table_lock);
1918 return rc;
1919 out_free:
1920 spin_unlock(&sg->guest_table_lock);
1921 page_table_free_pgste(page);
1922 return rc;
1923
1924 }
1925 EXPORT_SYMBOL_GPL(gmap_shadow_pgt);
1926
1927 /**
1928 * gmap_shadow_page - create a shadow page mapping
1929 * @sg: pointer to the shadow guest address space structure
1930 * @saddr: faulting address in the shadow gmap
1931 * @pte: pte in parent gmap address space to get shadowed
1932 *
1933 * Returns 0 if successfully shadowed or already shadowed, -EAGAIN if the
1934 * shadow table structure is incomplete, -ENOMEM if out of memory and
1935 * -EFAULT if an address in the parent gmap could not be resolved.
1936 *
1937 * Called with sg->mm->mmap_sem in read.
1938 */
1939 int gmap_shadow_page(struct gmap *sg, unsigned long saddr, pte_t pte)
1940 {
1941 struct gmap *parent;
1942 struct gmap_rmap *rmap;
1943 unsigned long vmaddr, paddr;
1944 spinlock_t *ptl;
1945 pte_t *sptep, *tptep;
1946 int prot;
1947 int rc;
1948
1949 BUG_ON(!gmap_is_shadow(sg));
1950 parent = sg->parent;
1951 prot = (pte_val(pte) & _PAGE_PROTECT) ? PROT_READ : PROT_WRITE;
1952
1953 rmap = kzalloc(sizeof(*rmap), GFP_KERNEL);
1954 if (!rmap)
1955 return -ENOMEM;
1956 rmap->raddr = (saddr & PAGE_MASK) | _SHADOW_RMAP_PGTABLE;
1957
1958 while (1) {
1959 paddr = pte_val(pte) & PAGE_MASK;
1960 vmaddr = __gmap_translate(parent, paddr);
1961 if (IS_ERR_VALUE(vmaddr)) {
1962 rc = vmaddr;
1963 break;
1964 }
1965 rc = radix_tree_preload(GFP_KERNEL);
1966 if (rc)
1967 break;
1968 rc = -EAGAIN;
1969 sptep = gmap_pte_op_walk(parent, paddr, &ptl);
1970 if (sptep) {
1971 spin_lock(&sg->guest_table_lock);
1972 /* Get page table pointer */
1973 tptep = (pte_t *) gmap_table_walk(sg, saddr, 0);
1974 if (!tptep) {
1975 spin_unlock(&sg->guest_table_lock);
1976 gmap_pte_op_end(ptl);
1977 radix_tree_preload_end();
1978 break;
1979 }
1980 rc = ptep_shadow_pte(sg->mm, saddr, sptep, tptep, pte);
1981 if (rc > 0) {
1982 /* Success and a new mapping */
1983 gmap_insert_rmap(sg, vmaddr, rmap);
1984 rmap = NULL;
1985 rc = 0;
1986 }
1987 gmap_pte_op_end(ptl);
1988 spin_unlock(&sg->guest_table_lock);
1989 }
1990 radix_tree_preload_end();
1991 if (!rc)
1992 break;
1993 rc = gmap_pte_op_fixup(parent, paddr, vmaddr, prot);
1994 if (rc)
1995 break;
1996 }
1997 kfree(rmap);
1998 return rc;
1999 }
2000 EXPORT_SYMBOL_GPL(gmap_shadow_page);
2001
2002 /**
2003 * gmap_shadow_notify - handle notifications for shadow gmap
2004 *
2005 * Called with sg->parent->shadow_lock.
2006 */
2007 static void gmap_shadow_notify(struct gmap *sg, unsigned long vmaddr,
2008 unsigned long gaddr, pte_t *pte)
2009 {
2010 struct gmap_rmap *rmap, *rnext, *head;
2011 unsigned long start, end, bits, raddr;
2012
2013 BUG_ON(!gmap_is_shadow(sg));
2014
2015 spin_lock(&sg->guest_table_lock);
2016 if (sg->removed) {
2017 spin_unlock(&sg->guest_table_lock);
2018 return;
2019 }
2020 /* Check for top level table */
2021 start = sg->orig_asce & _ASCE_ORIGIN;
2022 end = start + ((sg->orig_asce & _ASCE_TABLE_LENGTH) + 1) * PAGE_SIZE;
2023 if (!(sg->orig_asce & _ASCE_REAL_SPACE) && gaddr >= start &&
2024 gaddr < end) {
2025 /* The complete shadow table has to go */
2026 gmap_unshadow(sg);
2027 spin_unlock(&sg->guest_table_lock);
2028 list_del(&sg->list);
2029 gmap_put(sg);
2030 return;
2031 }
2032 /* Remove the page table tree from on specific entry */
2033 head = radix_tree_delete(&sg->host_to_rmap, vmaddr >> PAGE_SHIFT);
2034 gmap_for_each_rmap_safe(rmap, rnext, head) {
2035 bits = rmap->raddr & _SHADOW_RMAP_MASK;
2036 raddr = rmap->raddr ^ bits;
2037 switch (bits) {
2038 case _SHADOW_RMAP_REGION1:
2039 gmap_unshadow_r2t(sg, raddr);
2040 break;
2041 case _SHADOW_RMAP_REGION2:
2042 gmap_unshadow_r3t(sg, raddr);
2043 break;
2044 case _SHADOW_RMAP_REGION3:
2045 gmap_unshadow_sgt(sg, raddr);
2046 break;
2047 case _SHADOW_RMAP_SEGMENT:
2048 gmap_unshadow_pgt(sg, raddr);
2049 break;
2050 case _SHADOW_RMAP_PGTABLE:
2051 gmap_unshadow_page(sg, raddr);
2052 break;
2053 }
2054 kfree(rmap);
2055 }
2056 spin_unlock(&sg->guest_table_lock);
2057 }
2058
2059 /**
2060 * ptep_notify - call all invalidation callbacks for a specific pte.
2061 * @mm: pointer to the process mm_struct
2062 * @addr: virtual address in the process address space
2063 * @pte: pointer to the page table entry
2064 * @bits: bits from the pgste that caused the notify call
2065 *
2066 * This function is assumed to be called with the page table lock held
2067 * for the pte to notify.
2068 */
2069 void ptep_notify(struct mm_struct *mm, unsigned long vmaddr,
2070 pte_t *pte, unsigned long bits)
2071 {
2072 unsigned long offset, gaddr = 0;
2073 unsigned long *table;
2074 struct gmap *gmap, *sg, *next;
2075
2076 offset = ((unsigned long) pte) & (255 * sizeof(pte_t));
2077 offset = offset * (PAGE_SIZE / sizeof(pte_t));
2078 rcu_read_lock();
2079 list_for_each_entry_rcu(gmap, &mm->context.gmap_list, list) {
2080 spin_lock(&gmap->guest_table_lock);
2081 table = radix_tree_lookup(&gmap->host_to_guest,
2082 vmaddr >> PMD_SHIFT);
2083 if (table)
2084 gaddr = __gmap_segment_gaddr(table) + offset;
2085 spin_unlock(&gmap->guest_table_lock);
2086 if (!table)
2087 continue;
2088
2089 if (!list_empty(&gmap->children) && (bits & PGSTE_VSIE_BIT)) {
2090 spin_lock(&gmap->shadow_lock);
2091 list_for_each_entry_safe(sg, next,
2092 &gmap->children, list)
2093 gmap_shadow_notify(sg, vmaddr, gaddr, pte);
2094 spin_unlock(&gmap->shadow_lock);
2095 }
2096 if (bits & PGSTE_IN_BIT)
2097 gmap_call_notifier(gmap, gaddr, gaddr + PAGE_SIZE - 1);
2098 }
2099 rcu_read_unlock();
2100 }
2101 EXPORT_SYMBOL_GPL(ptep_notify);
2102
2103 static inline void thp_split_mm(struct mm_struct *mm)
2104 {
2105 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
2106 struct vm_area_struct *vma;
2107 unsigned long addr;
2108
2109 for (vma = mm->mmap; vma != NULL; vma = vma->vm_next) {
2110 for (addr = vma->vm_start;
2111 addr < vma->vm_end;
2112 addr += PAGE_SIZE)
2113 follow_page(vma, addr, FOLL_SPLIT);
2114 vma->vm_flags &= ~VM_HUGEPAGE;
2115 vma->vm_flags |= VM_NOHUGEPAGE;
2116 }
2117 mm->def_flags |= VM_NOHUGEPAGE;
2118 #endif
2119 }
2120
2121 /*
2122 * Remove all empty zero pages from the mapping for lazy refaulting
2123 * - This must be called after mm->context.has_pgste is set, to avoid
2124 * future creation of zero pages
2125 * - This must be called after THP was enabled
2126 */
2127 static int __zap_zero_pages(pmd_t *pmd, unsigned long start,
2128 unsigned long end, struct mm_walk *walk)
2129 {
2130 unsigned long addr;
2131
2132 for (addr = start; addr != end; addr += PAGE_SIZE) {
2133 pte_t *ptep;
2134 spinlock_t *ptl;
2135
2136 ptep = pte_offset_map_lock(walk->mm, pmd, addr, &ptl);
2137 if (is_zero_pfn(pte_pfn(*ptep)))
2138 ptep_xchg_direct(walk->mm, addr, ptep, __pte(_PAGE_INVALID));
2139 pte_unmap_unlock(ptep, ptl);
2140 }
2141 return 0;
2142 }
2143
2144 static inline void zap_zero_pages(struct mm_struct *mm)
2145 {
2146 struct mm_walk walk = { .pmd_entry = __zap_zero_pages };
2147
2148 walk.mm = mm;
2149 walk_page_range(0, TASK_SIZE, &walk);
2150 }
2151
2152 /*
2153 * switch on pgstes for its userspace process (for kvm)
2154 */
2155 int s390_enable_sie(void)
2156 {
2157 struct mm_struct *mm = current->mm;
2158
2159 /* Do we have pgstes? if yes, we are done */
2160 if (mm_has_pgste(mm))
2161 return 0;
2162 /* Fail if the page tables are 2K */
2163 if (!mm_alloc_pgste(mm))
2164 return -EINVAL;
2165 down_write(&mm->mmap_sem);
2166 mm->context.has_pgste = 1;
2167 /* split thp mappings and disable thp for future mappings */
2168 thp_split_mm(mm);
2169 zap_zero_pages(mm);
2170 up_write(&mm->mmap_sem);
2171 return 0;
2172 }
2173 EXPORT_SYMBOL_GPL(s390_enable_sie);
2174
2175 /*
2176 * Enable storage key handling from now on and initialize the storage
2177 * keys with the default key.
2178 */
2179 static int __s390_enable_skey(pte_t *pte, unsigned long addr,
2180 unsigned long next, struct mm_walk *walk)
2181 {
2182 /* Clear storage key */
2183 ptep_zap_key(walk->mm, addr, pte);
2184 return 0;
2185 }
2186
2187 int s390_enable_skey(void)
2188 {
2189 struct mm_walk walk = { .pte_entry = __s390_enable_skey };
2190 struct mm_struct *mm = current->mm;
2191 struct vm_area_struct *vma;
2192 int rc = 0;
2193
2194 down_write(&mm->mmap_sem);
2195 if (mm_use_skey(mm))
2196 goto out_up;
2197
2198 mm->context.use_skey = 1;
2199 for (vma = mm->mmap; vma; vma = vma->vm_next) {
2200 if (ksm_madvise(vma, vma->vm_start, vma->vm_end,
2201 MADV_UNMERGEABLE, &vma->vm_flags)) {
2202 mm->context.use_skey = 0;
2203 rc = -ENOMEM;
2204 goto out_up;
2205 }
2206 }
2207 mm->def_flags &= ~VM_MERGEABLE;
2208
2209 walk.mm = mm;
2210 walk_page_range(0, TASK_SIZE, &walk);
2211
2212 out_up:
2213 up_write(&mm->mmap_sem);
2214 return rc;
2215 }
2216 EXPORT_SYMBOL_GPL(s390_enable_skey);
2217
2218 /*
2219 * Reset CMMA state, make all pages stable again.
2220 */
2221 static int __s390_reset_cmma(pte_t *pte, unsigned long addr,
2222 unsigned long next, struct mm_walk *walk)
2223 {
2224 ptep_zap_unused(walk->mm, addr, pte, 1);
2225 return 0;
2226 }
2227
2228 void s390_reset_cmma(struct mm_struct *mm)
2229 {
2230 struct mm_walk walk = { .pte_entry = __s390_reset_cmma };
2231
2232 down_write(&mm->mmap_sem);
2233 walk.mm = mm;
2234 walk_page_range(0, TASK_SIZE, &walk);
2235 up_write(&mm->mmap_sem);
2236 }
2237 EXPORT_SYMBOL_GPL(s390_reset_cmma);