x86/speculation/mds: Fix documentation typo
[linux/fpc-iii.git] / arch / s390 / mm / gmap.c
blobec9292917d3f24852ec931c5abb82eecc2e516a2
1 /*
2 * KVM guest address space mapping code
4 * Copyright IBM Corp. 2007, 2016
5 * Author(s): Martin Schwidefsky <schwidefsky@de.ibm.com>
6 */
8 #include <linux/kernel.h>
9 #include <linux/mm.h>
10 #include <linux/swap.h>
11 #include <linux/smp.h>
12 #include <linux/spinlock.h>
13 #include <linux/slab.h>
14 #include <linux/swapops.h>
15 #include <linux/ksm.h>
16 #include <linux/mman.h>
18 #include <asm/pgtable.h>
19 #include <asm/pgalloc.h>
20 #include <asm/gmap.h>
21 #include <asm/tlb.h>
23 #define GMAP_SHADOW_FAKE_TABLE 1ULL
25 /**
26 * gmap_alloc - allocate and initialize a guest address space
27 * @mm: pointer to the parent mm_struct
28 * @limit: maximum address of the gmap address space
30 * Returns a guest address space structure.
32 static struct gmap *gmap_alloc(unsigned long limit)
34 struct gmap *gmap;
35 struct page *page;
36 unsigned long *table;
37 unsigned long etype, atype;
39 if (limit < _REGION3_SIZE) {
40 limit = _REGION3_SIZE - 1;
41 atype = _ASCE_TYPE_SEGMENT;
42 etype = _SEGMENT_ENTRY_EMPTY;
43 } else if (limit < _REGION2_SIZE) {
44 limit = _REGION2_SIZE - 1;
45 atype = _ASCE_TYPE_REGION3;
46 etype = _REGION3_ENTRY_EMPTY;
47 } else if (limit < _REGION1_SIZE) {
48 limit = _REGION1_SIZE - 1;
49 atype = _ASCE_TYPE_REGION2;
50 etype = _REGION2_ENTRY_EMPTY;
51 } else {
52 limit = -1UL;
53 atype = _ASCE_TYPE_REGION1;
54 etype = _REGION1_ENTRY_EMPTY;
56 gmap = kzalloc(sizeof(struct gmap), GFP_KERNEL);
57 if (!gmap)
58 goto out;
59 INIT_LIST_HEAD(&gmap->crst_list);
60 INIT_LIST_HEAD(&gmap->children);
61 INIT_LIST_HEAD(&gmap->pt_list);
62 INIT_RADIX_TREE(&gmap->guest_to_host, GFP_KERNEL);
63 INIT_RADIX_TREE(&gmap->host_to_guest, GFP_ATOMIC);
64 INIT_RADIX_TREE(&gmap->host_to_rmap, GFP_ATOMIC);
65 spin_lock_init(&gmap->guest_table_lock);
66 spin_lock_init(&gmap->shadow_lock);
67 atomic_set(&gmap->ref_count, 1);
68 page = alloc_pages(GFP_KERNEL, CRST_ALLOC_ORDER);
69 if (!page)
70 goto out_free;
71 page->index = 0;
72 list_add(&page->lru, &gmap->crst_list);
73 table = (unsigned long *) page_to_phys(page);
74 crst_table_init(table, etype);
75 gmap->table = table;
76 gmap->asce = atype | _ASCE_TABLE_LENGTH |
77 _ASCE_USER_BITS | __pa(table);
78 gmap->asce_end = limit;
79 return gmap;
81 out_free:
82 kfree(gmap);
83 out:
84 return NULL;
87 /**
88 * gmap_create - create a guest address space
89 * @mm: pointer to the parent mm_struct
90 * @limit: maximum size of the gmap address space
92 * Returns a guest address space structure.
94 struct gmap *gmap_create(struct mm_struct *mm, unsigned long limit)
96 struct gmap *gmap;
97 unsigned long gmap_asce;
99 gmap = gmap_alloc(limit);
100 if (!gmap)
101 return NULL;
102 gmap->mm = mm;
103 spin_lock(&mm->context.lock);
104 list_add_rcu(&gmap->list, &mm->context.gmap_list);
105 if (list_is_singular(&mm->context.gmap_list))
106 gmap_asce = gmap->asce;
107 else
108 gmap_asce = -1UL;
109 WRITE_ONCE(mm->context.gmap_asce, gmap_asce);
110 spin_unlock(&mm->context.lock);
111 return gmap;
113 EXPORT_SYMBOL_GPL(gmap_create);
115 static void gmap_flush_tlb(struct gmap *gmap)
117 if (MACHINE_HAS_IDTE)
118 __tlb_flush_idte(gmap->asce);
119 else
120 __tlb_flush_global();
123 static void gmap_radix_tree_free(struct radix_tree_root *root)
125 struct radix_tree_iter iter;
126 unsigned long indices[16];
127 unsigned long index;
128 void __rcu **slot;
129 int i, nr;
131 /* A radix tree is freed by deleting all of its entries */
132 index = 0;
133 do {
134 nr = 0;
135 radix_tree_for_each_slot(slot, root, &iter, index) {
136 indices[nr] = iter.index;
137 if (++nr == 16)
138 break;
140 for (i = 0; i < nr; i++) {
141 index = indices[i];
142 radix_tree_delete(root, index);
144 } while (nr > 0);
147 static void gmap_rmap_radix_tree_free(struct radix_tree_root *root)
149 struct gmap_rmap *rmap, *rnext, *head;
150 struct radix_tree_iter iter;
151 unsigned long indices[16];
152 unsigned long index;
153 void __rcu **slot;
154 int i, nr;
156 /* A radix tree is freed by deleting all of its entries */
157 index = 0;
158 do {
159 nr = 0;
160 radix_tree_for_each_slot(slot, root, &iter, index) {
161 indices[nr] = iter.index;
162 if (++nr == 16)
163 break;
165 for (i = 0; i < nr; i++) {
166 index = indices[i];
167 head = radix_tree_delete(root, index);
168 gmap_for_each_rmap_safe(rmap, rnext, head)
169 kfree(rmap);
171 } while (nr > 0);
175 * gmap_free - free a guest address space
176 * @gmap: pointer to the guest address space structure
178 * No locks required. There are no references to this gmap anymore.
180 static void gmap_free(struct gmap *gmap)
182 struct page *page, *next;
184 /* Flush tlb of all gmaps (if not already done for shadows) */
185 if (!(gmap_is_shadow(gmap) && gmap->removed))
186 gmap_flush_tlb(gmap);
187 /* Free all segment & region tables. */
188 list_for_each_entry_safe(page, next, &gmap->crst_list, lru)
189 __free_pages(page, CRST_ALLOC_ORDER);
190 gmap_radix_tree_free(&gmap->guest_to_host);
191 gmap_radix_tree_free(&gmap->host_to_guest);
193 /* Free additional data for a shadow gmap */
194 if (gmap_is_shadow(gmap)) {
195 /* Free all page tables. */
196 list_for_each_entry_safe(page, next, &gmap->pt_list, lru)
197 page_table_free_pgste(page);
198 gmap_rmap_radix_tree_free(&gmap->host_to_rmap);
199 /* Release reference to the parent */
200 gmap_put(gmap->parent);
203 kfree(gmap);
207 * gmap_get - increase reference counter for guest address space
208 * @gmap: pointer to the guest address space structure
210 * Returns the gmap pointer
212 struct gmap *gmap_get(struct gmap *gmap)
214 atomic_inc(&gmap->ref_count);
215 return gmap;
217 EXPORT_SYMBOL_GPL(gmap_get);
220 * gmap_put - decrease reference counter for guest address space
221 * @gmap: pointer to the guest address space structure
223 * If the reference counter reaches zero the guest address space is freed.
225 void gmap_put(struct gmap *gmap)
227 if (atomic_dec_return(&gmap->ref_count) == 0)
228 gmap_free(gmap);
230 EXPORT_SYMBOL_GPL(gmap_put);
233 * gmap_remove - remove a guest address space but do not free it yet
234 * @gmap: pointer to the guest address space structure
236 void gmap_remove(struct gmap *gmap)
238 struct gmap *sg, *next;
239 unsigned long gmap_asce;
241 /* Remove all shadow gmaps linked to this gmap */
242 if (!list_empty(&gmap->children)) {
243 spin_lock(&gmap->shadow_lock);
244 list_for_each_entry_safe(sg, next, &gmap->children, list) {
245 list_del(&sg->list);
246 gmap_put(sg);
248 spin_unlock(&gmap->shadow_lock);
250 /* Remove gmap from the pre-mm list */
251 spin_lock(&gmap->mm->context.lock);
252 list_del_rcu(&gmap->list);
253 if (list_empty(&gmap->mm->context.gmap_list))
254 gmap_asce = 0;
255 else if (list_is_singular(&gmap->mm->context.gmap_list))
256 gmap_asce = list_first_entry(&gmap->mm->context.gmap_list,
257 struct gmap, list)->asce;
258 else
259 gmap_asce = -1UL;
260 WRITE_ONCE(gmap->mm->context.gmap_asce, gmap_asce);
261 spin_unlock(&gmap->mm->context.lock);
262 synchronize_rcu();
263 /* Put reference */
264 gmap_put(gmap);
266 EXPORT_SYMBOL_GPL(gmap_remove);
269 * gmap_enable - switch primary space to the guest address space
270 * @gmap: pointer to the guest address space structure
272 void gmap_enable(struct gmap *gmap)
274 S390_lowcore.gmap = (unsigned long) gmap;
276 EXPORT_SYMBOL_GPL(gmap_enable);
279 * gmap_disable - switch back to the standard primary address space
280 * @gmap: pointer to the guest address space structure
282 void gmap_disable(struct gmap *gmap)
284 S390_lowcore.gmap = 0UL;
286 EXPORT_SYMBOL_GPL(gmap_disable);
289 * gmap_get_enabled - get a pointer to the currently enabled gmap
291 * Returns a pointer to the currently enabled gmap. 0 if none is enabled.
293 struct gmap *gmap_get_enabled(void)
295 return (struct gmap *) S390_lowcore.gmap;
297 EXPORT_SYMBOL_GPL(gmap_get_enabled);
300 * gmap_alloc_table is assumed to be called with mmap_sem held
302 static int gmap_alloc_table(struct gmap *gmap, unsigned long *table,
303 unsigned long init, unsigned long gaddr)
305 struct page *page;
306 unsigned long *new;
308 /* since we dont free the gmap table until gmap_free we can unlock */
309 page = alloc_pages(GFP_KERNEL, CRST_ALLOC_ORDER);
310 if (!page)
311 return -ENOMEM;
312 new = (unsigned long *) page_to_phys(page);
313 crst_table_init(new, init);
314 spin_lock(&gmap->guest_table_lock);
315 if (*table & _REGION_ENTRY_INVALID) {
316 list_add(&page->lru, &gmap->crst_list);
317 *table = (unsigned long) new | _REGION_ENTRY_LENGTH |
318 (*table & _REGION_ENTRY_TYPE_MASK);
319 page->index = gaddr;
320 page = NULL;
322 spin_unlock(&gmap->guest_table_lock);
323 if (page)
324 __free_pages(page, CRST_ALLOC_ORDER);
325 return 0;
329 * __gmap_segment_gaddr - find virtual address from segment pointer
330 * @entry: pointer to a segment table entry in the guest address space
332 * Returns the virtual address in the guest address space for the segment
334 static unsigned long __gmap_segment_gaddr(unsigned long *entry)
336 struct page *page;
337 unsigned long offset, mask;
339 offset = (unsigned long) entry / sizeof(unsigned long);
340 offset = (offset & (PTRS_PER_PMD - 1)) * PMD_SIZE;
341 mask = ~(PTRS_PER_PMD * sizeof(pmd_t) - 1);
342 page = virt_to_page((void *)((unsigned long) entry & mask));
343 return page->index + offset;
347 * __gmap_unlink_by_vmaddr - unlink a single segment via a host address
348 * @gmap: pointer to the guest address space structure
349 * @vmaddr: address in the host process address space
351 * Returns 1 if a TLB flush is required
353 static int __gmap_unlink_by_vmaddr(struct gmap *gmap, unsigned long vmaddr)
355 unsigned long *entry;
356 int flush = 0;
358 BUG_ON(gmap_is_shadow(gmap));
359 spin_lock(&gmap->guest_table_lock);
360 entry = radix_tree_delete(&gmap->host_to_guest, vmaddr >> PMD_SHIFT);
361 if (entry) {
362 flush = (*entry != _SEGMENT_ENTRY_EMPTY);
363 *entry = _SEGMENT_ENTRY_EMPTY;
365 spin_unlock(&gmap->guest_table_lock);
366 return flush;
370 * __gmap_unmap_by_gaddr - unmap a single segment via a guest address
371 * @gmap: pointer to the guest address space structure
372 * @gaddr: address in the guest address space
374 * Returns 1 if a TLB flush is required
376 static int __gmap_unmap_by_gaddr(struct gmap *gmap, unsigned long gaddr)
378 unsigned long vmaddr;
380 vmaddr = (unsigned long) radix_tree_delete(&gmap->guest_to_host,
381 gaddr >> PMD_SHIFT);
382 return vmaddr ? __gmap_unlink_by_vmaddr(gmap, vmaddr) : 0;
386 * gmap_unmap_segment - unmap segment from the guest address space
387 * @gmap: pointer to the guest address space structure
388 * @to: address in the guest address space
389 * @len: length of the memory area to unmap
391 * Returns 0 if the unmap succeeded, -EINVAL if not.
393 int gmap_unmap_segment(struct gmap *gmap, unsigned long to, unsigned long len)
395 unsigned long off;
396 int flush;
398 BUG_ON(gmap_is_shadow(gmap));
399 if ((to | len) & (PMD_SIZE - 1))
400 return -EINVAL;
401 if (len == 0 || to + len < to)
402 return -EINVAL;
404 flush = 0;
405 down_write(&gmap->mm->mmap_sem);
406 for (off = 0; off < len; off += PMD_SIZE)
407 flush |= __gmap_unmap_by_gaddr(gmap, to + off);
408 up_write(&gmap->mm->mmap_sem);
409 if (flush)
410 gmap_flush_tlb(gmap);
411 return 0;
413 EXPORT_SYMBOL_GPL(gmap_unmap_segment);
416 * gmap_map_segment - map a segment to the guest address space
417 * @gmap: pointer to the guest address space structure
418 * @from: source address in the parent address space
419 * @to: target address in the guest address space
420 * @len: length of the memory area to map
422 * Returns 0 if the mmap succeeded, -EINVAL or -ENOMEM if not.
424 int gmap_map_segment(struct gmap *gmap, unsigned long from,
425 unsigned long to, unsigned long len)
427 unsigned long off;
428 int flush;
430 BUG_ON(gmap_is_shadow(gmap));
431 if ((from | to | len) & (PMD_SIZE - 1))
432 return -EINVAL;
433 if (len == 0 || from + len < from || to + len < to ||
434 from + len - 1 > TASK_SIZE_MAX || to + len - 1 > gmap->asce_end)
435 return -EINVAL;
437 flush = 0;
438 down_write(&gmap->mm->mmap_sem);
439 for (off = 0; off < len; off += PMD_SIZE) {
440 /* Remove old translation */
441 flush |= __gmap_unmap_by_gaddr(gmap, to + off);
442 /* Store new translation */
443 if (radix_tree_insert(&gmap->guest_to_host,
444 (to + off) >> PMD_SHIFT,
445 (void *) from + off))
446 break;
448 up_write(&gmap->mm->mmap_sem);
449 if (flush)
450 gmap_flush_tlb(gmap);
451 if (off >= len)
452 return 0;
453 gmap_unmap_segment(gmap, to, len);
454 return -ENOMEM;
456 EXPORT_SYMBOL_GPL(gmap_map_segment);
459 * __gmap_translate - translate a guest address to a user space address
460 * @gmap: pointer to guest mapping meta data structure
461 * @gaddr: guest address
463 * Returns user space address which corresponds to the guest address or
464 * -EFAULT if no such mapping exists.
465 * This function does not establish potentially missing page table entries.
466 * The mmap_sem of the mm that belongs to the address space must be held
467 * when this function gets called.
469 * Note: Can also be called for shadow gmaps.
471 unsigned long __gmap_translate(struct gmap *gmap, unsigned long gaddr)
473 unsigned long vmaddr;
475 vmaddr = (unsigned long)
476 radix_tree_lookup(&gmap->guest_to_host, gaddr >> PMD_SHIFT);
477 /* Note: guest_to_host is empty for a shadow gmap */
478 return vmaddr ? (vmaddr | (gaddr & ~PMD_MASK)) : -EFAULT;
480 EXPORT_SYMBOL_GPL(__gmap_translate);
483 * gmap_translate - translate a guest address to a user space address
484 * @gmap: pointer to guest mapping meta data structure
485 * @gaddr: guest address
487 * Returns user space address which corresponds to the guest address or
488 * -EFAULT if no such mapping exists.
489 * This function does not establish potentially missing page table entries.
491 unsigned long gmap_translate(struct gmap *gmap, unsigned long gaddr)
493 unsigned long rc;
495 down_read(&gmap->mm->mmap_sem);
496 rc = __gmap_translate(gmap, gaddr);
497 up_read(&gmap->mm->mmap_sem);
498 return rc;
500 EXPORT_SYMBOL_GPL(gmap_translate);
503 * gmap_unlink - disconnect a page table from the gmap shadow tables
504 * @gmap: pointer to guest mapping meta data structure
505 * @table: pointer to the host page table
506 * @vmaddr: vm address associated with the host page table
508 void gmap_unlink(struct mm_struct *mm, unsigned long *table,
509 unsigned long vmaddr)
511 struct gmap *gmap;
512 int flush;
514 rcu_read_lock();
515 list_for_each_entry_rcu(gmap, &mm->context.gmap_list, list) {
516 flush = __gmap_unlink_by_vmaddr(gmap, vmaddr);
517 if (flush)
518 gmap_flush_tlb(gmap);
520 rcu_read_unlock();
524 * gmap_link - set up shadow page tables to connect a host to a guest address
525 * @gmap: pointer to guest mapping meta data structure
526 * @gaddr: guest address
527 * @vmaddr: vm address
529 * Returns 0 on success, -ENOMEM for out of memory conditions, and -EFAULT
530 * if the vm address is already mapped to a different guest segment.
531 * The mmap_sem of the mm that belongs to the address space must be held
532 * when this function gets called.
534 int __gmap_link(struct gmap *gmap, unsigned long gaddr, unsigned long vmaddr)
536 struct mm_struct *mm;
537 unsigned long *table;
538 spinlock_t *ptl;
539 pgd_t *pgd;
540 p4d_t *p4d;
541 pud_t *pud;
542 pmd_t *pmd;
543 int rc;
545 BUG_ON(gmap_is_shadow(gmap));
546 /* Create higher level tables in the gmap page table */
547 table = gmap->table;
548 if ((gmap->asce & _ASCE_TYPE_MASK) >= _ASCE_TYPE_REGION1) {
549 table += (gaddr & _REGION1_INDEX) >> _REGION1_SHIFT;
550 if ((*table & _REGION_ENTRY_INVALID) &&
551 gmap_alloc_table(gmap, table, _REGION2_ENTRY_EMPTY,
552 gaddr & _REGION1_MASK))
553 return -ENOMEM;
554 table = (unsigned long *)(*table & _REGION_ENTRY_ORIGIN);
556 if ((gmap->asce & _ASCE_TYPE_MASK) >= _ASCE_TYPE_REGION2) {
557 table += (gaddr & _REGION2_INDEX) >> _REGION2_SHIFT;
558 if ((*table & _REGION_ENTRY_INVALID) &&
559 gmap_alloc_table(gmap, table, _REGION3_ENTRY_EMPTY,
560 gaddr & _REGION2_MASK))
561 return -ENOMEM;
562 table = (unsigned long *)(*table & _REGION_ENTRY_ORIGIN);
564 if ((gmap->asce & _ASCE_TYPE_MASK) >= _ASCE_TYPE_REGION3) {
565 table += (gaddr & _REGION3_INDEX) >> _REGION3_SHIFT;
566 if ((*table & _REGION_ENTRY_INVALID) &&
567 gmap_alloc_table(gmap, table, _SEGMENT_ENTRY_EMPTY,
568 gaddr & _REGION3_MASK))
569 return -ENOMEM;
570 table = (unsigned long *)(*table & _REGION_ENTRY_ORIGIN);
572 table += (gaddr & _SEGMENT_INDEX) >> _SEGMENT_SHIFT;
573 /* Walk the parent mm page table */
574 mm = gmap->mm;
575 pgd = pgd_offset(mm, vmaddr);
576 VM_BUG_ON(pgd_none(*pgd));
577 p4d = p4d_offset(pgd, vmaddr);
578 VM_BUG_ON(p4d_none(*p4d));
579 pud = pud_offset(p4d, vmaddr);
580 VM_BUG_ON(pud_none(*pud));
581 /* large puds cannot yet be handled */
582 if (pud_large(*pud))
583 return -EFAULT;
584 pmd = pmd_offset(pud, vmaddr);
585 VM_BUG_ON(pmd_none(*pmd));
586 /* large pmds cannot yet be handled */
587 if (pmd_large(*pmd))
588 return -EFAULT;
589 /* Link gmap segment table entry location to page table. */
590 rc = radix_tree_preload(GFP_KERNEL);
591 if (rc)
592 return rc;
593 ptl = pmd_lock(mm, pmd);
594 spin_lock(&gmap->guest_table_lock);
595 if (*table == _SEGMENT_ENTRY_EMPTY) {
596 rc = radix_tree_insert(&gmap->host_to_guest,
597 vmaddr >> PMD_SHIFT, table);
598 if (!rc)
599 *table = pmd_val(*pmd);
600 } else
601 rc = 0;
602 spin_unlock(&gmap->guest_table_lock);
603 spin_unlock(ptl);
604 radix_tree_preload_end();
605 return rc;
609 * gmap_fault - resolve a fault on a guest address
610 * @gmap: pointer to guest mapping meta data structure
611 * @gaddr: guest address
612 * @fault_flags: flags to pass down to handle_mm_fault()
614 * Returns 0 on success, -ENOMEM for out of memory conditions, and -EFAULT
615 * if the vm address is already mapped to a different guest segment.
617 int gmap_fault(struct gmap *gmap, unsigned long gaddr,
618 unsigned int fault_flags)
620 unsigned long vmaddr;
621 int rc;
622 bool unlocked;
624 down_read(&gmap->mm->mmap_sem);
626 retry:
627 unlocked = false;
628 vmaddr = __gmap_translate(gmap, gaddr);
629 if (IS_ERR_VALUE(vmaddr)) {
630 rc = vmaddr;
631 goto out_up;
633 if (fixup_user_fault(current, gmap->mm, vmaddr, fault_flags,
634 &unlocked)) {
635 rc = -EFAULT;
636 goto out_up;
639 * In the case that fixup_user_fault unlocked the mmap_sem during
640 * faultin redo __gmap_translate to not race with a map/unmap_segment.
642 if (unlocked)
643 goto retry;
645 rc = __gmap_link(gmap, gaddr, vmaddr);
646 out_up:
647 up_read(&gmap->mm->mmap_sem);
648 return rc;
650 EXPORT_SYMBOL_GPL(gmap_fault);
653 * this function is assumed to be called with mmap_sem held
655 void __gmap_zap(struct gmap *gmap, unsigned long gaddr)
657 unsigned long vmaddr;
658 spinlock_t *ptl;
659 pte_t *ptep;
661 /* Find the vm address for the guest address */
662 vmaddr = (unsigned long) radix_tree_lookup(&gmap->guest_to_host,
663 gaddr >> PMD_SHIFT);
664 if (vmaddr) {
665 vmaddr |= gaddr & ~PMD_MASK;
666 /* Get pointer to the page table entry */
667 ptep = get_locked_pte(gmap->mm, vmaddr, &ptl);
668 if (likely(ptep))
669 ptep_zap_unused(gmap->mm, vmaddr, ptep, 0);
670 pte_unmap_unlock(ptep, ptl);
673 EXPORT_SYMBOL_GPL(__gmap_zap);
675 void gmap_discard(struct gmap *gmap, unsigned long from, unsigned long to)
677 unsigned long gaddr, vmaddr, size;
678 struct vm_area_struct *vma;
680 down_read(&gmap->mm->mmap_sem);
681 for (gaddr = from; gaddr < to;
682 gaddr = (gaddr + PMD_SIZE) & PMD_MASK) {
683 /* Find the vm address for the guest address */
684 vmaddr = (unsigned long)
685 radix_tree_lookup(&gmap->guest_to_host,
686 gaddr >> PMD_SHIFT);
687 if (!vmaddr)
688 continue;
689 vmaddr |= gaddr & ~PMD_MASK;
690 /* Find vma in the parent mm */
691 vma = find_vma(gmap->mm, vmaddr);
692 if (!vma)
693 continue;
694 size = min(to - gaddr, PMD_SIZE - (gaddr & ~PMD_MASK));
695 zap_page_range(vma, vmaddr, size);
697 up_read(&gmap->mm->mmap_sem);
699 EXPORT_SYMBOL_GPL(gmap_discard);
701 static LIST_HEAD(gmap_notifier_list);
702 static DEFINE_SPINLOCK(gmap_notifier_lock);
705 * gmap_register_pte_notifier - register a pte invalidation callback
706 * @nb: pointer to the gmap notifier block
708 void gmap_register_pte_notifier(struct gmap_notifier *nb)
710 spin_lock(&gmap_notifier_lock);
711 list_add_rcu(&nb->list, &gmap_notifier_list);
712 spin_unlock(&gmap_notifier_lock);
714 EXPORT_SYMBOL_GPL(gmap_register_pte_notifier);
717 * gmap_unregister_pte_notifier - remove a pte invalidation callback
718 * @nb: pointer to the gmap notifier block
720 void gmap_unregister_pte_notifier(struct gmap_notifier *nb)
722 spin_lock(&gmap_notifier_lock);
723 list_del_rcu(&nb->list);
724 spin_unlock(&gmap_notifier_lock);
725 synchronize_rcu();
727 EXPORT_SYMBOL_GPL(gmap_unregister_pte_notifier);
730 * gmap_call_notifier - call all registered invalidation callbacks
731 * @gmap: pointer to guest mapping meta data structure
732 * @start: start virtual address in the guest address space
733 * @end: end virtual address in the guest address space
735 static void gmap_call_notifier(struct gmap *gmap, unsigned long start,
736 unsigned long end)
738 struct gmap_notifier *nb;
740 list_for_each_entry(nb, &gmap_notifier_list, list)
741 nb->notifier_call(gmap, start, end);
745 * gmap_table_walk - walk the gmap page tables
746 * @gmap: pointer to guest mapping meta data structure
747 * @gaddr: virtual address in the guest address space
748 * @level: page table level to stop at
750 * Returns a table entry pointer for the given guest address and @level
751 * @level=0 : returns a pointer to a page table table entry (or NULL)
752 * @level=1 : returns a pointer to a segment table entry (or NULL)
753 * @level=2 : returns a pointer to a region-3 table entry (or NULL)
754 * @level=3 : returns a pointer to a region-2 table entry (or NULL)
755 * @level=4 : returns a pointer to a region-1 table entry (or NULL)
757 * Returns NULL if the gmap page tables could not be walked to the
758 * requested level.
760 * Note: Can also be called for shadow gmaps.
762 static inline unsigned long *gmap_table_walk(struct gmap *gmap,
763 unsigned long gaddr, int level)
765 unsigned long *table;
767 if ((gmap->asce & _ASCE_TYPE_MASK) + 4 < (level * 4))
768 return NULL;
769 if (gmap_is_shadow(gmap) && gmap->removed)
770 return NULL;
771 if (gaddr & (-1UL << (31 + ((gmap->asce & _ASCE_TYPE_MASK) >> 2)*11)))
772 return NULL;
773 table = gmap->table;
774 switch (gmap->asce & _ASCE_TYPE_MASK) {
775 case _ASCE_TYPE_REGION1:
776 table += (gaddr & _REGION1_INDEX) >> _REGION1_SHIFT;
777 if (level == 4)
778 break;
779 if (*table & _REGION_ENTRY_INVALID)
780 return NULL;
781 table = (unsigned long *)(*table & _REGION_ENTRY_ORIGIN);
782 /* Fallthrough */
783 case _ASCE_TYPE_REGION2:
784 table += (gaddr & _REGION2_INDEX) >> _REGION2_SHIFT;
785 if (level == 3)
786 break;
787 if (*table & _REGION_ENTRY_INVALID)
788 return NULL;
789 table = (unsigned long *)(*table & _REGION_ENTRY_ORIGIN);
790 /* Fallthrough */
791 case _ASCE_TYPE_REGION3:
792 table += (gaddr & _REGION3_INDEX) >> _REGION3_SHIFT;
793 if (level == 2)
794 break;
795 if (*table & _REGION_ENTRY_INVALID)
796 return NULL;
797 table = (unsigned long *)(*table & _REGION_ENTRY_ORIGIN);
798 /* Fallthrough */
799 case _ASCE_TYPE_SEGMENT:
800 table += (gaddr & _SEGMENT_INDEX) >> _SEGMENT_SHIFT;
801 if (level == 1)
802 break;
803 if (*table & _REGION_ENTRY_INVALID)
804 return NULL;
805 table = (unsigned long *)(*table & _SEGMENT_ENTRY_ORIGIN);
806 table += (gaddr & _PAGE_INDEX) >> _PAGE_SHIFT;
808 return table;
812 * gmap_pte_op_walk - walk the gmap page table, get the page table lock
813 * and return the pte pointer
814 * @gmap: pointer to guest mapping meta data structure
815 * @gaddr: virtual address in the guest address space
816 * @ptl: pointer to the spinlock pointer
818 * Returns a pointer to the locked pte for a guest address, or NULL
820 * Note: Can also be called for shadow gmaps.
822 static pte_t *gmap_pte_op_walk(struct gmap *gmap, unsigned long gaddr,
823 spinlock_t **ptl)
825 unsigned long *table;
827 if (gmap_is_shadow(gmap))
828 spin_lock(&gmap->guest_table_lock);
829 /* Walk the gmap page table, lock and get pte pointer */
830 table = gmap_table_walk(gmap, gaddr, 1); /* get segment pointer */
831 if (!table || *table & _SEGMENT_ENTRY_INVALID) {
832 if (gmap_is_shadow(gmap))
833 spin_unlock(&gmap->guest_table_lock);
834 return NULL;
836 if (gmap_is_shadow(gmap)) {
837 *ptl = &gmap->guest_table_lock;
838 return pte_offset_map((pmd_t *) table, gaddr);
840 return pte_alloc_map_lock(gmap->mm, (pmd_t *) table, gaddr, ptl);
844 * gmap_pte_op_fixup - force a page in and connect the gmap page table
845 * @gmap: pointer to guest mapping meta data structure
846 * @gaddr: virtual address in the guest address space
847 * @vmaddr: address in the host process address space
848 * @prot: indicates access rights: PROT_NONE, PROT_READ or PROT_WRITE
850 * Returns 0 if the caller can retry __gmap_translate (might fail again),
851 * -ENOMEM if out of memory and -EFAULT if anything goes wrong while fixing
852 * up or connecting the gmap page table.
854 static int gmap_pte_op_fixup(struct gmap *gmap, unsigned long gaddr,
855 unsigned long vmaddr, int prot)
857 struct mm_struct *mm = gmap->mm;
858 unsigned int fault_flags;
859 bool unlocked = false;
861 BUG_ON(gmap_is_shadow(gmap));
862 fault_flags = (prot == PROT_WRITE) ? FAULT_FLAG_WRITE : 0;
863 if (fixup_user_fault(current, mm, vmaddr, fault_flags, &unlocked))
864 return -EFAULT;
865 if (unlocked)
866 /* lost mmap_sem, caller has to retry __gmap_translate */
867 return 0;
868 /* Connect the page tables */
869 return __gmap_link(gmap, gaddr, vmaddr);
873 * gmap_pte_op_end - release the page table lock
874 * @ptl: pointer to the spinlock pointer
876 static void gmap_pte_op_end(spinlock_t *ptl)
878 spin_unlock(ptl);
882 * gmap_protect_range - remove access rights to memory and set pgste bits
883 * @gmap: pointer to guest mapping meta data structure
884 * @gaddr: virtual address in the guest address space
885 * @len: size of area
886 * @prot: indicates access rights: PROT_NONE, PROT_READ or PROT_WRITE
887 * @bits: pgste notification bits to set
889 * Returns 0 if successfully protected, -ENOMEM if out of memory and
890 * -EFAULT if gaddr is invalid (or mapping for shadows is missing).
892 * Called with sg->mm->mmap_sem in read.
894 * Note: Can also be called for shadow gmaps.
896 static int gmap_protect_range(struct gmap *gmap, unsigned long gaddr,
897 unsigned long len, int prot, unsigned long bits)
899 unsigned long vmaddr;
900 spinlock_t *ptl;
901 pte_t *ptep;
902 int rc;
904 while (len) {
905 rc = -EAGAIN;
906 ptep = gmap_pte_op_walk(gmap, gaddr, &ptl);
907 if (ptep) {
908 rc = ptep_force_prot(gmap->mm, gaddr, ptep, prot, bits);
909 gmap_pte_op_end(ptl);
911 if (rc) {
912 vmaddr = __gmap_translate(gmap, gaddr);
913 if (IS_ERR_VALUE(vmaddr))
914 return vmaddr;
915 rc = gmap_pte_op_fixup(gmap, gaddr, vmaddr, prot);
916 if (rc)
917 return rc;
918 continue;
920 gaddr += PAGE_SIZE;
921 len -= PAGE_SIZE;
923 return 0;
927 * gmap_mprotect_notify - change access rights for a range of ptes and
928 * call the notifier if any pte changes again
929 * @gmap: pointer to guest mapping meta data structure
930 * @gaddr: virtual address in the guest address space
931 * @len: size of area
932 * @prot: indicates access rights: PROT_NONE, PROT_READ or PROT_WRITE
934 * Returns 0 if for each page in the given range a gmap mapping exists,
935 * the new access rights could be set and the notifier could be armed.
936 * If the gmap mapping is missing for one or more pages -EFAULT is
937 * returned. If no memory could be allocated -ENOMEM is returned.
938 * This function establishes missing page table entries.
940 int gmap_mprotect_notify(struct gmap *gmap, unsigned long gaddr,
941 unsigned long len, int prot)
943 int rc;
945 if ((gaddr & ~PAGE_MASK) || (len & ~PAGE_MASK) || gmap_is_shadow(gmap))
946 return -EINVAL;
947 if (!MACHINE_HAS_ESOP && prot == PROT_READ)
948 return -EINVAL;
949 down_read(&gmap->mm->mmap_sem);
950 rc = gmap_protect_range(gmap, gaddr, len, prot, PGSTE_IN_BIT);
951 up_read(&gmap->mm->mmap_sem);
952 return rc;
954 EXPORT_SYMBOL_GPL(gmap_mprotect_notify);
957 * gmap_read_table - get an unsigned long value from a guest page table using
958 * absolute addressing, without marking the page referenced.
959 * @gmap: pointer to guest mapping meta data structure
960 * @gaddr: virtual address in the guest address space
961 * @val: pointer to the unsigned long value to return
963 * Returns 0 if the value was read, -ENOMEM if out of memory and -EFAULT
964 * if reading using the virtual address failed.
966 * Called with gmap->mm->mmap_sem in read.
968 int gmap_read_table(struct gmap *gmap, unsigned long gaddr, unsigned long *val)
970 unsigned long address, vmaddr;
971 spinlock_t *ptl;
972 pte_t *ptep, pte;
973 int rc;
975 while (1) {
976 rc = -EAGAIN;
977 ptep = gmap_pte_op_walk(gmap, gaddr, &ptl);
978 if (ptep) {
979 pte = *ptep;
980 if (pte_present(pte) && (pte_val(pte) & _PAGE_READ)) {
981 address = pte_val(pte) & PAGE_MASK;
982 address += gaddr & ~PAGE_MASK;
983 *val = *(unsigned long *) address;
984 pte_val(*ptep) |= _PAGE_YOUNG;
985 /* Do *NOT* clear the _PAGE_INVALID bit! */
986 rc = 0;
988 gmap_pte_op_end(ptl);
990 if (!rc)
991 break;
992 vmaddr = __gmap_translate(gmap, gaddr);
993 if (IS_ERR_VALUE(vmaddr)) {
994 rc = vmaddr;
995 break;
997 rc = gmap_pte_op_fixup(gmap, gaddr, vmaddr, PROT_READ);
998 if (rc)
999 break;
1001 return rc;
1003 EXPORT_SYMBOL_GPL(gmap_read_table);
1006 * gmap_insert_rmap - add a rmap to the host_to_rmap radix tree
1007 * @sg: pointer to the shadow guest address space structure
1008 * @vmaddr: vm address associated with the rmap
1009 * @rmap: pointer to the rmap structure
1011 * Called with the sg->guest_table_lock
1013 static inline void gmap_insert_rmap(struct gmap *sg, unsigned long vmaddr,
1014 struct gmap_rmap *rmap)
1016 void __rcu **slot;
1018 BUG_ON(!gmap_is_shadow(sg));
1019 slot = radix_tree_lookup_slot(&sg->host_to_rmap, vmaddr >> PAGE_SHIFT);
1020 if (slot) {
1021 rmap->next = radix_tree_deref_slot_protected(slot,
1022 &sg->guest_table_lock);
1023 radix_tree_replace_slot(&sg->host_to_rmap, slot, rmap);
1024 } else {
1025 rmap->next = NULL;
1026 radix_tree_insert(&sg->host_to_rmap, vmaddr >> PAGE_SHIFT,
1027 rmap);
1032 * gmap_protect_rmap - modify access rights to memory and create an rmap
1033 * @sg: pointer to the shadow guest address space structure
1034 * @raddr: rmap address in the shadow gmap
1035 * @paddr: address in the parent guest address space
1036 * @len: length of the memory area to protect
1037 * @prot: indicates access rights: none, read-only or read-write
1039 * Returns 0 if successfully protected and the rmap was created, -ENOMEM
1040 * if out of memory and -EFAULT if paddr is invalid.
1042 static int gmap_protect_rmap(struct gmap *sg, unsigned long raddr,
1043 unsigned long paddr, unsigned long len, int prot)
1045 struct gmap *parent;
1046 struct gmap_rmap *rmap;
1047 unsigned long vmaddr;
1048 spinlock_t *ptl;
1049 pte_t *ptep;
1050 int rc;
1052 BUG_ON(!gmap_is_shadow(sg));
1053 parent = sg->parent;
1054 while (len) {
1055 vmaddr = __gmap_translate(parent, paddr);
1056 if (IS_ERR_VALUE(vmaddr))
1057 return vmaddr;
1058 rmap = kzalloc(sizeof(*rmap), GFP_KERNEL);
1059 if (!rmap)
1060 return -ENOMEM;
1061 rmap->raddr = raddr;
1062 rc = radix_tree_preload(GFP_KERNEL);
1063 if (rc) {
1064 kfree(rmap);
1065 return rc;
1067 rc = -EAGAIN;
1068 ptep = gmap_pte_op_walk(parent, paddr, &ptl);
1069 if (ptep) {
1070 spin_lock(&sg->guest_table_lock);
1071 rc = ptep_force_prot(parent->mm, paddr, ptep, prot,
1072 PGSTE_VSIE_BIT);
1073 if (!rc)
1074 gmap_insert_rmap(sg, vmaddr, rmap);
1075 spin_unlock(&sg->guest_table_lock);
1076 gmap_pte_op_end(ptl);
1078 radix_tree_preload_end();
1079 if (rc) {
1080 kfree(rmap);
1081 rc = gmap_pte_op_fixup(parent, paddr, vmaddr, prot);
1082 if (rc)
1083 return rc;
1084 continue;
1086 paddr += PAGE_SIZE;
1087 len -= PAGE_SIZE;
1089 return 0;
1092 #define _SHADOW_RMAP_MASK 0x7
1093 #define _SHADOW_RMAP_REGION1 0x5
1094 #define _SHADOW_RMAP_REGION2 0x4
1095 #define _SHADOW_RMAP_REGION3 0x3
1096 #define _SHADOW_RMAP_SEGMENT 0x2
1097 #define _SHADOW_RMAP_PGTABLE 0x1
1100 * gmap_idte_one - invalidate a single region or segment table entry
1101 * @asce: region or segment table *origin* + table-type bits
1102 * @vaddr: virtual address to identify the table entry to flush
1104 * The invalid bit of a single region or segment table entry is set
1105 * and the associated TLB entries depending on the entry are flushed.
1106 * The table-type of the @asce identifies the portion of the @vaddr
1107 * that is used as the invalidation index.
1109 static inline void gmap_idte_one(unsigned long asce, unsigned long vaddr)
1111 asm volatile(
1112 " .insn rrf,0xb98e0000,%0,%1,0,0"
1113 : : "a" (asce), "a" (vaddr) : "cc", "memory");
1117 * gmap_unshadow_page - remove a page from a shadow page table
1118 * @sg: pointer to the shadow guest address space structure
1119 * @raddr: rmap address in the shadow guest address space
1121 * Called with the sg->guest_table_lock
1123 static void gmap_unshadow_page(struct gmap *sg, unsigned long raddr)
1125 unsigned long *table;
1127 BUG_ON(!gmap_is_shadow(sg));
1128 table = gmap_table_walk(sg, raddr, 0); /* get page table pointer */
1129 if (!table || *table & _PAGE_INVALID)
1130 return;
1131 gmap_call_notifier(sg, raddr, raddr + _PAGE_SIZE - 1);
1132 ptep_unshadow_pte(sg->mm, raddr, (pte_t *) table);
1136 * __gmap_unshadow_pgt - remove all entries from a shadow page table
1137 * @sg: pointer to the shadow guest address space structure
1138 * @raddr: rmap address in the shadow guest address space
1139 * @pgt: pointer to the start of a shadow page table
1141 * Called with the sg->guest_table_lock
1143 static void __gmap_unshadow_pgt(struct gmap *sg, unsigned long raddr,
1144 unsigned long *pgt)
1146 int i;
1148 BUG_ON(!gmap_is_shadow(sg));
1149 for (i = 0; i < _PAGE_ENTRIES; i++, raddr += _PAGE_SIZE)
1150 pgt[i] = _PAGE_INVALID;
1154 * gmap_unshadow_pgt - remove a shadow page table from a segment entry
1155 * @sg: pointer to the shadow guest address space structure
1156 * @raddr: address in the shadow guest address space
1158 * Called with the sg->guest_table_lock
1160 static void gmap_unshadow_pgt(struct gmap *sg, unsigned long raddr)
1162 unsigned long sto, *ste, *pgt;
1163 struct page *page;
1165 BUG_ON(!gmap_is_shadow(sg));
1166 ste = gmap_table_walk(sg, raddr, 1); /* get segment pointer */
1167 if (!ste || !(*ste & _SEGMENT_ENTRY_ORIGIN))
1168 return;
1169 gmap_call_notifier(sg, raddr, raddr + _SEGMENT_SIZE - 1);
1170 sto = (unsigned long) (ste - ((raddr & _SEGMENT_INDEX) >> _SEGMENT_SHIFT));
1171 gmap_idte_one(sto | _ASCE_TYPE_SEGMENT, raddr);
1172 pgt = (unsigned long *)(*ste & _SEGMENT_ENTRY_ORIGIN);
1173 *ste = _SEGMENT_ENTRY_EMPTY;
1174 __gmap_unshadow_pgt(sg, raddr, pgt);
1175 /* Free page table */
1176 page = pfn_to_page(__pa(pgt) >> PAGE_SHIFT);
1177 list_del(&page->lru);
1178 page_table_free_pgste(page);
1182 * __gmap_unshadow_sgt - remove all entries from a shadow segment table
1183 * @sg: pointer to the shadow guest address space structure
1184 * @raddr: rmap address in the shadow guest address space
1185 * @sgt: pointer to the start of a shadow segment table
1187 * Called with the sg->guest_table_lock
1189 static void __gmap_unshadow_sgt(struct gmap *sg, unsigned long raddr,
1190 unsigned long *sgt)
1192 unsigned long asce, *pgt;
1193 struct page *page;
1194 int i;
1196 BUG_ON(!gmap_is_shadow(sg));
1197 asce = (unsigned long) sgt | _ASCE_TYPE_SEGMENT;
1198 for (i = 0; i < _CRST_ENTRIES; i++, raddr += _SEGMENT_SIZE) {
1199 if (!(sgt[i] & _SEGMENT_ENTRY_ORIGIN))
1200 continue;
1201 pgt = (unsigned long *)(sgt[i] & _REGION_ENTRY_ORIGIN);
1202 sgt[i] = _SEGMENT_ENTRY_EMPTY;
1203 __gmap_unshadow_pgt(sg, raddr, pgt);
1204 /* Free page table */
1205 page = pfn_to_page(__pa(pgt) >> PAGE_SHIFT);
1206 list_del(&page->lru);
1207 page_table_free_pgste(page);
1212 * gmap_unshadow_sgt - remove a shadow segment table from a region-3 entry
1213 * @sg: pointer to the shadow guest address space structure
1214 * @raddr: rmap address in the shadow guest address space
1216 * Called with the shadow->guest_table_lock
1218 static void gmap_unshadow_sgt(struct gmap *sg, unsigned long raddr)
1220 unsigned long r3o, *r3e, *sgt;
1221 struct page *page;
1223 BUG_ON(!gmap_is_shadow(sg));
1224 r3e = gmap_table_walk(sg, raddr, 2); /* get region-3 pointer */
1225 if (!r3e || !(*r3e & _REGION_ENTRY_ORIGIN))
1226 return;
1227 gmap_call_notifier(sg, raddr, raddr + _REGION3_SIZE - 1);
1228 r3o = (unsigned long) (r3e - ((raddr & _REGION3_INDEX) >> _REGION3_SHIFT));
1229 gmap_idte_one(r3o | _ASCE_TYPE_REGION3, raddr);
1230 sgt = (unsigned long *)(*r3e & _REGION_ENTRY_ORIGIN);
1231 *r3e = _REGION3_ENTRY_EMPTY;
1232 __gmap_unshadow_sgt(sg, raddr, sgt);
1233 /* Free segment table */
1234 page = pfn_to_page(__pa(sgt) >> PAGE_SHIFT);
1235 list_del(&page->lru);
1236 __free_pages(page, CRST_ALLOC_ORDER);
1240 * __gmap_unshadow_r3t - remove all entries from a shadow region-3 table
1241 * @sg: pointer to the shadow guest address space structure
1242 * @raddr: address in the shadow guest address space
1243 * @r3t: pointer to the start of a shadow region-3 table
1245 * Called with the sg->guest_table_lock
1247 static void __gmap_unshadow_r3t(struct gmap *sg, unsigned long raddr,
1248 unsigned long *r3t)
1250 unsigned long asce, *sgt;
1251 struct page *page;
1252 int i;
1254 BUG_ON(!gmap_is_shadow(sg));
1255 asce = (unsigned long) r3t | _ASCE_TYPE_REGION3;
1256 for (i = 0; i < _CRST_ENTRIES; i++, raddr += _REGION3_SIZE) {
1257 if (!(r3t[i] & _REGION_ENTRY_ORIGIN))
1258 continue;
1259 sgt = (unsigned long *)(r3t[i] & _REGION_ENTRY_ORIGIN);
1260 r3t[i] = _REGION3_ENTRY_EMPTY;
1261 __gmap_unshadow_sgt(sg, raddr, sgt);
1262 /* Free segment table */
1263 page = pfn_to_page(__pa(sgt) >> PAGE_SHIFT);
1264 list_del(&page->lru);
1265 __free_pages(page, CRST_ALLOC_ORDER);
1270 * gmap_unshadow_r3t - remove a shadow region-3 table from a region-2 entry
1271 * @sg: pointer to the shadow guest address space structure
1272 * @raddr: rmap address in the shadow guest address space
1274 * Called with the sg->guest_table_lock
1276 static void gmap_unshadow_r3t(struct gmap *sg, unsigned long raddr)
1278 unsigned long r2o, *r2e, *r3t;
1279 struct page *page;
1281 BUG_ON(!gmap_is_shadow(sg));
1282 r2e = gmap_table_walk(sg, raddr, 3); /* get region-2 pointer */
1283 if (!r2e || !(*r2e & _REGION_ENTRY_ORIGIN))
1284 return;
1285 gmap_call_notifier(sg, raddr, raddr + _REGION2_SIZE - 1);
1286 r2o = (unsigned long) (r2e - ((raddr & _REGION2_INDEX) >> _REGION2_SHIFT));
1287 gmap_idte_one(r2o | _ASCE_TYPE_REGION2, raddr);
1288 r3t = (unsigned long *)(*r2e & _REGION_ENTRY_ORIGIN);
1289 *r2e = _REGION2_ENTRY_EMPTY;
1290 __gmap_unshadow_r3t(sg, raddr, r3t);
1291 /* Free region 3 table */
1292 page = pfn_to_page(__pa(r3t) >> PAGE_SHIFT);
1293 list_del(&page->lru);
1294 __free_pages(page, CRST_ALLOC_ORDER);
1298 * __gmap_unshadow_r2t - remove all entries from a shadow region-2 table
1299 * @sg: pointer to the shadow guest address space structure
1300 * @raddr: rmap address in the shadow guest address space
1301 * @r2t: pointer to the start of a shadow region-2 table
1303 * Called with the sg->guest_table_lock
1305 static void __gmap_unshadow_r2t(struct gmap *sg, unsigned long raddr,
1306 unsigned long *r2t)
1308 unsigned long asce, *r3t;
1309 struct page *page;
1310 int i;
1312 BUG_ON(!gmap_is_shadow(sg));
1313 asce = (unsigned long) r2t | _ASCE_TYPE_REGION2;
1314 for (i = 0; i < _CRST_ENTRIES; i++, raddr += _REGION2_SIZE) {
1315 if (!(r2t[i] & _REGION_ENTRY_ORIGIN))
1316 continue;
1317 r3t = (unsigned long *)(r2t[i] & _REGION_ENTRY_ORIGIN);
1318 r2t[i] = _REGION2_ENTRY_EMPTY;
1319 __gmap_unshadow_r3t(sg, raddr, r3t);
1320 /* Free region 3 table */
1321 page = pfn_to_page(__pa(r3t) >> PAGE_SHIFT);
1322 list_del(&page->lru);
1323 __free_pages(page, CRST_ALLOC_ORDER);
1328 * gmap_unshadow_r2t - remove a shadow region-2 table from a region-1 entry
1329 * @sg: pointer to the shadow guest address space structure
1330 * @raddr: rmap address in the shadow guest address space
1332 * Called with the sg->guest_table_lock
1334 static void gmap_unshadow_r2t(struct gmap *sg, unsigned long raddr)
1336 unsigned long r1o, *r1e, *r2t;
1337 struct page *page;
1339 BUG_ON(!gmap_is_shadow(sg));
1340 r1e = gmap_table_walk(sg, raddr, 4); /* get region-1 pointer */
1341 if (!r1e || !(*r1e & _REGION_ENTRY_ORIGIN))
1342 return;
1343 gmap_call_notifier(sg, raddr, raddr + _REGION1_SIZE - 1);
1344 r1o = (unsigned long) (r1e - ((raddr & _REGION1_INDEX) >> _REGION1_SHIFT));
1345 gmap_idte_one(r1o | _ASCE_TYPE_REGION1, raddr);
1346 r2t = (unsigned long *)(*r1e & _REGION_ENTRY_ORIGIN);
1347 *r1e = _REGION1_ENTRY_EMPTY;
1348 __gmap_unshadow_r2t(sg, raddr, r2t);
1349 /* Free region 2 table */
1350 page = pfn_to_page(__pa(r2t) >> PAGE_SHIFT);
1351 list_del(&page->lru);
1352 __free_pages(page, CRST_ALLOC_ORDER);
1356 * __gmap_unshadow_r1t - remove all entries from a shadow region-1 table
1357 * @sg: pointer to the shadow guest address space structure
1358 * @raddr: rmap address in the shadow guest address space
1359 * @r1t: pointer to the start of a shadow region-1 table
1361 * Called with the shadow->guest_table_lock
1363 static void __gmap_unshadow_r1t(struct gmap *sg, unsigned long raddr,
1364 unsigned long *r1t)
1366 unsigned long asce, *r2t;
1367 struct page *page;
1368 int i;
1370 BUG_ON(!gmap_is_shadow(sg));
1371 asce = (unsigned long) r1t | _ASCE_TYPE_REGION1;
1372 for (i = 0; i < _CRST_ENTRIES; i++, raddr += _REGION1_SIZE) {
1373 if (!(r1t[i] & _REGION_ENTRY_ORIGIN))
1374 continue;
1375 r2t = (unsigned long *)(r1t[i] & _REGION_ENTRY_ORIGIN);
1376 __gmap_unshadow_r2t(sg, raddr, r2t);
1377 /* Clear entry and flush translation r1t -> r2t */
1378 gmap_idte_one(asce, raddr);
1379 r1t[i] = _REGION1_ENTRY_EMPTY;
1380 /* Free region 2 table */
1381 page = pfn_to_page(__pa(r2t) >> PAGE_SHIFT);
1382 list_del(&page->lru);
1383 __free_pages(page, CRST_ALLOC_ORDER);
1388 * gmap_unshadow - remove a shadow page table completely
1389 * @sg: pointer to the shadow guest address space structure
1391 * Called with sg->guest_table_lock
1393 static void gmap_unshadow(struct gmap *sg)
1395 unsigned long *table;
1397 BUG_ON(!gmap_is_shadow(sg));
1398 if (sg->removed)
1399 return;
1400 sg->removed = 1;
1401 gmap_call_notifier(sg, 0, -1UL);
1402 gmap_flush_tlb(sg);
1403 table = (unsigned long *)(sg->asce & _ASCE_ORIGIN);
1404 switch (sg->asce & _ASCE_TYPE_MASK) {
1405 case _ASCE_TYPE_REGION1:
1406 __gmap_unshadow_r1t(sg, 0, table);
1407 break;
1408 case _ASCE_TYPE_REGION2:
1409 __gmap_unshadow_r2t(sg, 0, table);
1410 break;
1411 case _ASCE_TYPE_REGION3:
1412 __gmap_unshadow_r3t(sg, 0, table);
1413 break;
1414 case _ASCE_TYPE_SEGMENT:
1415 __gmap_unshadow_sgt(sg, 0, table);
1416 break;
1421 * gmap_find_shadow - find a specific asce in the list of shadow tables
1422 * @parent: pointer to the parent gmap
1423 * @asce: ASCE for which the shadow table is created
1424 * @edat_level: edat level to be used for the shadow translation
1426 * Returns the pointer to a gmap if a shadow table with the given asce is
1427 * already available, ERR_PTR(-EAGAIN) if another one is just being created,
1428 * otherwise NULL
1430 static struct gmap *gmap_find_shadow(struct gmap *parent, unsigned long asce,
1431 int edat_level)
1433 struct gmap *sg;
1435 list_for_each_entry(sg, &parent->children, list) {
1436 if (sg->orig_asce != asce || sg->edat_level != edat_level ||
1437 sg->removed)
1438 continue;
1439 if (!sg->initialized)
1440 return ERR_PTR(-EAGAIN);
1441 atomic_inc(&sg->ref_count);
1442 return sg;
1444 return NULL;
1448 * gmap_shadow_valid - check if a shadow guest address space matches the
1449 * given properties and is still valid
1450 * @sg: pointer to the shadow guest address space structure
1451 * @asce: ASCE for which the shadow table is requested
1452 * @edat_level: edat level to be used for the shadow translation
1454 * Returns 1 if the gmap shadow is still valid and matches the given
1455 * properties, the caller can continue using it. Returns 0 otherwise, the
1456 * caller has to request a new shadow gmap in this case.
1459 int gmap_shadow_valid(struct gmap *sg, unsigned long asce, int edat_level)
1461 if (sg->removed)
1462 return 0;
1463 return sg->orig_asce == asce && sg->edat_level == edat_level;
1465 EXPORT_SYMBOL_GPL(gmap_shadow_valid);
1468 * gmap_shadow - create/find a shadow guest address space
1469 * @parent: pointer to the parent gmap
1470 * @asce: ASCE for which the shadow table is created
1471 * @edat_level: edat level to be used for the shadow translation
1473 * The pages of the top level page table referred by the asce parameter
1474 * will be set to read-only and marked in the PGSTEs of the kvm process.
1475 * The shadow table will be removed automatically on any change to the
1476 * PTE mapping for the source table.
1478 * Returns a guest address space structure, ERR_PTR(-ENOMEM) if out of memory,
1479 * ERR_PTR(-EAGAIN) if the caller has to retry and ERR_PTR(-EFAULT) if the
1480 * parent gmap table could not be protected.
1482 struct gmap *gmap_shadow(struct gmap *parent, unsigned long asce,
1483 int edat_level)
1485 struct gmap *sg, *new;
1486 unsigned long limit;
1487 int rc;
1489 BUG_ON(gmap_is_shadow(parent));
1490 spin_lock(&parent->shadow_lock);
1491 sg = gmap_find_shadow(parent, asce, edat_level);
1492 spin_unlock(&parent->shadow_lock);
1493 if (sg)
1494 return sg;
1495 /* Create a new shadow gmap */
1496 limit = -1UL >> (33 - (((asce & _ASCE_TYPE_MASK) >> 2) * 11));
1497 if (asce & _ASCE_REAL_SPACE)
1498 limit = -1UL;
1499 new = gmap_alloc(limit);
1500 if (!new)
1501 return ERR_PTR(-ENOMEM);
1502 new->mm = parent->mm;
1503 new->parent = gmap_get(parent);
1504 new->orig_asce = asce;
1505 new->edat_level = edat_level;
1506 new->initialized = false;
1507 spin_lock(&parent->shadow_lock);
1508 /* Recheck if another CPU created the same shadow */
1509 sg = gmap_find_shadow(parent, asce, edat_level);
1510 if (sg) {
1511 spin_unlock(&parent->shadow_lock);
1512 gmap_free(new);
1513 return sg;
1515 if (asce & _ASCE_REAL_SPACE) {
1516 /* only allow one real-space gmap shadow */
1517 list_for_each_entry(sg, &parent->children, list) {
1518 if (sg->orig_asce & _ASCE_REAL_SPACE) {
1519 spin_lock(&sg->guest_table_lock);
1520 gmap_unshadow(sg);
1521 spin_unlock(&sg->guest_table_lock);
1522 list_del(&sg->list);
1523 gmap_put(sg);
1524 break;
1528 atomic_set(&new->ref_count, 2);
1529 list_add(&new->list, &parent->children);
1530 if (asce & _ASCE_REAL_SPACE) {
1531 /* nothing to protect, return right away */
1532 new->initialized = true;
1533 spin_unlock(&parent->shadow_lock);
1534 return new;
1536 spin_unlock(&parent->shadow_lock);
1537 /* protect after insertion, so it will get properly invalidated */
1538 down_read(&parent->mm->mmap_sem);
1539 rc = gmap_protect_range(parent, asce & _ASCE_ORIGIN,
1540 ((asce & _ASCE_TABLE_LENGTH) + 1) * PAGE_SIZE,
1541 PROT_READ, PGSTE_VSIE_BIT);
1542 up_read(&parent->mm->mmap_sem);
1543 spin_lock(&parent->shadow_lock);
1544 new->initialized = true;
1545 if (rc) {
1546 list_del(&new->list);
1547 gmap_free(new);
1548 new = ERR_PTR(rc);
1550 spin_unlock(&parent->shadow_lock);
1551 return new;
1553 EXPORT_SYMBOL_GPL(gmap_shadow);
1556 * gmap_shadow_r2t - create an empty shadow region 2 table
1557 * @sg: pointer to the shadow guest address space structure
1558 * @saddr: faulting address in the shadow gmap
1559 * @r2t: parent gmap address of the region 2 table to get shadowed
1560 * @fake: r2t references contiguous guest memory block, not a r2t
1562 * The r2t parameter specifies the address of the source table. The
1563 * four pages of the source table are made read-only in the parent gmap
1564 * address space. A write to the source table area @r2t will automatically
1565 * remove the shadow r2 table and all of its decendents.
1567 * Returns 0 if successfully shadowed or already shadowed, -EAGAIN if the
1568 * shadow table structure is incomplete, -ENOMEM if out of memory and
1569 * -EFAULT if an address in the parent gmap could not be resolved.
1571 * Called with sg->mm->mmap_sem in read.
1573 int gmap_shadow_r2t(struct gmap *sg, unsigned long saddr, unsigned long r2t,
1574 int fake)
1576 unsigned long raddr, origin, offset, len;
1577 unsigned long *s_r2t, *table;
1578 struct page *page;
1579 int rc;
1581 BUG_ON(!gmap_is_shadow(sg));
1582 /* Allocate a shadow region second table */
1583 page = alloc_pages(GFP_KERNEL, CRST_ALLOC_ORDER);
1584 if (!page)
1585 return -ENOMEM;
1586 page->index = r2t & _REGION_ENTRY_ORIGIN;
1587 if (fake)
1588 page->index |= GMAP_SHADOW_FAKE_TABLE;
1589 s_r2t = (unsigned long *) page_to_phys(page);
1590 /* Install shadow region second table */
1591 spin_lock(&sg->guest_table_lock);
1592 table = gmap_table_walk(sg, saddr, 4); /* get region-1 pointer */
1593 if (!table) {
1594 rc = -EAGAIN; /* Race with unshadow */
1595 goto out_free;
1597 if (!(*table & _REGION_ENTRY_INVALID)) {
1598 rc = 0; /* Already established */
1599 goto out_free;
1600 } else if (*table & _REGION_ENTRY_ORIGIN) {
1601 rc = -EAGAIN; /* Race with shadow */
1602 goto out_free;
1604 crst_table_init(s_r2t, _REGION2_ENTRY_EMPTY);
1605 /* mark as invalid as long as the parent table is not protected */
1606 *table = (unsigned long) s_r2t | _REGION_ENTRY_LENGTH |
1607 _REGION_ENTRY_TYPE_R1 | _REGION_ENTRY_INVALID;
1608 if (sg->edat_level >= 1)
1609 *table |= (r2t & _REGION_ENTRY_PROTECT);
1610 list_add(&page->lru, &sg->crst_list);
1611 if (fake) {
1612 /* nothing to protect for fake tables */
1613 *table &= ~_REGION_ENTRY_INVALID;
1614 spin_unlock(&sg->guest_table_lock);
1615 return 0;
1617 spin_unlock(&sg->guest_table_lock);
1618 /* Make r2t read-only in parent gmap page table */
1619 raddr = (saddr & _REGION1_MASK) | _SHADOW_RMAP_REGION1;
1620 origin = r2t & _REGION_ENTRY_ORIGIN;
1621 offset = ((r2t & _REGION_ENTRY_OFFSET) >> 6) * PAGE_SIZE;
1622 len = ((r2t & _REGION_ENTRY_LENGTH) + 1) * PAGE_SIZE - offset;
1623 rc = gmap_protect_rmap(sg, raddr, origin + offset, len, PROT_READ);
1624 spin_lock(&sg->guest_table_lock);
1625 if (!rc) {
1626 table = gmap_table_walk(sg, saddr, 4);
1627 if (!table || (*table & _REGION_ENTRY_ORIGIN) !=
1628 (unsigned long) s_r2t)
1629 rc = -EAGAIN; /* Race with unshadow */
1630 else
1631 *table &= ~_REGION_ENTRY_INVALID;
1632 } else {
1633 gmap_unshadow_r2t(sg, raddr);
1635 spin_unlock(&sg->guest_table_lock);
1636 return rc;
1637 out_free:
1638 spin_unlock(&sg->guest_table_lock);
1639 __free_pages(page, CRST_ALLOC_ORDER);
1640 return rc;
1642 EXPORT_SYMBOL_GPL(gmap_shadow_r2t);
1645 * gmap_shadow_r3t - create a shadow region 3 table
1646 * @sg: pointer to the shadow guest address space structure
1647 * @saddr: faulting address in the shadow gmap
1648 * @r3t: parent gmap address of the region 3 table to get shadowed
1649 * @fake: r3t references contiguous guest memory block, not a r3t
1651 * Returns 0 if successfully shadowed or already shadowed, -EAGAIN if the
1652 * shadow table structure is incomplete, -ENOMEM if out of memory and
1653 * -EFAULT if an address in the parent gmap could not be resolved.
1655 * Called with sg->mm->mmap_sem in read.
1657 int gmap_shadow_r3t(struct gmap *sg, unsigned long saddr, unsigned long r3t,
1658 int fake)
1660 unsigned long raddr, origin, offset, len;
1661 unsigned long *s_r3t, *table;
1662 struct page *page;
1663 int rc;
1665 BUG_ON(!gmap_is_shadow(sg));
1666 /* Allocate a shadow region second table */
1667 page = alloc_pages(GFP_KERNEL, CRST_ALLOC_ORDER);
1668 if (!page)
1669 return -ENOMEM;
1670 page->index = r3t & _REGION_ENTRY_ORIGIN;
1671 if (fake)
1672 page->index |= GMAP_SHADOW_FAKE_TABLE;
1673 s_r3t = (unsigned long *) page_to_phys(page);
1674 /* Install shadow region second table */
1675 spin_lock(&sg->guest_table_lock);
1676 table = gmap_table_walk(sg, saddr, 3); /* get region-2 pointer */
1677 if (!table) {
1678 rc = -EAGAIN; /* Race with unshadow */
1679 goto out_free;
1681 if (!(*table & _REGION_ENTRY_INVALID)) {
1682 rc = 0; /* Already established */
1683 goto out_free;
1684 } else if (*table & _REGION_ENTRY_ORIGIN) {
1685 rc = -EAGAIN; /* Race with shadow */
1687 crst_table_init(s_r3t, _REGION3_ENTRY_EMPTY);
1688 /* mark as invalid as long as the parent table is not protected */
1689 *table = (unsigned long) s_r3t | _REGION_ENTRY_LENGTH |
1690 _REGION_ENTRY_TYPE_R2 | _REGION_ENTRY_INVALID;
1691 if (sg->edat_level >= 1)
1692 *table |= (r3t & _REGION_ENTRY_PROTECT);
1693 list_add(&page->lru, &sg->crst_list);
1694 if (fake) {
1695 /* nothing to protect for fake tables */
1696 *table &= ~_REGION_ENTRY_INVALID;
1697 spin_unlock(&sg->guest_table_lock);
1698 return 0;
1700 spin_unlock(&sg->guest_table_lock);
1701 /* Make r3t read-only in parent gmap page table */
1702 raddr = (saddr & _REGION2_MASK) | _SHADOW_RMAP_REGION2;
1703 origin = r3t & _REGION_ENTRY_ORIGIN;
1704 offset = ((r3t & _REGION_ENTRY_OFFSET) >> 6) * PAGE_SIZE;
1705 len = ((r3t & _REGION_ENTRY_LENGTH) + 1) * PAGE_SIZE - offset;
1706 rc = gmap_protect_rmap(sg, raddr, origin + offset, len, PROT_READ);
1707 spin_lock(&sg->guest_table_lock);
1708 if (!rc) {
1709 table = gmap_table_walk(sg, saddr, 3);
1710 if (!table || (*table & _REGION_ENTRY_ORIGIN) !=
1711 (unsigned long) s_r3t)
1712 rc = -EAGAIN; /* Race with unshadow */
1713 else
1714 *table &= ~_REGION_ENTRY_INVALID;
1715 } else {
1716 gmap_unshadow_r3t(sg, raddr);
1718 spin_unlock(&sg->guest_table_lock);
1719 return rc;
1720 out_free:
1721 spin_unlock(&sg->guest_table_lock);
1722 __free_pages(page, CRST_ALLOC_ORDER);
1723 return rc;
1725 EXPORT_SYMBOL_GPL(gmap_shadow_r3t);
1728 * gmap_shadow_sgt - create a shadow segment table
1729 * @sg: pointer to the shadow guest address space structure
1730 * @saddr: faulting address in the shadow gmap
1731 * @sgt: parent gmap address of the segment table to get shadowed
1732 * @fake: sgt references contiguous guest memory block, not a sgt
1734 * Returns: 0 if successfully shadowed or already shadowed, -EAGAIN if the
1735 * shadow table structure is incomplete, -ENOMEM if out of memory and
1736 * -EFAULT if an address in the parent gmap could not be resolved.
1738 * Called with sg->mm->mmap_sem in read.
1740 int gmap_shadow_sgt(struct gmap *sg, unsigned long saddr, unsigned long sgt,
1741 int fake)
1743 unsigned long raddr, origin, offset, len;
1744 unsigned long *s_sgt, *table;
1745 struct page *page;
1746 int rc;
1748 BUG_ON(!gmap_is_shadow(sg) || (sgt & _REGION3_ENTRY_LARGE));
1749 /* Allocate a shadow segment table */
1750 page = alloc_pages(GFP_KERNEL, CRST_ALLOC_ORDER);
1751 if (!page)
1752 return -ENOMEM;
1753 page->index = sgt & _REGION_ENTRY_ORIGIN;
1754 if (fake)
1755 page->index |= GMAP_SHADOW_FAKE_TABLE;
1756 s_sgt = (unsigned long *) page_to_phys(page);
1757 /* Install shadow region second table */
1758 spin_lock(&sg->guest_table_lock);
1759 table = gmap_table_walk(sg, saddr, 2); /* get region-3 pointer */
1760 if (!table) {
1761 rc = -EAGAIN; /* Race with unshadow */
1762 goto out_free;
1764 if (!(*table & _REGION_ENTRY_INVALID)) {
1765 rc = 0; /* Already established */
1766 goto out_free;
1767 } else if (*table & _REGION_ENTRY_ORIGIN) {
1768 rc = -EAGAIN; /* Race with shadow */
1769 goto out_free;
1771 crst_table_init(s_sgt, _SEGMENT_ENTRY_EMPTY);
1772 /* mark as invalid as long as the parent table is not protected */
1773 *table = (unsigned long) s_sgt | _REGION_ENTRY_LENGTH |
1774 _REGION_ENTRY_TYPE_R3 | _REGION_ENTRY_INVALID;
1775 if (sg->edat_level >= 1)
1776 *table |= sgt & _REGION_ENTRY_PROTECT;
1777 list_add(&page->lru, &sg->crst_list);
1778 if (fake) {
1779 /* nothing to protect for fake tables */
1780 *table &= ~_REGION_ENTRY_INVALID;
1781 spin_unlock(&sg->guest_table_lock);
1782 return 0;
1784 spin_unlock(&sg->guest_table_lock);
1785 /* Make sgt read-only in parent gmap page table */
1786 raddr = (saddr & _REGION3_MASK) | _SHADOW_RMAP_REGION3;
1787 origin = sgt & _REGION_ENTRY_ORIGIN;
1788 offset = ((sgt & _REGION_ENTRY_OFFSET) >> 6) * PAGE_SIZE;
1789 len = ((sgt & _REGION_ENTRY_LENGTH) + 1) * PAGE_SIZE - offset;
1790 rc = gmap_protect_rmap(sg, raddr, origin + offset, len, PROT_READ);
1791 spin_lock(&sg->guest_table_lock);
1792 if (!rc) {
1793 table = gmap_table_walk(sg, saddr, 2);
1794 if (!table || (*table & _REGION_ENTRY_ORIGIN) !=
1795 (unsigned long) s_sgt)
1796 rc = -EAGAIN; /* Race with unshadow */
1797 else
1798 *table &= ~_REGION_ENTRY_INVALID;
1799 } else {
1800 gmap_unshadow_sgt(sg, raddr);
1802 spin_unlock(&sg->guest_table_lock);
1803 return rc;
1804 out_free:
1805 spin_unlock(&sg->guest_table_lock);
1806 __free_pages(page, CRST_ALLOC_ORDER);
1807 return rc;
1809 EXPORT_SYMBOL_GPL(gmap_shadow_sgt);
1812 * gmap_shadow_lookup_pgtable - find a shadow page table
1813 * @sg: pointer to the shadow guest address space structure
1814 * @saddr: the address in the shadow aguest address space
1815 * @pgt: parent gmap address of the page table to get shadowed
1816 * @dat_protection: if the pgtable is marked as protected by dat
1817 * @fake: pgt references contiguous guest memory block, not a pgtable
1819 * Returns 0 if the shadow page table was found and -EAGAIN if the page
1820 * table was not found.
1822 * Called with sg->mm->mmap_sem in read.
1824 int gmap_shadow_pgt_lookup(struct gmap *sg, unsigned long saddr,
1825 unsigned long *pgt, int *dat_protection,
1826 int *fake)
1828 unsigned long *table;
1829 struct page *page;
1830 int rc;
1832 BUG_ON(!gmap_is_shadow(sg));
1833 spin_lock(&sg->guest_table_lock);
1834 table = gmap_table_walk(sg, saddr, 1); /* get segment pointer */
1835 if (table && !(*table & _SEGMENT_ENTRY_INVALID)) {
1836 /* Shadow page tables are full pages (pte+pgste) */
1837 page = pfn_to_page(*table >> PAGE_SHIFT);
1838 *pgt = page->index & ~GMAP_SHADOW_FAKE_TABLE;
1839 *dat_protection = !!(*table & _SEGMENT_ENTRY_PROTECT);
1840 *fake = !!(page->index & GMAP_SHADOW_FAKE_TABLE);
1841 rc = 0;
1842 } else {
1843 rc = -EAGAIN;
1845 spin_unlock(&sg->guest_table_lock);
1846 return rc;
1849 EXPORT_SYMBOL_GPL(gmap_shadow_pgt_lookup);
1852 * gmap_shadow_pgt - instantiate a shadow page table
1853 * @sg: pointer to the shadow guest address space structure
1854 * @saddr: faulting address in the shadow gmap
1855 * @pgt: parent gmap address of the page table to get shadowed
1856 * @fake: pgt references contiguous guest memory block, not a pgtable
1858 * Returns 0 if successfully shadowed or already shadowed, -EAGAIN if the
1859 * shadow table structure is incomplete, -ENOMEM if out of memory,
1860 * -EFAULT if an address in the parent gmap could not be resolved and
1862 * Called with gmap->mm->mmap_sem in read
1864 int gmap_shadow_pgt(struct gmap *sg, unsigned long saddr, unsigned long pgt,
1865 int fake)
1867 unsigned long raddr, origin;
1868 unsigned long *s_pgt, *table;
1869 struct page *page;
1870 int rc;
1872 BUG_ON(!gmap_is_shadow(sg) || (pgt & _SEGMENT_ENTRY_LARGE));
1873 /* Allocate a shadow page table */
1874 page = page_table_alloc_pgste(sg->mm);
1875 if (!page)
1876 return -ENOMEM;
1877 page->index = pgt & _SEGMENT_ENTRY_ORIGIN;
1878 if (fake)
1879 page->index |= GMAP_SHADOW_FAKE_TABLE;
1880 s_pgt = (unsigned long *) page_to_phys(page);
1881 /* Install shadow page table */
1882 spin_lock(&sg->guest_table_lock);
1883 table = gmap_table_walk(sg, saddr, 1); /* get segment pointer */
1884 if (!table) {
1885 rc = -EAGAIN; /* Race with unshadow */
1886 goto out_free;
1888 if (!(*table & _SEGMENT_ENTRY_INVALID)) {
1889 rc = 0; /* Already established */
1890 goto out_free;
1891 } else if (*table & _SEGMENT_ENTRY_ORIGIN) {
1892 rc = -EAGAIN; /* Race with shadow */
1893 goto out_free;
1895 /* mark as invalid as long as the parent table is not protected */
1896 *table = (unsigned long) s_pgt | _SEGMENT_ENTRY |
1897 (pgt & _SEGMENT_ENTRY_PROTECT) | _SEGMENT_ENTRY_INVALID;
1898 list_add(&page->lru, &sg->pt_list);
1899 if (fake) {
1900 /* nothing to protect for fake tables */
1901 *table &= ~_SEGMENT_ENTRY_INVALID;
1902 spin_unlock(&sg->guest_table_lock);
1903 return 0;
1905 spin_unlock(&sg->guest_table_lock);
1906 /* Make pgt read-only in parent gmap page table (not the pgste) */
1907 raddr = (saddr & _SEGMENT_MASK) | _SHADOW_RMAP_SEGMENT;
1908 origin = pgt & _SEGMENT_ENTRY_ORIGIN & PAGE_MASK;
1909 rc = gmap_protect_rmap(sg, raddr, origin, PAGE_SIZE, PROT_READ);
1910 spin_lock(&sg->guest_table_lock);
1911 if (!rc) {
1912 table = gmap_table_walk(sg, saddr, 1);
1913 if (!table || (*table & _SEGMENT_ENTRY_ORIGIN) !=
1914 (unsigned long) s_pgt)
1915 rc = -EAGAIN; /* Race with unshadow */
1916 else
1917 *table &= ~_SEGMENT_ENTRY_INVALID;
1918 } else {
1919 gmap_unshadow_pgt(sg, raddr);
1921 spin_unlock(&sg->guest_table_lock);
1922 return rc;
1923 out_free:
1924 spin_unlock(&sg->guest_table_lock);
1925 page_table_free_pgste(page);
1926 return rc;
1929 EXPORT_SYMBOL_GPL(gmap_shadow_pgt);
1932 * gmap_shadow_page - create a shadow page mapping
1933 * @sg: pointer to the shadow guest address space structure
1934 * @saddr: faulting address in the shadow gmap
1935 * @pte: pte in parent gmap address space to get shadowed
1937 * Returns 0 if successfully shadowed or already shadowed, -EAGAIN if the
1938 * shadow table structure is incomplete, -ENOMEM if out of memory and
1939 * -EFAULT if an address in the parent gmap could not be resolved.
1941 * Called with sg->mm->mmap_sem in read.
1943 int gmap_shadow_page(struct gmap *sg, unsigned long saddr, pte_t pte)
1945 struct gmap *parent;
1946 struct gmap_rmap *rmap;
1947 unsigned long vmaddr, paddr;
1948 spinlock_t *ptl;
1949 pte_t *sptep, *tptep;
1950 int prot;
1951 int rc;
1953 BUG_ON(!gmap_is_shadow(sg));
1954 parent = sg->parent;
1955 prot = (pte_val(pte) & _PAGE_PROTECT) ? PROT_READ : PROT_WRITE;
1957 rmap = kzalloc(sizeof(*rmap), GFP_KERNEL);
1958 if (!rmap)
1959 return -ENOMEM;
1960 rmap->raddr = (saddr & PAGE_MASK) | _SHADOW_RMAP_PGTABLE;
1962 while (1) {
1963 paddr = pte_val(pte) & PAGE_MASK;
1964 vmaddr = __gmap_translate(parent, paddr);
1965 if (IS_ERR_VALUE(vmaddr)) {
1966 rc = vmaddr;
1967 break;
1969 rc = radix_tree_preload(GFP_KERNEL);
1970 if (rc)
1971 break;
1972 rc = -EAGAIN;
1973 sptep = gmap_pte_op_walk(parent, paddr, &ptl);
1974 if (sptep) {
1975 spin_lock(&sg->guest_table_lock);
1976 /* Get page table pointer */
1977 tptep = (pte_t *) gmap_table_walk(sg, saddr, 0);
1978 if (!tptep) {
1979 spin_unlock(&sg->guest_table_lock);
1980 gmap_pte_op_end(ptl);
1981 radix_tree_preload_end();
1982 break;
1984 rc = ptep_shadow_pte(sg->mm, saddr, sptep, tptep, pte);
1985 if (rc > 0) {
1986 /* Success and a new mapping */
1987 gmap_insert_rmap(sg, vmaddr, rmap);
1988 rmap = NULL;
1989 rc = 0;
1991 gmap_pte_op_end(ptl);
1992 spin_unlock(&sg->guest_table_lock);
1994 radix_tree_preload_end();
1995 if (!rc)
1996 break;
1997 rc = gmap_pte_op_fixup(parent, paddr, vmaddr, prot);
1998 if (rc)
1999 break;
2001 kfree(rmap);
2002 return rc;
2004 EXPORT_SYMBOL_GPL(gmap_shadow_page);
2007 * gmap_shadow_notify - handle notifications for shadow gmap
2009 * Called with sg->parent->shadow_lock.
2011 static void gmap_shadow_notify(struct gmap *sg, unsigned long vmaddr,
2012 unsigned long gaddr, pte_t *pte)
2014 struct gmap_rmap *rmap, *rnext, *head;
2015 unsigned long start, end, bits, raddr;
2017 BUG_ON(!gmap_is_shadow(sg));
2019 spin_lock(&sg->guest_table_lock);
2020 if (sg->removed) {
2021 spin_unlock(&sg->guest_table_lock);
2022 return;
2024 /* Check for top level table */
2025 start = sg->orig_asce & _ASCE_ORIGIN;
2026 end = start + ((sg->orig_asce & _ASCE_TABLE_LENGTH) + 1) * PAGE_SIZE;
2027 if (!(sg->orig_asce & _ASCE_REAL_SPACE) && gaddr >= start &&
2028 gaddr < end) {
2029 /* The complete shadow table has to go */
2030 gmap_unshadow(sg);
2031 spin_unlock(&sg->guest_table_lock);
2032 list_del(&sg->list);
2033 gmap_put(sg);
2034 return;
2036 /* Remove the page table tree from on specific entry */
2037 head = radix_tree_delete(&sg->host_to_rmap, vmaddr >> PAGE_SHIFT);
2038 gmap_for_each_rmap_safe(rmap, rnext, head) {
2039 bits = rmap->raddr & _SHADOW_RMAP_MASK;
2040 raddr = rmap->raddr ^ bits;
2041 switch (bits) {
2042 case _SHADOW_RMAP_REGION1:
2043 gmap_unshadow_r2t(sg, raddr);
2044 break;
2045 case _SHADOW_RMAP_REGION2:
2046 gmap_unshadow_r3t(sg, raddr);
2047 break;
2048 case _SHADOW_RMAP_REGION3:
2049 gmap_unshadow_sgt(sg, raddr);
2050 break;
2051 case _SHADOW_RMAP_SEGMENT:
2052 gmap_unshadow_pgt(sg, raddr);
2053 break;
2054 case _SHADOW_RMAP_PGTABLE:
2055 gmap_unshadow_page(sg, raddr);
2056 break;
2058 kfree(rmap);
2060 spin_unlock(&sg->guest_table_lock);
2064 * ptep_notify - call all invalidation callbacks for a specific pte.
2065 * @mm: pointer to the process mm_struct
2066 * @addr: virtual address in the process address space
2067 * @pte: pointer to the page table entry
2068 * @bits: bits from the pgste that caused the notify call
2070 * This function is assumed to be called with the page table lock held
2071 * for the pte to notify.
2073 void ptep_notify(struct mm_struct *mm, unsigned long vmaddr,
2074 pte_t *pte, unsigned long bits)
2076 unsigned long offset, gaddr = 0;
2077 unsigned long *table;
2078 struct gmap *gmap, *sg, *next;
2080 offset = ((unsigned long) pte) & (255 * sizeof(pte_t));
2081 offset = offset * (PAGE_SIZE / sizeof(pte_t));
2082 rcu_read_lock();
2083 list_for_each_entry_rcu(gmap, &mm->context.gmap_list, list) {
2084 spin_lock(&gmap->guest_table_lock);
2085 table = radix_tree_lookup(&gmap->host_to_guest,
2086 vmaddr >> PMD_SHIFT);
2087 if (table)
2088 gaddr = __gmap_segment_gaddr(table) + offset;
2089 spin_unlock(&gmap->guest_table_lock);
2090 if (!table)
2091 continue;
2093 if (!list_empty(&gmap->children) && (bits & PGSTE_VSIE_BIT)) {
2094 spin_lock(&gmap->shadow_lock);
2095 list_for_each_entry_safe(sg, next,
2096 &gmap->children, list)
2097 gmap_shadow_notify(sg, vmaddr, gaddr, pte);
2098 spin_unlock(&gmap->shadow_lock);
2100 if (bits & PGSTE_IN_BIT)
2101 gmap_call_notifier(gmap, gaddr, gaddr + PAGE_SIZE - 1);
2103 rcu_read_unlock();
2105 EXPORT_SYMBOL_GPL(ptep_notify);
2107 static inline void thp_split_mm(struct mm_struct *mm)
2109 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
2110 struct vm_area_struct *vma;
2111 unsigned long addr;
2113 for (vma = mm->mmap; vma != NULL; vma = vma->vm_next) {
2114 for (addr = vma->vm_start;
2115 addr < vma->vm_end;
2116 addr += PAGE_SIZE)
2117 follow_page(vma, addr, FOLL_SPLIT);
2118 vma->vm_flags &= ~VM_HUGEPAGE;
2119 vma->vm_flags |= VM_NOHUGEPAGE;
2121 mm->def_flags |= VM_NOHUGEPAGE;
2122 #endif
2126 * Remove all empty zero pages from the mapping for lazy refaulting
2127 * - This must be called after mm->context.has_pgste is set, to avoid
2128 * future creation of zero pages
2129 * - This must be called after THP was enabled
2131 static int __zap_zero_pages(pmd_t *pmd, unsigned long start,
2132 unsigned long end, struct mm_walk *walk)
2134 unsigned long addr;
2136 for (addr = start; addr != end; addr += PAGE_SIZE) {
2137 pte_t *ptep;
2138 spinlock_t *ptl;
2140 ptep = pte_offset_map_lock(walk->mm, pmd, addr, &ptl);
2141 if (is_zero_pfn(pte_pfn(*ptep)))
2142 ptep_xchg_direct(walk->mm, addr, ptep, __pte(_PAGE_INVALID));
2143 pte_unmap_unlock(ptep, ptl);
2145 return 0;
2148 static inline void zap_zero_pages(struct mm_struct *mm)
2150 struct mm_walk walk = { .pmd_entry = __zap_zero_pages };
2152 walk.mm = mm;
2153 walk_page_range(0, TASK_SIZE, &walk);
2157 * switch on pgstes for its userspace process (for kvm)
2159 int s390_enable_sie(void)
2161 struct mm_struct *mm = current->mm;
2163 /* Do we have pgstes? if yes, we are done */
2164 if (mm_has_pgste(mm))
2165 return 0;
2166 /* Fail if the page tables are 2K */
2167 if (!mm_alloc_pgste(mm))
2168 return -EINVAL;
2169 down_write(&mm->mmap_sem);
2170 mm->context.has_pgste = 1;
2171 /* split thp mappings and disable thp for future mappings */
2172 thp_split_mm(mm);
2173 zap_zero_pages(mm);
2174 up_write(&mm->mmap_sem);
2175 return 0;
2177 EXPORT_SYMBOL_GPL(s390_enable_sie);
2180 * Enable storage key handling from now on and initialize the storage
2181 * keys with the default key.
2183 static int __s390_enable_skey(pte_t *pte, unsigned long addr,
2184 unsigned long next, struct mm_walk *walk)
2186 /* Clear storage key */
2187 ptep_zap_key(walk->mm, addr, pte);
2188 return 0;
2191 int s390_enable_skey(void)
2193 struct mm_walk walk = { .pte_entry = __s390_enable_skey };
2194 struct mm_struct *mm = current->mm;
2195 struct vm_area_struct *vma;
2196 int rc = 0;
2198 down_write(&mm->mmap_sem);
2199 if (mm_use_skey(mm))
2200 goto out_up;
2202 mm->context.use_skey = 1;
2203 for (vma = mm->mmap; vma; vma = vma->vm_next) {
2204 if (ksm_madvise(vma, vma->vm_start, vma->vm_end,
2205 MADV_UNMERGEABLE, &vma->vm_flags)) {
2206 mm->context.use_skey = 0;
2207 rc = -ENOMEM;
2208 goto out_up;
2211 mm->def_flags &= ~VM_MERGEABLE;
2213 walk.mm = mm;
2214 walk_page_range(0, TASK_SIZE, &walk);
2216 out_up:
2217 up_write(&mm->mmap_sem);
2218 return rc;
2220 EXPORT_SYMBOL_GPL(s390_enable_skey);
2223 * Reset CMMA state, make all pages stable again.
2225 static int __s390_reset_cmma(pte_t *pte, unsigned long addr,
2226 unsigned long next, struct mm_walk *walk)
2228 ptep_zap_unused(walk->mm, addr, pte, 1);
2229 return 0;
2232 void s390_reset_cmma(struct mm_struct *mm)
2234 struct mm_walk walk = { .pte_entry = __s390_reset_cmma };
2236 down_write(&mm->mmap_sem);
2237 walk.mm = mm;
2238 walk_page_range(0, TASK_SIZE, &walk);
2239 up_write(&mm->mmap_sem);
2241 EXPORT_SYMBOL_GPL(s390_reset_cmma);