2 * KVM guest address space mapping code
4 * Copyright IBM Corp. 2007, 2016
5 * Author(s): Martin Schwidefsky <schwidefsky@de.ibm.com>
8 #include <linux/kernel.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>
23 #define GMAP_SHADOW_FAKE_TABLE 1ULL
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)
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;
53 atype = _ASCE_TYPE_REGION1;
54 etype = _REGION1_ENTRY_EMPTY;
56 gmap = kzalloc(sizeof(struct gmap), GFP_KERNEL);
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);
72 list_add(&page->lru, &gmap->crst_list);
73 table = (unsigned long *) page_to_phys(page);
74 crst_table_init(table, etype);
76 gmap->asce = atype | _ASCE_TABLE_LENGTH |
77 _ASCE_USER_BITS | __pa(table);
78 gmap->asce_end = limit;
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)
97 unsigned long gmap_asce;
99 gmap = gmap_alloc(limit);
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;
109 WRITE_ONCE(mm->context.gmap_asce, gmap_asce);
110 spin_unlock(&mm->context.lock);
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);
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];
131 /* A radix tree is freed by deleting all of its entries */
135 radix_tree_for_each_slot(slot, root, &iter, index) {
136 indices[nr] = iter.index;
140 for (i = 0; i < nr; i++) {
142 radix_tree_delete(root, index);
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];
156 /* A radix tree is freed by deleting all of its entries */
160 radix_tree_for_each_slot(slot, root, &iter, index) {
161 indices[nr] = iter.index;
165 for (i = 0; i < nr; i++) {
167 head = radix_tree_delete(root, index);
168 gmap_for_each_rmap_safe(rmap, rnext, head)
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);
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);
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)
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) {
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))
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;
260 WRITE_ONCE(gmap->mm->context.gmap_asce, gmap_asce);
261 spin_unlock(&gmap->mm->context.lock);
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)
308 /* since we dont free the gmap table until gmap_free we can unlock */
309 page = alloc_pages(GFP_KERNEL, CRST_ALLOC_ORDER);
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);
322 spin_unlock(&gmap->guest_table_lock);
324 __free_pages(page, CRST_ALLOC_ORDER);
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)
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;
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);
362 flush = (*entry != _SEGMENT_ENTRY_EMPTY);
363 *entry = _SEGMENT_ENTRY_EMPTY;
365 spin_unlock(&gmap->guest_table_lock);
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,
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)
398 BUG_ON(gmap_is_shadow(gmap));
399 if ((to | len) & (PMD_SIZE - 1))
401 if (len == 0 || to + len < to)
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);
410 gmap_flush_tlb(gmap);
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)
430 BUG_ON(gmap_is_shadow(gmap));
431 if ((from | to | len) & (PMD_SIZE - 1))
433 if (len == 0 || from + len < from || to + len < to ||
434 from + len - 1 > TASK_SIZE_MAX || to + len - 1 > gmap->asce_end)
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))
448 up_write(&gmap->mm->mmap_sem);
450 gmap_flush_tlb(gmap);
453 gmap_unmap_segment(gmap, to, len);
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)
495 down_read(&gmap->mm->mmap_sem);
496 rc = __gmap_translate(gmap, gaddr);
497 up_read(&gmap->mm->mmap_sem);
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)
515 list_for_each_entry_rcu(gmap, &mm->context.gmap_list, list) {
516 flush = __gmap_unlink_by_vmaddr(gmap, vmaddr);
518 gmap_flush_tlb(gmap);
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;
545 BUG_ON(gmap_is_shadow(gmap));
546 /* Create higher level tables in the gmap page 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))
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))
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))
570 table = (unsigned long *)(*table & _REGION_ENTRY_ORIGIN);
572 table += (gaddr & _SEGMENT_INDEX) >> _SEGMENT_SHIFT;
573 /* Walk the parent mm page table */
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 */
584 pmd = pmd_offset(pud, vmaddr);
585 VM_BUG_ON(pmd_none(*pmd));
586 /* large pmds cannot yet be handled */
589 /* Link gmap segment table entry location to page table. */
590 rc = radix_tree_preload(GFP_KERNEL);
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);
599 *table = pmd_val(*pmd);
602 spin_unlock(&gmap->guest_table_lock);
604 radix_tree_preload_end();
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;
624 down_read(&gmap->mm->mmap_sem);
628 vmaddr = __gmap_translate(gmap, gaddr);
629 if (IS_ERR_VALUE(vmaddr)) {
633 if (fixup_user_fault(current, gmap->mm, vmaddr, fault_flags,
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.
645 rc = __gmap_link(gmap, gaddr, vmaddr);
647 up_read(&gmap->mm->mmap_sem);
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;
661 /* Find the vm address for the guest address */
662 vmaddr = (unsigned long) radix_tree_lookup(&gmap->guest_to_host,
665 vmaddr |= gaddr & ~PMD_MASK;
666 /* Get pointer to the page table entry */
667 ptep = get_locked_pte(gmap->mm, vmaddr, &ptl);
669 ptep_zap_unused(gmap->mm, vmaddr, ptep, 0);
670 pte_unmap_unlock(ptep, ptl);
674 EXPORT_SYMBOL_GPL(__gmap_zap);
676 void gmap_discard(struct gmap *gmap, unsigned long from, unsigned long to)
678 unsigned long gaddr, vmaddr, size;
679 struct vm_area_struct *vma;
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,
690 vmaddr |= gaddr & ~PMD_MASK;
691 /* Find vma in the parent mm */
692 vma = find_vma(gmap->mm, vmaddr);
695 size = min(to - gaddr, PMD_SIZE - (gaddr & ~PMD_MASK));
696 zap_page_range(vma, vmaddr, size);
698 up_read(&gmap->mm->mmap_sem);
700 EXPORT_SYMBOL_GPL(gmap_discard);
702 static LIST_HEAD(gmap_notifier_list);
703 static DEFINE_SPINLOCK(gmap_notifier_lock);
706 * gmap_register_pte_notifier - register a pte invalidation callback
707 * @nb: pointer to the gmap notifier block
709 void gmap_register_pte_notifier(struct gmap_notifier *nb)
711 spin_lock(&gmap_notifier_lock);
712 list_add_rcu(&nb->list, &gmap_notifier_list);
713 spin_unlock(&gmap_notifier_lock);
715 EXPORT_SYMBOL_GPL(gmap_register_pte_notifier);
718 * gmap_unregister_pte_notifier - remove a pte invalidation callback
719 * @nb: pointer to the gmap notifier block
721 void gmap_unregister_pte_notifier(struct gmap_notifier *nb)
723 spin_lock(&gmap_notifier_lock);
724 list_del_rcu(&nb->list);
725 spin_unlock(&gmap_notifier_lock);
728 EXPORT_SYMBOL_GPL(gmap_unregister_pte_notifier);
731 * gmap_call_notifier - call all registered invalidation callbacks
732 * @gmap: pointer to guest mapping meta data structure
733 * @start: start virtual address in the guest address space
734 * @end: end virtual address in the guest address space
736 static void gmap_call_notifier(struct gmap *gmap, unsigned long start,
739 struct gmap_notifier *nb;
741 list_for_each_entry(nb, &gmap_notifier_list, list)
742 nb->notifier_call(gmap, start, end);
746 * gmap_table_walk - walk the gmap page tables
747 * @gmap: pointer to guest mapping meta data structure
748 * @gaddr: virtual address in the guest address space
749 * @level: page table level to stop at
751 * Returns a table entry pointer for the given guest address and @level
752 * @level=0 : returns a pointer to a page table table entry (or NULL)
753 * @level=1 : returns a pointer to a segment table entry (or NULL)
754 * @level=2 : returns a pointer to a region-3 table entry (or NULL)
755 * @level=3 : returns a pointer to a region-2 table entry (or NULL)
756 * @level=4 : returns a pointer to a region-1 table entry (or NULL)
758 * Returns NULL if the gmap page tables could not be walked to the
761 * Note: Can also be called for shadow gmaps.
763 static inline unsigned long *gmap_table_walk(struct gmap *gmap,
764 unsigned long gaddr, int level)
766 const int asce_type = gmap->asce & _ASCE_TYPE_MASK;
767 unsigned long *table;
769 if ((gmap->asce & _ASCE_TYPE_MASK) + 4 < (level * 4))
771 if (gmap_is_shadow(gmap) && gmap->removed)
774 if (asce_type != _ASCE_TYPE_REGION1 &&
775 gaddr & (-1UL << (31 + (asce_type >> 2) * 11)))
779 switch (gmap->asce & _ASCE_TYPE_MASK) {
780 case _ASCE_TYPE_REGION1:
781 table += (gaddr & _REGION1_INDEX) >> _REGION1_SHIFT;
784 if (*table & _REGION_ENTRY_INVALID)
786 table = (unsigned long *)(*table & _REGION_ENTRY_ORIGIN);
788 case _ASCE_TYPE_REGION2:
789 table += (gaddr & _REGION2_INDEX) >> _REGION2_SHIFT;
792 if (*table & _REGION_ENTRY_INVALID)
794 table = (unsigned long *)(*table & _REGION_ENTRY_ORIGIN);
796 case _ASCE_TYPE_REGION3:
797 table += (gaddr & _REGION3_INDEX) >> _REGION3_SHIFT;
800 if (*table & _REGION_ENTRY_INVALID)
802 table = (unsigned long *)(*table & _REGION_ENTRY_ORIGIN);
804 case _ASCE_TYPE_SEGMENT:
805 table += (gaddr & _SEGMENT_INDEX) >> _SEGMENT_SHIFT;
808 if (*table & _REGION_ENTRY_INVALID)
810 table = (unsigned long *)(*table & _SEGMENT_ENTRY_ORIGIN);
811 table += (gaddr & _PAGE_INDEX) >> _PAGE_SHIFT;
817 * gmap_pte_op_walk - walk the gmap page table, get the page table lock
818 * and return the pte pointer
819 * @gmap: pointer to guest mapping meta data structure
820 * @gaddr: virtual address in the guest address space
821 * @ptl: pointer to the spinlock pointer
823 * Returns a pointer to the locked pte for a guest address, or NULL
825 * Note: Can also be called for shadow gmaps.
827 static pte_t *gmap_pte_op_walk(struct gmap *gmap, unsigned long gaddr,
830 unsigned long *table;
832 if (gmap_is_shadow(gmap))
833 spin_lock(&gmap->guest_table_lock);
834 /* Walk the gmap page table, lock and get pte pointer */
835 table = gmap_table_walk(gmap, gaddr, 1); /* get segment pointer */
836 if (!table || *table & _SEGMENT_ENTRY_INVALID) {
837 if (gmap_is_shadow(gmap))
838 spin_unlock(&gmap->guest_table_lock);
841 if (gmap_is_shadow(gmap)) {
842 *ptl = &gmap->guest_table_lock;
843 return pte_offset_map((pmd_t *) table, gaddr);
845 return pte_alloc_map_lock(gmap->mm, (pmd_t *) table, gaddr, ptl);
849 * gmap_pte_op_fixup - force a page in and connect the gmap page table
850 * @gmap: pointer to guest mapping meta data structure
851 * @gaddr: virtual address in the guest address space
852 * @vmaddr: address in the host process address space
853 * @prot: indicates access rights: PROT_NONE, PROT_READ or PROT_WRITE
855 * Returns 0 if the caller can retry __gmap_translate (might fail again),
856 * -ENOMEM if out of memory and -EFAULT if anything goes wrong while fixing
857 * up or connecting the gmap page table.
859 static int gmap_pte_op_fixup(struct gmap *gmap, unsigned long gaddr,
860 unsigned long vmaddr, int prot)
862 struct mm_struct *mm = gmap->mm;
863 unsigned int fault_flags;
864 bool unlocked = false;
866 BUG_ON(gmap_is_shadow(gmap));
867 fault_flags = (prot == PROT_WRITE) ? FAULT_FLAG_WRITE : 0;
868 if (fixup_user_fault(current, mm, vmaddr, fault_flags, &unlocked))
871 /* lost mmap_sem, caller has to retry __gmap_translate */
873 /* Connect the page tables */
874 return __gmap_link(gmap, gaddr, vmaddr);
878 * gmap_pte_op_end - release the page table lock
879 * @ptl: pointer to the spinlock pointer
881 static void gmap_pte_op_end(spinlock_t *ptl)
887 * gmap_protect_range - remove access rights to memory and set pgste bits
888 * @gmap: pointer to guest mapping meta data structure
889 * @gaddr: virtual address in the guest address space
891 * @prot: indicates access rights: PROT_NONE, PROT_READ or PROT_WRITE
892 * @bits: pgste notification bits to set
894 * Returns 0 if successfully protected, -ENOMEM if out of memory and
895 * -EFAULT if gaddr is invalid (or mapping for shadows is missing).
897 * Called with sg->mm->mmap_sem in read.
899 * Note: Can also be called for shadow gmaps.
901 static int gmap_protect_range(struct gmap *gmap, unsigned long gaddr,
902 unsigned long len, int prot, unsigned long bits)
904 unsigned long vmaddr;
911 ptep = gmap_pte_op_walk(gmap, gaddr, &ptl);
913 rc = ptep_force_prot(gmap->mm, gaddr, ptep, prot, bits);
914 gmap_pte_op_end(ptl);
917 vmaddr = __gmap_translate(gmap, gaddr);
918 if (IS_ERR_VALUE(vmaddr))
920 rc = gmap_pte_op_fixup(gmap, gaddr, vmaddr, prot);
932 * gmap_mprotect_notify - change access rights for a range of ptes and
933 * call the notifier if any pte changes again
934 * @gmap: pointer to guest mapping meta data structure
935 * @gaddr: virtual address in the guest address space
937 * @prot: indicates access rights: PROT_NONE, PROT_READ or PROT_WRITE
939 * Returns 0 if for each page in the given range a gmap mapping exists,
940 * the new access rights could be set and the notifier could be armed.
941 * If the gmap mapping is missing for one or more pages -EFAULT is
942 * returned. If no memory could be allocated -ENOMEM is returned.
943 * This function establishes missing page table entries.
945 int gmap_mprotect_notify(struct gmap *gmap, unsigned long gaddr,
946 unsigned long len, int prot)
950 if ((gaddr & ~PAGE_MASK) || (len & ~PAGE_MASK) || gmap_is_shadow(gmap))
952 if (!MACHINE_HAS_ESOP && prot == PROT_READ)
954 down_read(&gmap->mm->mmap_sem);
955 rc = gmap_protect_range(gmap, gaddr, len, prot, PGSTE_IN_BIT);
956 up_read(&gmap->mm->mmap_sem);
959 EXPORT_SYMBOL_GPL(gmap_mprotect_notify);
962 * gmap_read_table - get an unsigned long value from a guest page table using
963 * absolute addressing, without marking the page referenced.
964 * @gmap: pointer to guest mapping meta data structure
965 * @gaddr: virtual address in the guest address space
966 * @val: pointer to the unsigned long value to return
968 * Returns 0 if the value was read, -ENOMEM if out of memory and -EFAULT
969 * if reading using the virtual address failed.
971 * Called with gmap->mm->mmap_sem in read.
973 int gmap_read_table(struct gmap *gmap, unsigned long gaddr, unsigned long *val)
975 unsigned long address, vmaddr;
982 ptep = gmap_pte_op_walk(gmap, gaddr, &ptl);
985 if (pte_present(pte) && (pte_val(pte) & _PAGE_READ)) {
986 address = pte_val(pte) & PAGE_MASK;
987 address += gaddr & ~PAGE_MASK;
988 *val = *(unsigned long *) address;
989 pte_val(*ptep) |= _PAGE_YOUNG;
990 /* Do *NOT* clear the _PAGE_INVALID bit! */
993 gmap_pte_op_end(ptl);
997 vmaddr = __gmap_translate(gmap, gaddr);
998 if (IS_ERR_VALUE(vmaddr)) {
1002 rc = gmap_pte_op_fixup(gmap, gaddr, vmaddr, PROT_READ);
1008 EXPORT_SYMBOL_GPL(gmap_read_table);
1011 * gmap_insert_rmap - add a rmap to the host_to_rmap radix tree
1012 * @sg: pointer to the shadow guest address space structure
1013 * @vmaddr: vm address associated with the rmap
1014 * @rmap: pointer to the rmap structure
1016 * Called with the sg->guest_table_lock
1018 static inline void gmap_insert_rmap(struct gmap *sg, unsigned long vmaddr,
1019 struct gmap_rmap *rmap)
1023 BUG_ON(!gmap_is_shadow(sg));
1024 slot = radix_tree_lookup_slot(&sg->host_to_rmap, vmaddr >> PAGE_SHIFT);
1026 rmap->next = radix_tree_deref_slot_protected(slot,
1027 &sg->guest_table_lock);
1028 radix_tree_replace_slot(&sg->host_to_rmap, slot, rmap);
1031 radix_tree_insert(&sg->host_to_rmap, vmaddr >> PAGE_SHIFT,
1037 * gmap_protect_rmap - modify access rights to memory and create an rmap
1038 * @sg: pointer to the shadow guest address space structure
1039 * @raddr: rmap address in the shadow gmap
1040 * @paddr: address in the parent guest address space
1041 * @len: length of the memory area to protect
1042 * @prot: indicates access rights: none, read-only or read-write
1044 * Returns 0 if successfully protected and the rmap was created, -ENOMEM
1045 * if out of memory and -EFAULT if paddr is invalid.
1047 static int gmap_protect_rmap(struct gmap *sg, unsigned long raddr,
1048 unsigned long paddr, unsigned long len, int prot)
1050 struct gmap *parent;
1051 struct gmap_rmap *rmap;
1052 unsigned long vmaddr;
1057 BUG_ON(!gmap_is_shadow(sg));
1058 parent = sg->parent;
1060 vmaddr = __gmap_translate(parent, paddr);
1061 if (IS_ERR_VALUE(vmaddr))
1063 rmap = kzalloc(sizeof(*rmap), GFP_KERNEL);
1066 rmap->raddr = raddr;
1067 rc = radix_tree_preload(GFP_KERNEL);
1073 ptep = gmap_pte_op_walk(parent, paddr, &ptl);
1075 spin_lock(&sg->guest_table_lock);
1076 rc = ptep_force_prot(parent->mm, paddr, ptep, prot,
1079 gmap_insert_rmap(sg, vmaddr, rmap);
1080 spin_unlock(&sg->guest_table_lock);
1081 gmap_pte_op_end(ptl);
1083 radix_tree_preload_end();
1086 rc = gmap_pte_op_fixup(parent, paddr, vmaddr, prot);
1097 #define _SHADOW_RMAP_MASK 0x7
1098 #define _SHADOW_RMAP_REGION1 0x5
1099 #define _SHADOW_RMAP_REGION2 0x4
1100 #define _SHADOW_RMAP_REGION3 0x3
1101 #define _SHADOW_RMAP_SEGMENT 0x2
1102 #define _SHADOW_RMAP_PGTABLE 0x1
1105 * gmap_idte_one - invalidate a single region or segment table entry
1106 * @asce: region or segment table *origin* + table-type bits
1107 * @vaddr: virtual address to identify the table entry to flush
1109 * The invalid bit of a single region or segment table entry is set
1110 * and the associated TLB entries depending on the entry are flushed.
1111 * The table-type of the @asce identifies the portion of the @vaddr
1112 * that is used as the invalidation index.
1114 static inline void gmap_idte_one(unsigned long asce, unsigned long vaddr)
1117 " .insn rrf,0xb98e0000,%0,%1,0,0"
1118 : : "a" (asce), "a" (vaddr) : "cc", "memory");
1122 * gmap_unshadow_page - remove a page from a shadow page table
1123 * @sg: pointer to the shadow guest address space structure
1124 * @raddr: rmap address in the shadow guest address space
1126 * Called with the sg->guest_table_lock
1128 static void gmap_unshadow_page(struct gmap *sg, unsigned long raddr)
1130 unsigned long *table;
1132 BUG_ON(!gmap_is_shadow(sg));
1133 table = gmap_table_walk(sg, raddr, 0); /* get page table pointer */
1134 if (!table || *table & _PAGE_INVALID)
1136 gmap_call_notifier(sg, raddr, raddr + _PAGE_SIZE - 1);
1137 ptep_unshadow_pte(sg->mm, raddr, (pte_t *) table);
1141 * __gmap_unshadow_pgt - remove all entries from a shadow page table
1142 * @sg: pointer to the shadow guest address space structure
1143 * @raddr: rmap address in the shadow guest address space
1144 * @pgt: pointer to the start of a shadow page table
1146 * Called with the sg->guest_table_lock
1148 static void __gmap_unshadow_pgt(struct gmap *sg, unsigned long raddr,
1153 BUG_ON(!gmap_is_shadow(sg));
1154 for (i = 0; i < _PAGE_ENTRIES; i++, raddr += _PAGE_SIZE)
1155 pgt[i] = _PAGE_INVALID;
1159 * gmap_unshadow_pgt - remove a shadow page table from a segment entry
1160 * @sg: pointer to the shadow guest address space structure
1161 * @raddr: address in the shadow guest address space
1163 * Called with the sg->guest_table_lock
1165 static void gmap_unshadow_pgt(struct gmap *sg, unsigned long raddr)
1167 unsigned long sto, *ste, *pgt;
1170 BUG_ON(!gmap_is_shadow(sg));
1171 ste = gmap_table_walk(sg, raddr, 1); /* get segment pointer */
1172 if (!ste || !(*ste & _SEGMENT_ENTRY_ORIGIN))
1174 gmap_call_notifier(sg, raddr, raddr + _SEGMENT_SIZE - 1);
1175 sto = (unsigned long) (ste - ((raddr & _SEGMENT_INDEX) >> _SEGMENT_SHIFT));
1176 gmap_idte_one(sto | _ASCE_TYPE_SEGMENT, raddr);
1177 pgt = (unsigned long *)(*ste & _SEGMENT_ENTRY_ORIGIN);
1178 *ste = _SEGMENT_ENTRY_EMPTY;
1179 __gmap_unshadow_pgt(sg, raddr, pgt);
1180 /* Free page table */
1181 page = pfn_to_page(__pa(pgt) >> PAGE_SHIFT);
1182 list_del(&page->lru);
1183 page_table_free_pgste(page);
1187 * __gmap_unshadow_sgt - remove all entries from a shadow segment table
1188 * @sg: pointer to the shadow guest address space structure
1189 * @raddr: rmap address in the shadow guest address space
1190 * @sgt: pointer to the start of a shadow segment table
1192 * Called with the sg->guest_table_lock
1194 static void __gmap_unshadow_sgt(struct gmap *sg, unsigned long raddr,
1197 unsigned long asce, *pgt;
1201 BUG_ON(!gmap_is_shadow(sg));
1202 asce = (unsigned long) sgt | _ASCE_TYPE_SEGMENT;
1203 for (i = 0; i < _CRST_ENTRIES; i++, raddr += _SEGMENT_SIZE) {
1204 if (!(sgt[i] & _SEGMENT_ENTRY_ORIGIN))
1206 pgt = (unsigned long *)(sgt[i] & _REGION_ENTRY_ORIGIN);
1207 sgt[i] = _SEGMENT_ENTRY_EMPTY;
1208 __gmap_unshadow_pgt(sg, raddr, pgt);
1209 /* Free page table */
1210 page = pfn_to_page(__pa(pgt) >> PAGE_SHIFT);
1211 list_del(&page->lru);
1212 page_table_free_pgste(page);
1217 * gmap_unshadow_sgt - remove a shadow segment table from a region-3 entry
1218 * @sg: pointer to the shadow guest address space structure
1219 * @raddr: rmap address in the shadow guest address space
1221 * Called with the shadow->guest_table_lock
1223 static void gmap_unshadow_sgt(struct gmap *sg, unsigned long raddr)
1225 unsigned long r3o, *r3e, *sgt;
1228 BUG_ON(!gmap_is_shadow(sg));
1229 r3e = gmap_table_walk(sg, raddr, 2); /* get region-3 pointer */
1230 if (!r3e || !(*r3e & _REGION_ENTRY_ORIGIN))
1232 gmap_call_notifier(sg, raddr, raddr + _REGION3_SIZE - 1);
1233 r3o = (unsigned long) (r3e - ((raddr & _REGION3_INDEX) >> _REGION3_SHIFT));
1234 gmap_idte_one(r3o | _ASCE_TYPE_REGION3, raddr);
1235 sgt = (unsigned long *)(*r3e & _REGION_ENTRY_ORIGIN);
1236 *r3e = _REGION3_ENTRY_EMPTY;
1237 __gmap_unshadow_sgt(sg, raddr, sgt);
1238 /* Free segment table */
1239 page = pfn_to_page(__pa(sgt) >> PAGE_SHIFT);
1240 list_del(&page->lru);
1241 __free_pages(page, CRST_ALLOC_ORDER);
1245 * __gmap_unshadow_r3t - remove all entries from a shadow region-3 table
1246 * @sg: pointer to the shadow guest address space structure
1247 * @raddr: address in the shadow guest address space
1248 * @r3t: pointer to the start of a shadow region-3 table
1250 * Called with the sg->guest_table_lock
1252 static void __gmap_unshadow_r3t(struct gmap *sg, unsigned long raddr,
1255 unsigned long asce, *sgt;
1259 BUG_ON(!gmap_is_shadow(sg));
1260 asce = (unsigned long) r3t | _ASCE_TYPE_REGION3;
1261 for (i = 0; i < _CRST_ENTRIES; i++, raddr += _REGION3_SIZE) {
1262 if (!(r3t[i] & _REGION_ENTRY_ORIGIN))
1264 sgt = (unsigned long *)(r3t[i] & _REGION_ENTRY_ORIGIN);
1265 r3t[i] = _REGION3_ENTRY_EMPTY;
1266 __gmap_unshadow_sgt(sg, raddr, sgt);
1267 /* Free segment table */
1268 page = pfn_to_page(__pa(sgt) >> PAGE_SHIFT);
1269 list_del(&page->lru);
1270 __free_pages(page, CRST_ALLOC_ORDER);
1275 * gmap_unshadow_r3t - remove a shadow region-3 table from a region-2 entry
1276 * @sg: pointer to the shadow guest address space structure
1277 * @raddr: rmap address in the shadow guest address space
1279 * Called with the sg->guest_table_lock
1281 static void gmap_unshadow_r3t(struct gmap *sg, unsigned long raddr)
1283 unsigned long r2o, *r2e, *r3t;
1286 BUG_ON(!gmap_is_shadow(sg));
1287 r2e = gmap_table_walk(sg, raddr, 3); /* get region-2 pointer */
1288 if (!r2e || !(*r2e & _REGION_ENTRY_ORIGIN))
1290 gmap_call_notifier(sg, raddr, raddr + _REGION2_SIZE - 1);
1291 r2o = (unsigned long) (r2e - ((raddr & _REGION2_INDEX) >> _REGION2_SHIFT));
1292 gmap_idte_one(r2o | _ASCE_TYPE_REGION2, raddr);
1293 r3t = (unsigned long *)(*r2e & _REGION_ENTRY_ORIGIN);
1294 *r2e = _REGION2_ENTRY_EMPTY;
1295 __gmap_unshadow_r3t(sg, raddr, r3t);
1296 /* Free region 3 table */
1297 page = pfn_to_page(__pa(r3t) >> PAGE_SHIFT);
1298 list_del(&page->lru);
1299 __free_pages(page, CRST_ALLOC_ORDER);
1303 * __gmap_unshadow_r2t - remove all entries from a shadow region-2 table
1304 * @sg: pointer to the shadow guest address space structure
1305 * @raddr: rmap address in the shadow guest address space
1306 * @r2t: pointer to the start of a shadow region-2 table
1308 * Called with the sg->guest_table_lock
1310 static void __gmap_unshadow_r2t(struct gmap *sg, unsigned long raddr,
1313 unsigned long asce, *r3t;
1317 BUG_ON(!gmap_is_shadow(sg));
1318 asce = (unsigned long) r2t | _ASCE_TYPE_REGION2;
1319 for (i = 0; i < _CRST_ENTRIES; i++, raddr += _REGION2_SIZE) {
1320 if (!(r2t[i] & _REGION_ENTRY_ORIGIN))
1322 r3t = (unsigned long *)(r2t[i] & _REGION_ENTRY_ORIGIN);
1323 r2t[i] = _REGION2_ENTRY_EMPTY;
1324 __gmap_unshadow_r3t(sg, raddr, r3t);
1325 /* Free region 3 table */
1326 page = pfn_to_page(__pa(r3t) >> PAGE_SHIFT);
1327 list_del(&page->lru);
1328 __free_pages(page, CRST_ALLOC_ORDER);
1333 * gmap_unshadow_r2t - remove a shadow region-2 table from a region-1 entry
1334 * @sg: pointer to the shadow guest address space structure
1335 * @raddr: rmap address in the shadow guest address space
1337 * Called with the sg->guest_table_lock
1339 static void gmap_unshadow_r2t(struct gmap *sg, unsigned long raddr)
1341 unsigned long r1o, *r1e, *r2t;
1344 BUG_ON(!gmap_is_shadow(sg));
1345 r1e = gmap_table_walk(sg, raddr, 4); /* get region-1 pointer */
1346 if (!r1e || !(*r1e & _REGION_ENTRY_ORIGIN))
1348 gmap_call_notifier(sg, raddr, raddr + _REGION1_SIZE - 1);
1349 r1o = (unsigned long) (r1e - ((raddr & _REGION1_INDEX) >> _REGION1_SHIFT));
1350 gmap_idte_one(r1o | _ASCE_TYPE_REGION1, raddr);
1351 r2t = (unsigned long *)(*r1e & _REGION_ENTRY_ORIGIN);
1352 *r1e = _REGION1_ENTRY_EMPTY;
1353 __gmap_unshadow_r2t(sg, raddr, r2t);
1354 /* Free region 2 table */
1355 page = pfn_to_page(__pa(r2t) >> PAGE_SHIFT);
1356 list_del(&page->lru);
1357 __free_pages(page, CRST_ALLOC_ORDER);
1361 * __gmap_unshadow_r1t - remove all entries from a shadow region-1 table
1362 * @sg: pointer to the shadow guest address space structure
1363 * @raddr: rmap address in the shadow guest address space
1364 * @r1t: pointer to the start of a shadow region-1 table
1366 * Called with the shadow->guest_table_lock
1368 static void __gmap_unshadow_r1t(struct gmap *sg, unsigned long raddr,
1371 unsigned long asce, *r2t;
1375 BUG_ON(!gmap_is_shadow(sg));
1376 asce = (unsigned long) r1t | _ASCE_TYPE_REGION1;
1377 for (i = 0; i < _CRST_ENTRIES; i++, raddr += _REGION1_SIZE) {
1378 if (!(r1t[i] & _REGION_ENTRY_ORIGIN))
1380 r2t = (unsigned long *)(r1t[i] & _REGION_ENTRY_ORIGIN);
1381 __gmap_unshadow_r2t(sg, raddr, r2t);
1382 /* Clear entry and flush translation r1t -> r2t */
1383 gmap_idte_one(asce, raddr);
1384 r1t[i] = _REGION1_ENTRY_EMPTY;
1385 /* Free region 2 table */
1386 page = pfn_to_page(__pa(r2t) >> PAGE_SHIFT);
1387 list_del(&page->lru);
1388 __free_pages(page, CRST_ALLOC_ORDER);
1393 * gmap_unshadow - remove a shadow page table completely
1394 * @sg: pointer to the shadow guest address space structure
1396 * Called with sg->guest_table_lock
1398 static void gmap_unshadow(struct gmap *sg)
1400 unsigned long *table;
1402 BUG_ON(!gmap_is_shadow(sg));
1406 gmap_call_notifier(sg, 0, -1UL);
1408 table = (unsigned long *)(sg->asce & _ASCE_ORIGIN);
1409 switch (sg->asce & _ASCE_TYPE_MASK) {
1410 case _ASCE_TYPE_REGION1:
1411 __gmap_unshadow_r1t(sg, 0, table);
1413 case _ASCE_TYPE_REGION2:
1414 __gmap_unshadow_r2t(sg, 0, table);
1416 case _ASCE_TYPE_REGION3:
1417 __gmap_unshadow_r3t(sg, 0, table);
1419 case _ASCE_TYPE_SEGMENT:
1420 __gmap_unshadow_sgt(sg, 0, table);
1426 * gmap_find_shadow - find a specific asce in the list of shadow tables
1427 * @parent: pointer to the parent gmap
1428 * @asce: ASCE for which the shadow table is created
1429 * @edat_level: edat level to be used for the shadow translation
1431 * Returns the pointer to a gmap if a shadow table with the given asce is
1432 * already available, ERR_PTR(-EAGAIN) if another one is just being created,
1435 static struct gmap *gmap_find_shadow(struct gmap *parent, unsigned long asce,
1440 list_for_each_entry(sg, &parent->children, list) {
1441 if (sg->orig_asce != asce || sg->edat_level != edat_level ||
1444 if (!sg->initialized)
1445 return ERR_PTR(-EAGAIN);
1446 atomic_inc(&sg->ref_count);
1453 * gmap_shadow_valid - check if a shadow guest address space matches the
1454 * given properties and is still valid
1455 * @sg: pointer to the shadow guest address space structure
1456 * @asce: ASCE for which the shadow table is requested
1457 * @edat_level: edat level to be used for the shadow translation
1459 * Returns 1 if the gmap shadow is still valid and matches the given
1460 * properties, the caller can continue using it. Returns 0 otherwise, the
1461 * caller has to request a new shadow gmap in this case.
1464 int gmap_shadow_valid(struct gmap *sg, unsigned long asce, int edat_level)
1468 return sg->orig_asce == asce && sg->edat_level == edat_level;
1470 EXPORT_SYMBOL_GPL(gmap_shadow_valid);
1473 * gmap_shadow - create/find a shadow guest address space
1474 * @parent: pointer to the parent gmap
1475 * @asce: ASCE for which the shadow table is created
1476 * @edat_level: edat level to be used for the shadow translation
1478 * The pages of the top level page table referred by the asce parameter
1479 * will be set to read-only and marked in the PGSTEs of the kvm process.
1480 * The shadow table will be removed automatically on any change to the
1481 * PTE mapping for the source table.
1483 * Returns a guest address space structure, ERR_PTR(-ENOMEM) if out of memory,
1484 * ERR_PTR(-EAGAIN) if the caller has to retry and ERR_PTR(-EFAULT) if the
1485 * parent gmap table could not be protected.
1487 struct gmap *gmap_shadow(struct gmap *parent, unsigned long asce,
1490 struct gmap *sg, *new;
1491 unsigned long limit;
1494 BUG_ON(gmap_is_shadow(parent));
1495 spin_lock(&parent->shadow_lock);
1496 sg = gmap_find_shadow(parent, asce, edat_level);
1497 spin_unlock(&parent->shadow_lock);
1500 /* Create a new shadow gmap */
1501 limit = -1UL >> (33 - (((asce & _ASCE_TYPE_MASK) >> 2) * 11));
1502 if (asce & _ASCE_REAL_SPACE)
1504 new = gmap_alloc(limit);
1506 return ERR_PTR(-ENOMEM);
1507 new->mm = parent->mm;
1508 new->parent = gmap_get(parent);
1509 new->orig_asce = asce;
1510 new->edat_level = edat_level;
1511 new->initialized = false;
1512 spin_lock(&parent->shadow_lock);
1513 /* Recheck if another CPU created the same shadow */
1514 sg = gmap_find_shadow(parent, asce, edat_level);
1516 spin_unlock(&parent->shadow_lock);
1520 if (asce & _ASCE_REAL_SPACE) {
1521 /* only allow one real-space gmap shadow */
1522 list_for_each_entry(sg, &parent->children, list) {
1523 if (sg->orig_asce & _ASCE_REAL_SPACE) {
1524 spin_lock(&sg->guest_table_lock);
1526 spin_unlock(&sg->guest_table_lock);
1527 list_del(&sg->list);
1533 atomic_set(&new->ref_count, 2);
1534 list_add(&new->list, &parent->children);
1535 if (asce & _ASCE_REAL_SPACE) {
1536 /* nothing to protect, return right away */
1537 new->initialized = true;
1538 spin_unlock(&parent->shadow_lock);
1541 spin_unlock(&parent->shadow_lock);
1542 /* protect after insertion, so it will get properly invalidated */
1543 down_read(&parent->mm->mmap_sem);
1544 rc = gmap_protect_range(parent, asce & _ASCE_ORIGIN,
1545 ((asce & _ASCE_TABLE_LENGTH) + 1) * PAGE_SIZE,
1546 PROT_READ, PGSTE_VSIE_BIT);
1547 up_read(&parent->mm->mmap_sem);
1548 spin_lock(&parent->shadow_lock);
1549 new->initialized = true;
1551 list_del(&new->list);
1555 spin_unlock(&parent->shadow_lock);
1558 EXPORT_SYMBOL_GPL(gmap_shadow);
1561 * gmap_shadow_r2t - create an empty shadow region 2 table
1562 * @sg: pointer to the shadow guest address space structure
1563 * @saddr: faulting address in the shadow gmap
1564 * @r2t: parent gmap address of the region 2 table to get shadowed
1565 * @fake: r2t references contiguous guest memory block, not a r2t
1567 * The r2t parameter specifies the address of the source table. The
1568 * four pages of the source table are made read-only in the parent gmap
1569 * address space. A write to the source table area @r2t will automatically
1570 * remove the shadow r2 table and all of its decendents.
1572 * Returns 0 if successfully shadowed or already shadowed, -EAGAIN if the
1573 * shadow table structure is incomplete, -ENOMEM if out of memory and
1574 * -EFAULT if an address in the parent gmap could not be resolved.
1576 * Called with sg->mm->mmap_sem in read.
1578 int gmap_shadow_r2t(struct gmap *sg, unsigned long saddr, unsigned long r2t,
1581 unsigned long raddr, origin, offset, len;
1582 unsigned long *s_r2t, *table;
1586 BUG_ON(!gmap_is_shadow(sg));
1587 /* Allocate a shadow region second table */
1588 page = alloc_pages(GFP_KERNEL, CRST_ALLOC_ORDER);
1591 page->index = r2t & _REGION_ENTRY_ORIGIN;
1593 page->index |= GMAP_SHADOW_FAKE_TABLE;
1594 s_r2t = (unsigned long *) page_to_phys(page);
1595 /* Install shadow region second table */
1596 spin_lock(&sg->guest_table_lock);
1597 table = gmap_table_walk(sg, saddr, 4); /* get region-1 pointer */
1599 rc = -EAGAIN; /* Race with unshadow */
1602 if (!(*table & _REGION_ENTRY_INVALID)) {
1603 rc = 0; /* Already established */
1605 } else if (*table & _REGION_ENTRY_ORIGIN) {
1606 rc = -EAGAIN; /* Race with shadow */
1609 crst_table_init(s_r2t, _REGION2_ENTRY_EMPTY);
1610 /* mark as invalid as long as the parent table is not protected */
1611 *table = (unsigned long) s_r2t | _REGION_ENTRY_LENGTH |
1612 _REGION_ENTRY_TYPE_R1 | _REGION_ENTRY_INVALID;
1613 if (sg->edat_level >= 1)
1614 *table |= (r2t & _REGION_ENTRY_PROTECT);
1615 list_add(&page->lru, &sg->crst_list);
1617 /* nothing to protect for fake tables */
1618 *table &= ~_REGION_ENTRY_INVALID;
1619 spin_unlock(&sg->guest_table_lock);
1622 spin_unlock(&sg->guest_table_lock);
1623 /* Make r2t read-only in parent gmap page table */
1624 raddr = (saddr & _REGION1_MASK) | _SHADOW_RMAP_REGION1;
1625 origin = r2t & _REGION_ENTRY_ORIGIN;
1626 offset = ((r2t & _REGION_ENTRY_OFFSET) >> 6) * PAGE_SIZE;
1627 len = ((r2t & _REGION_ENTRY_LENGTH) + 1) * PAGE_SIZE - offset;
1628 rc = gmap_protect_rmap(sg, raddr, origin + offset, len, PROT_READ);
1629 spin_lock(&sg->guest_table_lock);
1631 table = gmap_table_walk(sg, saddr, 4);
1632 if (!table || (*table & _REGION_ENTRY_ORIGIN) !=
1633 (unsigned long) s_r2t)
1634 rc = -EAGAIN; /* Race with unshadow */
1636 *table &= ~_REGION_ENTRY_INVALID;
1638 gmap_unshadow_r2t(sg, raddr);
1640 spin_unlock(&sg->guest_table_lock);
1643 spin_unlock(&sg->guest_table_lock);
1644 __free_pages(page, CRST_ALLOC_ORDER);
1647 EXPORT_SYMBOL_GPL(gmap_shadow_r2t);
1650 * gmap_shadow_r3t - create a shadow region 3 table
1651 * @sg: pointer to the shadow guest address space structure
1652 * @saddr: faulting address in the shadow gmap
1653 * @r3t: parent gmap address of the region 3 table to get shadowed
1654 * @fake: r3t references contiguous guest memory block, not a r3t
1656 * Returns 0 if successfully shadowed or already shadowed, -EAGAIN if the
1657 * shadow table structure is incomplete, -ENOMEM if out of memory and
1658 * -EFAULT if an address in the parent gmap could not be resolved.
1660 * Called with sg->mm->mmap_sem in read.
1662 int gmap_shadow_r3t(struct gmap *sg, unsigned long saddr, unsigned long r3t,
1665 unsigned long raddr, origin, offset, len;
1666 unsigned long *s_r3t, *table;
1670 BUG_ON(!gmap_is_shadow(sg));
1671 /* Allocate a shadow region second table */
1672 page = alloc_pages(GFP_KERNEL, CRST_ALLOC_ORDER);
1675 page->index = r3t & _REGION_ENTRY_ORIGIN;
1677 page->index |= GMAP_SHADOW_FAKE_TABLE;
1678 s_r3t = (unsigned long *) page_to_phys(page);
1679 /* Install shadow region second table */
1680 spin_lock(&sg->guest_table_lock);
1681 table = gmap_table_walk(sg, saddr, 3); /* get region-2 pointer */
1683 rc = -EAGAIN; /* Race with unshadow */
1686 if (!(*table & _REGION_ENTRY_INVALID)) {
1687 rc = 0; /* Already established */
1689 } else if (*table & _REGION_ENTRY_ORIGIN) {
1690 rc = -EAGAIN; /* Race with shadow */
1693 crst_table_init(s_r3t, _REGION3_ENTRY_EMPTY);
1694 /* mark as invalid as long as the parent table is not protected */
1695 *table = (unsigned long) s_r3t | _REGION_ENTRY_LENGTH |
1696 _REGION_ENTRY_TYPE_R2 | _REGION_ENTRY_INVALID;
1697 if (sg->edat_level >= 1)
1698 *table |= (r3t & _REGION_ENTRY_PROTECT);
1699 list_add(&page->lru, &sg->crst_list);
1701 /* nothing to protect for fake tables */
1702 *table &= ~_REGION_ENTRY_INVALID;
1703 spin_unlock(&sg->guest_table_lock);
1706 spin_unlock(&sg->guest_table_lock);
1707 /* Make r3t read-only in parent gmap page table */
1708 raddr = (saddr & _REGION2_MASK) | _SHADOW_RMAP_REGION2;
1709 origin = r3t & _REGION_ENTRY_ORIGIN;
1710 offset = ((r3t & _REGION_ENTRY_OFFSET) >> 6) * PAGE_SIZE;
1711 len = ((r3t & _REGION_ENTRY_LENGTH) + 1) * PAGE_SIZE - offset;
1712 rc = gmap_protect_rmap(sg, raddr, origin + offset, len, PROT_READ);
1713 spin_lock(&sg->guest_table_lock);
1715 table = gmap_table_walk(sg, saddr, 3);
1716 if (!table || (*table & _REGION_ENTRY_ORIGIN) !=
1717 (unsigned long) s_r3t)
1718 rc = -EAGAIN; /* Race with unshadow */
1720 *table &= ~_REGION_ENTRY_INVALID;
1722 gmap_unshadow_r3t(sg, raddr);
1724 spin_unlock(&sg->guest_table_lock);
1727 spin_unlock(&sg->guest_table_lock);
1728 __free_pages(page, CRST_ALLOC_ORDER);
1731 EXPORT_SYMBOL_GPL(gmap_shadow_r3t);
1734 * gmap_shadow_sgt - create a shadow segment table
1735 * @sg: pointer to the shadow guest address space structure
1736 * @saddr: faulting address in the shadow gmap
1737 * @sgt: parent gmap address of the segment table to get shadowed
1738 * @fake: sgt references contiguous guest memory block, not a sgt
1740 * Returns: 0 if successfully shadowed or already shadowed, -EAGAIN if the
1741 * shadow table structure is incomplete, -ENOMEM if out of memory and
1742 * -EFAULT if an address in the parent gmap could not be resolved.
1744 * Called with sg->mm->mmap_sem in read.
1746 int gmap_shadow_sgt(struct gmap *sg, unsigned long saddr, unsigned long sgt,
1749 unsigned long raddr, origin, offset, len;
1750 unsigned long *s_sgt, *table;
1754 BUG_ON(!gmap_is_shadow(sg) || (sgt & _REGION3_ENTRY_LARGE));
1755 /* Allocate a shadow segment table */
1756 page = alloc_pages(GFP_KERNEL, CRST_ALLOC_ORDER);
1759 page->index = sgt & _REGION_ENTRY_ORIGIN;
1761 page->index |= GMAP_SHADOW_FAKE_TABLE;
1762 s_sgt = (unsigned long *) page_to_phys(page);
1763 /* Install shadow region second table */
1764 spin_lock(&sg->guest_table_lock);
1765 table = gmap_table_walk(sg, saddr, 2); /* get region-3 pointer */
1767 rc = -EAGAIN; /* Race with unshadow */
1770 if (!(*table & _REGION_ENTRY_INVALID)) {
1771 rc = 0; /* Already established */
1773 } else if (*table & _REGION_ENTRY_ORIGIN) {
1774 rc = -EAGAIN; /* Race with shadow */
1777 crst_table_init(s_sgt, _SEGMENT_ENTRY_EMPTY);
1778 /* mark as invalid as long as the parent table is not protected */
1779 *table = (unsigned long) s_sgt | _REGION_ENTRY_LENGTH |
1780 _REGION_ENTRY_TYPE_R3 | _REGION_ENTRY_INVALID;
1781 if (sg->edat_level >= 1)
1782 *table |= sgt & _REGION_ENTRY_PROTECT;
1783 list_add(&page->lru, &sg->crst_list);
1785 /* nothing to protect for fake tables */
1786 *table &= ~_REGION_ENTRY_INVALID;
1787 spin_unlock(&sg->guest_table_lock);
1790 spin_unlock(&sg->guest_table_lock);
1791 /* Make sgt read-only in parent gmap page table */
1792 raddr = (saddr & _REGION3_MASK) | _SHADOW_RMAP_REGION3;
1793 origin = sgt & _REGION_ENTRY_ORIGIN;
1794 offset = ((sgt & _REGION_ENTRY_OFFSET) >> 6) * PAGE_SIZE;
1795 len = ((sgt & _REGION_ENTRY_LENGTH) + 1) * PAGE_SIZE - offset;
1796 rc = gmap_protect_rmap(sg, raddr, origin + offset, len, PROT_READ);
1797 spin_lock(&sg->guest_table_lock);
1799 table = gmap_table_walk(sg, saddr, 2);
1800 if (!table || (*table & _REGION_ENTRY_ORIGIN) !=
1801 (unsigned long) s_sgt)
1802 rc = -EAGAIN; /* Race with unshadow */
1804 *table &= ~_REGION_ENTRY_INVALID;
1806 gmap_unshadow_sgt(sg, raddr);
1808 spin_unlock(&sg->guest_table_lock);
1811 spin_unlock(&sg->guest_table_lock);
1812 __free_pages(page, CRST_ALLOC_ORDER);
1815 EXPORT_SYMBOL_GPL(gmap_shadow_sgt);
1818 * gmap_shadow_lookup_pgtable - find a shadow page table
1819 * @sg: pointer to the shadow guest address space structure
1820 * @saddr: the address in the shadow aguest address space
1821 * @pgt: parent gmap address of the page table to get shadowed
1822 * @dat_protection: if the pgtable is marked as protected by dat
1823 * @fake: pgt references contiguous guest memory block, not a pgtable
1825 * Returns 0 if the shadow page table was found and -EAGAIN if the page
1826 * table was not found.
1828 * Called with sg->mm->mmap_sem in read.
1830 int gmap_shadow_pgt_lookup(struct gmap *sg, unsigned long saddr,
1831 unsigned long *pgt, int *dat_protection,
1834 unsigned long *table;
1838 BUG_ON(!gmap_is_shadow(sg));
1839 spin_lock(&sg->guest_table_lock);
1840 table = gmap_table_walk(sg, saddr, 1); /* get segment pointer */
1841 if (table && !(*table & _SEGMENT_ENTRY_INVALID)) {
1842 /* Shadow page tables are full pages (pte+pgste) */
1843 page = pfn_to_page(*table >> PAGE_SHIFT);
1844 *pgt = page->index & ~GMAP_SHADOW_FAKE_TABLE;
1845 *dat_protection = !!(*table & _SEGMENT_ENTRY_PROTECT);
1846 *fake = !!(page->index & GMAP_SHADOW_FAKE_TABLE);
1851 spin_unlock(&sg->guest_table_lock);
1855 EXPORT_SYMBOL_GPL(gmap_shadow_pgt_lookup);
1858 * gmap_shadow_pgt - instantiate a shadow page table
1859 * @sg: pointer to the shadow guest address space structure
1860 * @saddr: faulting address in the shadow gmap
1861 * @pgt: parent gmap address of the page table to get shadowed
1862 * @fake: pgt references contiguous guest memory block, not a pgtable
1864 * Returns 0 if successfully shadowed or already shadowed, -EAGAIN if the
1865 * shadow table structure is incomplete, -ENOMEM if out of memory,
1866 * -EFAULT if an address in the parent gmap could not be resolved and
1868 * Called with gmap->mm->mmap_sem in read
1870 int gmap_shadow_pgt(struct gmap *sg, unsigned long saddr, unsigned long pgt,
1873 unsigned long raddr, origin;
1874 unsigned long *s_pgt, *table;
1878 BUG_ON(!gmap_is_shadow(sg) || (pgt & _SEGMENT_ENTRY_LARGE));
1879 /* Allocate a shadow page table */
1880 page = page_table_alloc_pgste(sg->mm);
1883 page->index = pgt & _SEGMENT_ENTRY_ORIGIN;
1885 page->index |= GMAP_SHADOW_FAKE_TABLE;
1886 s_pgt = (unsigned long *) page_to_phys(page);
1887 /* Install shadow page table */
1888 spin_lock(&sg->guest_table_lock);
1889 table = gmap_table_walk(sg, saddr, 1); /* get segment pointer */
1891 rc = -EAGAIN; /* Race with unshadow */
1894 if (!(*table & _SEGMENT_ENTRY_INVALID)) {
1895 rc = 0; /* Already established */
1897 } else if (*table & _SEGMENT_ENTRY_ORIGIN) {
1898 rc = -EAGAIN; /* Race with shadow */
1901 /* mark as invalid as long as the parent table is not protected */
1902 *table = (unsigned long) s_pgt | _SEGMENT_ENTRY |
1903 (pgt & _SEGMENT_ENTRY_PROTECT) | _SEGMENT_ENTRY_INVALID;
1904 list_add(&page->lru, &sg->pt_list);
1906 /* nothing to protect for fake tables */
1907 *table &= ~_SEGMENT_ENTRY_INVALID;
1908 spin_unlock(&sg->guest_table_lock);
1911 spin_unlock(&sg->guest_table_lock);
1912 /* Make pgt read-only in parent gmap page table (not the pgste) */
1913 raddr = (saddr & _SEGMENT_MASK) | _SHADOW_RMAP_SEGMENT;
1914 origin = pgt & _SEGMENT_ENTRY_ORIGIN & PAGE_MASK;
1915 rc = gmap_protect_rmap(sg, raddr, origin, PAGE_SIZE, PROT_READ);
1916 spin_lock(&sg->guest_table_lock);
1918 table = gmap_table_walk(sg, saddr, 1);
1919 if (!table || (*table & _SEGMENT_ENTRY_ORIGIN) !=
1920 (unsigned long) s_pgt)
1921 rc = -EAGAIN; /* Race with unshadow */
1923 *table &= ~_SEGMENT_ENTRY_INVALID;
1925 gmap_unshadow_pgt(sg, raddr);
1927 spin_unlock(&sg->guest_table_lock);
1930 spin_unlock(&sg->guest_table_lock);
1931 page_table_free_pgste(page);
1935 EXPORT_SYMBOL_GPL(gmap_shadow_pgt);
1938 * gmap_shadow_page - create a shadow page mapping
1939 * @sg: pointer to the shadow guest address space structure
1940 * @saddr: faulting address in the shadow gmap
1941 * @pte: pte in parent gmap address space to get shadowed
1943 * Returns 0 if successfully shadowed or already shadowed, -EAGAIN if the
1944 * shadow table structure is incomplete, -ENOMEM if out of memory and
1945 * -EFAULT if an address in the parent gmap could not be resolved.
1947 * Called with sg->mm->mmap_sem in read.
1949 int gmap_shadow_page(struct gmap *sg, unsigned long saddr, pte_t pte)
1951 struct gmap *parent;
1952 struct gmap_rmap *rmap;
1953 unsigned long vmaddr, paddr;
1955 pte_t *sptep, *tptep;
1959 BUG_ON(!gmap_is_shadow(sg));
1960 parent = sg->parent;
1961 prot = (pte_val(pte) & _PAGE_PROTECT) ? PROT_READ : PROT_WRITE;
1963 rmap = kzalloc(sizeof(*rmap), GFP_KERNEL);
1966 rmap->raddr = (saddr & PAGE_MASK) | _SHADOW_RMAP_PGTABLE;
1969 paddr = pte_val(pte) & PAGE_MASK;
1970 vmaddr = __gmap_translate(parent, paddr);
1971 if (IS_ERR_VALUE(vmaddr)) {
1975 rc = radix_tree_preload(GFP_KERNEL);
1979 sptep = gmap_pte_op_walk(parent, paddr, &ptl);
1981 spin_lock(&sg->guest_table_lock);
1982 /* Get page table pointer */
1983 tptep = (pte_t *) gmap_table_walk(sg, saddr, 0);
1985 spin_unlock(&sg->guest_table_lock);
1986 gmap_pte_op_end(ptl);
1987 radix_tree_preload_end();
1990 rc = ptep_shadow_pte(sg->mm, saddr, sptep, tptep, pte);
1992 /* Success and a new mapping */
1993 gmap_insert_rmap(sg, vmaddr, rmap);
1997 gmap_pte_op_end(ptl);
1998 spin_unlock(&sg->guest_table_lock);
2000 radix_tree_preload_end();
2003 rc = gmap_pte_op_fixup(parent, paddr, vmaddr, prot);
2010 EXPORT_SYMBOL_GPL(gmap_shadow_page);
2013 * gmap_shadow_notify - handle notifications for shadow gmap
2015 * Called with sg->parent->shadow_lock.
2017 static void gmap_shadow_notify(struct gmap *sg, unsigned long vmaddr,
2018 unsigned long gaddr, pte_t *pte)
2020 struct gmap_rmap *rmap, *rnext, *head;
2021 unsigned long start, end, bits, raddr;
2023 BUG_ON(!gmap_is_shadow(sg));
2025 spin_lock(&sg->guest_table_lock);
2027 spin_unlock(&sg->guest_table_lock);
2030 /* Check for top level table */
2031 start = sg->orig_asce & _ASCE_ORIGIN;
2032 end = start + ((sg->orig_asce & _ASCE_TABLE_LENGTH) + 1) * PAGE_SIZE;
2033 if (!(sg->orig_asce & _ASCE_REAL_SPACE) && gaddr >= start &&
2035 /* The complete shadow table has to go */
2037 spin_unlock(&sg->guest_table_lock);
2038 list_del(&sg->list);
2042 /* Remove the page table tree from on specific entry */
2043 head = radix_tree_delete(&sg->host_to_rmap, vmaddr >> PAGE_SHIFT);
2044 gmap_for_each_rmap_safe(rmap, rnext, head) {
2045 bits = rmap->raddr & _SHADOW_RMAP_MASK;
2046 raddr = rmap->raddr ^ bits;
2048 case _SHADOW_RMAP_REGION1:
2049 gmap_unshadow_r2t(sg, raddr);
2051 case _SHADOW_RMAP_REGION2:
2052 gmap_unshadow_r3t(sg, raddr);
2054 case _SHADOW_RMAP_REGION3:
2055 gmap_unshadow_sgt(sg, raddr);
2057 case _SHADOW_RMAP_SEGMENT:
2058 gmap_unshadow_pgt(sg, raddr);
2060 case _SHADOW_RMAP_PGTABLE:
2061 gmap_unshadow_page(sg, raddr);
2066 spin_unlock(&sg->guest_table_lock);
2070 * ptep_notify - call all invalidation callbacks for a specific pte.
2071 * @mm: pointer to the process mm_struct
2072 * @addr: virtual address in the process address space
2073 * @pte: pointer to the page table entry
2074 * @bits: bits from the pgste that caused the notify call
2076 * This function is assumed to be called with the page table lock held
2077 * for the pte to notify.
2079 void ptep_notify(struct mm_struct *mm, unsigned long vmaddr,
2080 pte_t *pte, unsigned long bits)
2082 unsigned long offset, gaddr = 0;
2083 unsigned long *table;
2084 struct gmap *gmap, *sg, *next;
2086 offset = ((unsigned long) pte) & (255 * sizeof(pte_t));
2087 offset = offset * (PAGE_SIZE / sizeof(pte_t));
2089 list_for_each_entry_rcu(gmap, &mm->context.gmap_list, list) {
2090 spin_lock(&gmap->guest_table_lock);
2091 table = radix_tree_lookup(&gmap->host_to_guest,
2092 vmaddr >> PMD_SHIFT);
2094 gaddr = __gmap_segment_gaddr(table) + offset;
2095 spin_unlock(&gmap->guest_table_lock);
2099 if (!list_empty(&gmap->children) && (bits & PGSTE_VSIE_BIT)) {
2100 spin_lock(&gmap->shadow_lock);
2101 list_for_each_entry_safe(sg, next,
2102 &gmap->children, list)
2103 gmap_shadow_notify(sg, vmaddr, gaddr, pte);
2104 spin_unlock(&gmap->shadow_lock);
2106 if (bits & PGSTE_IN_BIT)
2107 gmap_call_notifier(gmap, gaddr, gaddr + PAGE_SIZE - 1);
2111 EXPORT_SYMBOL_GPL(ptep_notify);
2113 static inline void thp_split_mm(struct mm_struct *mm)
2115 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
2116 struct vm_area_struct *vma;
2119 for (vma = mm->mmap; vma != NULL; vma = vma->vm_next) {
2120 for (addr = vma->vm_start;
2123 follow_page(vma, addr, FOLL_SPLIT);
2124 vma->vm_flags &= ~VM_HUGEPAGE;
2125 vma->vm_flags |= VM_NOHUGEPAGE;
2127 mm->def_flags |= VM_NOHUGEPAGE;
2132 * Remove all empty zero pages from the mapping for lazy refaulting
2133 * - This must be called after mm->context.has_pgste is set, to avoid
2134 * future creation of zero pages
2135 * - This must be called after THP was enabled
2137 static int __zap_zero_pages(pmd_t *pmd, unsigned long start,
2138 unsigned long end, struct mm_walk *walk)
2142 for (addr = start; addr != end; addr += PAGE_SIZE) {
2146 ptep = pte_offset_map_lock(walk->mm, pmd, addr, &ptl);
2147 if (is_zero_pfn(pte_pfn(*ptep)))
2148 ptep_xchg_direct(walk->mm, addr, ptep, __pte(_PAGE_INVALID));
2149 pte_unmap_unlock(ptep, ptl);
2154 static inline void zap_zero_pages(struct mm_struct *mm)
2156 struct mm_walk walk = { .pmd_entry = __zap_zero_pages };
2159 walk_page_range(0, TASK_SIZE, &walk);
2163 * switch on pgstes for its userspace process (for kvm)
2165 int s390_enable_sie(void)
2167 struct mm_struct *mm = current->mm;
2169 /* Do we have pgstes? if yes, we are done */
2170 if (mm_has_pgste(mm))
2172 /* Fail if the page tables are 2K */
2173 if (!mm_alloc_pgste(mm))
2175 down_write(&mm->mmap_sem);
2176 mm->context.has_pgste = 1;
2177 /* split thp mappings and disable thp for future mappings */
2180 up_write(&mm->mmap_sem);
2183 EXPORT_SYMBOL_GPL(s390_enable_sie);
2186 * Enable storage key handling from now on and initialize the storage
2187 * keys with the default key.
2189 static int __s390_enable_skey(pte_t *pte, unsigned long addr,
2190 unsigned long next, struct mm_walk *walk)
2192 /* Clear storage key */
2193 ptep_zap_key(walk->mm, addr, pte);
2197 int s390_enable_skey(void)
2199 struct mm_walk walk = { .pte_entry = __s390_enable_skey };
2200 struct mm_struct *mm = current->mm;
2201 struct vm_area_struct *vma;
2204 down_write(&mm->mmap_sem);
2205 if (mm_use_skey(mm))
2208 mm->context.use_skey = 1;
2209 for (vma = mm->mmap; vma; vma = vma->vm_next) {
2210 if (ksm_madvise(vma, vma->vm_start, vma->vm_end,
2211 MADV_UNMERGEABLE, &vma->vm_flags)) {
2212 mm->context.use_skey = 0;
2217 mm->def_flags &= ~VM_MERGEABLE;
2220 walk_page_range(0, TASK_SIZE, &walk);
2223 up_write(&mm->mmap_sem);
2226 EXPORT_SYMBOL_GPL(s390_enable_skey);
2229 * Reset CMMA state, make all pages stable again.
2231 static int __s390_reset_cmma(pte_t *pte, unsigned long addr,
2232 unsigned long next, struct mm_walk *walk)
2234 ptep_zap_unused(walk->mm, addr, pte, 1);
2238 void s390_reset_cmma(struct mm_struct *mm)
2240 struct mm_walk walk = { .pte_entry = __s390_reset_cmma };
2242 down_write(&mm->mmap_sem);
2244 walk_page_range(0, TASK_SIZE, &walk);
2245 up_write(&mm->mmap_sem);
2247 EXPORT_SYMBOL_GPL(s390_reset_cmma);