1 // SPDX-License-Identifier: GPL-2.0
3 * KVM guest address space mapping code
5 * Copyright IBM Corp. 2007, 2020
6 * Author(s): Martin Schwidefsky <schwidefsky@de.ibm.com>
7 * David Hildenbrand <david@redhat.com>
8 * Janosch Frank <frankja@linux.vnet.ibm.com>
11 #include <linux/kernel.h>
12 #include <linux/pagewalk.h>
13 #include <linux/swap.h>
14 #include <linux/smp.h>
15 #include <linux/spinlock.h>
16 #include <linux/slab.h>
17 #include <linux/swapops.h>
18 #include <linux/ksm.h>
19 #include <linux/mman.h>
20 #include <linux/pgtable.h>
22 #include <asm/pgalloc.h>
26 #define GMAP_SHADOW_FAKE_TABLE 1ULL
29 * gmap_alloc - allocate and initialize a guest address space
30 * @limit: maximum address of the gmap address space
32 * Returns a guest address space structure.
34 static struct gmap *gmap_alloc(unsigned long limit)
39 unsigned long etype, atype;
41 if (limit < _REGION3_SIZE) {
42 limit = _REGION3_SIZE - 1;
43 atype = _ASCE_TYPE_SEGMENT;
44 etype = _SEGMENT_ENTRY_EMPTY;
45 } else if (limit < _REGION2_SIZE) {
46 limit = _REGION2_SIZE - 1;
47 atype = _ASCE_TYPE_REGION3;
48 etype = _REGION3_ENTRY_EMPTY;
49 } else if (limit < _REGION1_SIZE) {
50 limit = _REGION1_SIZE - 1;
51 atype = _ASCE_TYPE_REGION2;
52 etype = _REGION2_ENTRY_EMPTY;
55 atype = _ASCE_TYPE_REGION1;
56 etype = _REGION1_ENTRY_EMPTY;
58 gmap = kzalloc(sizeof(struct gmap), GFP_KERNEL_ACCOUNT);
61 INIT_LIST_HEAD(&gmap->crst_list);
62 INIT_LIST_HEAD(&gmap->children);
63 INIT_LIST_HEAD(&gmap->pt_list);
64 INIT_RADIX_TREE(&gmap->guest_to_host, GFP_KERNEL_ACCOUNT);
65 INIT_RADIX_TREE(&gmap->host_to_guest, GFP_ATOMIC | __GFP_ACCOUNT);
66 INIT_RADIX_TREE(&gmap->host_to_rmap, GFP_ATOMIC | __GFP_ACCOUNT);
67 spin_lock_init(&gmap->guest_table_lock);
68 spin_lock_init(&gmap->shadow_lock);
69 refcount_set(&gmap->ref_count, 1);
70 page = alloc_pages(GFP_KERNEL_ACCOUNT, CRST_ALLOC_ORDER);
74 list_add(&page->lru, &gmap->crst_list);
75 table = page_to_virt(page);
76 crst_table_init(table, etype);
78 gmap->asce = atype | _ASCE_TABLE_LENGTH |
79 _ASCE_USER_BITS | __pa(table);
80 gmap->asce_end = limit;
90 * gmap_create - create a guest address space
91 * @mm: pointer to the parent mm_struct
92 * @limit: maximum size of the gmap address space
94 * Returns a guest address space structure.
96 struct gmap *gmap_create(struct mm_struct *mm, unsigned long limit)
99 unsigned long gmap_asce;
101 gmap = gmap_alloc(limit);
105 spin_lock(&mm->context.lock);
106 list_add_rcu(&gmap->list, &mm->context.gmap_list);
107 if (list_is_singular(&mm->context.gmap_list))
108 gmap_asce = gmap->asce;
111 WRITE_ONCE(mm->context.gmap_asce, gmap_asce);
112 spin_unlock(&mm->context.lock);
115 EXPORT_SYMBOL_GPL(gmap_create);
117 static void gmap_flush_tlb(struct gmap *gmap)
119 if (MACHINE_HAS_IDTE)
120 __tlb_flush_idte(gmap->asce);
122 __tlb_flush_global();
125 static void gmap_radix_tree_free(struct radix_tree_root *root)
127 struct radix_tree_iter iter;
128 unsigned long indices[16];
133 /* A radix tree is freed by deleting all of its entries */
137 radix_tree_for_each_slot(slot, root, &iter, index) {
138 indices[nr] = iter.index;
142 for (i = 0; i < nr; i++) {
144 radix_tree_delete(root, index);
149 static void gmap_rmap_radix_tree_free(struct radix_tree_root *root)
151 struct gmap_rmap *rmap, *rnext, *head;
152 struct radix_tree_iter iter;
153 unsigned long indices[16];
158 /* A radix tree is freed by deleting all of its entries */
162 radix_tree_for_each_slot(slot, root, &iter, index) {
163 indices[nr] = iter.index;
167 for (i = 0; i < nr; i++) {
169 head = radix_tree_delete(root, index);
170 gmap_for_each_rmap_safe(rmap, rnext, head)
177 * gmap_free - free a guest address space
178 * @gmap: pointer to the guest address space structure
180 * No locks required. There are no references to this gmap anymore.
182 static void gmap_free(struct gmap *gmap)
184 struct page *page, *next;
186 /* Flush tlb of all gmaps (if not already done for shadows) */
187 if (!(gmap_is_shadow(gmap) && gmap->removed))
188 gmap_flush_tlb(gmap);
189 /* Free all segment & region tables. */
190 list_for_each_entry_safe(page, next, &gmap->crst_list, lru)
191 __free_pages(page, CRST_ALLOC_ORDER);
192 gmap_radix_tree_free(&gmap->guest_to_host);
193 gmap_radix_tree_free(&gmap->host_to_guest);
195 /* Free additional data for a shadow gmap */
196 if (gmap_is_shadow(gmap)) {
197 /* Free all page tables. */
198 list_for_each_entry_safe(page, next, &gmap->pt_list, lru)
199 page_table_free_pgste(page);
200 gmap_rmap_radix_tree_free(&gmap->host_to_rmap);
201 /* Release reference to the parent */
202 gmap_put(gmap->parent);
209 * gmap_get - increase reference counter for guest address space
210 * @gmap: pointer to the guest address space structure
212 * Returns the gmap pointer
214 struct gmap *gmap_get(struct gmap *gmap)
216 refcount_inc(&gmap->ref_count);
219 EXPORT_SYMBOL_GPL(gmap_get);
222 * gmap_put - decrease reference counter for guest address space
223 * @gmap: pointer to the guest address space structure
225 * If the reference counter reaches zero the guest address space is freed.
227 void gmap_put(struct gmap *gmap)
229 if (refcount_dec_and_test(&gmap->ref_count))
232 EXPORT_SYMBOL_GPL(gmap_put);
235 * gmap_remove - remove a guest address space but do not free it yet
236 * @gmap: pointer to the guest address space structure
238 void gmap_remove(struct gmap *gmap)
240 struct gmap *sg, *next;
241 unsigned long gmap_asce;
243 /* Remove all shadow gmaps linked to this gmap */
244 if (!list_empty(&gmap->children)) {
245 spin_lock(&gmap->shadow_lock);
246 list_for_each_entry_safe(sg, next, &gmap->children, list) {
250 spin_unlock(&gmap->shadow_lock);
252 /* Remove gmap from the pre-mm list */
253 spin_lock(&gmap->mm->context.lock);
254 list_del_rcu(&gmap->list);
255 if (list_empty(&gmap->mm->context.gmap_list))
257 else if (list_is_singular(&gmap->mm->context.gmap_list))
258 gmap_asce = list_first_entry(&gmap->mm->context.gmap_list,
259 struct gmap, list)->asce;
262 WRITE_ONCE(gmap->mm->context.gmap_asce, gmap_asce);
263 spin_unlock(&gmap->mm->context.lock);
268 EXPORT_SYMBOL_GPL(gmap_remove);
271 * gmap_enable - switch primary space to the guest address space
272 * @gmap: pointer to the guest address space structure
274 void gmap_enable(struct gmap *gmap)
276 S390_lowcore.gmap = (unsigned long) gmap;
278 EXPORT_SYMBOL_GPL(gmap_enable);
281 * gmap_disable - switch back to the standard primary address space
282 * @gmap: pointer to the guest address space structure
284 void gmap_disable(struct gmap *gmap)
286 S390_lowcore.gmap = 0UL;
288 EXPORT_SYMBOL_GPL(gmap_disable);
291 * gmap_get_enabled - get a pointer to the currently enabled gmap
293 * Returns a pointer to the currently enabled gmap. 0 if none is enabled.
295 struct gmap *gmap_get_enabled(void)
297 return (struct gmap *) S390_lowcore.gmap;
299 EXPORT_SYMBOL_GPL(gmap_get_enabled);
302 * gmap_alloc_table is assumed to be called with mmap_lock held
304 static int gmap_alloc_table(struct gmap *gmap, unsigned long *table,
305 unsigned long init, unsigned long gaddr)
310 /* since we dont free the gmap table until gmap_free we can unlock */
311 page = alloc_pages(GFP_KERNEL_ACCOUNT, CRST_ALLOC_ORDER);
314 new = page_to_virt(page);
315 crst_table_init(new, init);
316 spin_lock(&gmap->guest_table_lock);
317 if (*table & _REGION_ENTRY_INVALID) {
318 list_add(&page->lru, &gmap->crst_list);
319 *table = __pa(new) | _REGION_ENTRY_LENGTH |
320 (*table & _REGION_ENTRY_TYPE_MASK);
324 spin_unlock(&gmap->guest_table_lock);
326 __free_pages(page, CRST_ALLOC_ORDER);
331 * __gmap_segment_gaddr - find virtual address from segment pointer
332 * @entry: pointer to a segment table entry in the guest address space
334 * Returns the virtual address in the guest address space for the segment
336 static unsigned long __gmap_segment_gaddr(unsigned long *entry)
339 unsigned long offset;
341 offset = (unsigned long) entry / sizeof(unsigned long);
342 offset = (offset & (PTRS_PER_PMD - 1)) * PMD_SIZE;
343 page = pmd_pgtable_page((pmd_t *) entry);
344 return page->index + offset;
348 * __gmap_unlink_by_vmaddr - unlink a single segment via a host address
349 * @gmap: pointer to the guest address space structure
350 * @vmaddr: address in the host process address space
352 * Returns 1 if a TLB flush is required
354 static int __gmap_unlink_by_vmaddr(struct gmap *gmap, unsigned long vmaddr)
356 unsigned long *entry;
359 BUG_ON(gmap_is_shadow(gmap));
360 spin_lock(&gmap->guest_table_lock);
361 entry = radix_tree_delete(&gmap->host_to_guest, vmaddr >> PMD_SHIFT);
363 flush = (*entry != _SEGMENT_ENTRY_EMPTY);
364 *entry = _SEGMENT_ENTRY_EMPTY;
366 spin_unlock(&gmap->guest_table_lock);
371 * __gmap_unmap_by_gaddr - unmap a single segment via a guest address
372 * @gmap: pointer to the guest address space structure
373 * @gaddr: address in the guest address space
375 * Returns 1 if a TLB flush is required
377 static int __gmap_unmap_by_gaddr(struct gmap *gmap, unsigned long gaddr)
379 unsigned long vmaddr;
381 vmaddr = (unsigned long) radix_tree_delete(&gmap->guest_to_host,
383 return vmaddr ? __gmap_unlink_by_vmaddr(gmap, vmaddr) : 0;
387 * gmap_unmap_segment - unmap segment from the guest address space
388 * @gmap: pointer to the guest address space structure
389 * @to: address in the guest address space
390 * @len: length of the memory area to unmap
392 * Returns 0 if the unmap succeeded, -EINVAL if not.
394 int gmap_unmap_segment(struct gmap *gmap, unsigned long to, unsigned long len)
399 BUG_ON(gmap_is_shadow(gmap));
400 if ((to | len) & (PMD_SIZE - 1))
402 if (len == 0 || to + len < to)
406 mmap_write_lock(gmap->mm);
407 for (off = 0; off < len; off += PMD_SIZE)
408 flush |= __gmap_unmap_by_gaddr(gmap, to + off);
409 mmap_write_unlock(gmap->mm);
411 gmap_flush_tlb(gmap);
414 EXPORT_SYMBOL_GPL(gmap_unmap_segment);
417 * gmap_map_segment - map a segment to the guest address space
418 * @gmap: pointer to the guest address space structure
419 * @from: source address in the parent address space
420 * @to: target address in the guest address space
421 * @len: length of the memory area to map
423 * Returns 0 if the mmap succeeded, -EINVAL or -ENOMEM if not.
425 int gmap_map_segment(struct gmap *gmap, unsigned long from,
426 unsigned long to, unsigned long len)
431 BUG_ON(gmap_is_shadow(gmap));
432 if ((from | to | len) & (PMD_SIZE - 1))
434 if (len == 0 || from + len < from || to + len < to ||
435 from + len - 1 > TASK_SIZE_MAX || to + len - 1 > gmap->asce_end)
439 mmap_write_lock(gmap->mm);
440 for (off = 0; off < len; off += PMD_SIZE) {
441 /* Remove old translation */
442 flush |= __gmap_unmap_by_gaddr(gmap, to + off);
443 /* Store new translation */
444 if (radix_tree_insert(&gmap->guest_to_host,
445 (to + off) >> PMD_SHIFT,
446 (void *) from + off))
449 mmap_write_unlock(gmap->mm);
451 gmap_flush_tlb(gmap);
454 gmap_unmap_segment(gmap, to, len);
457 EXPORT_SYMBOL_GPL(gmap_map_segment);
460 * __gmap_translate - translate a guest address to a user space address
461 * @gmap: pointer to guest mapping meta data structure
462 * @gaddr: guest address
464 * Returns user space address which corresponds to the guest address or
465 * -EFAULT if no such mapping exists.
466 * This function does not establish potentially missing page table entries.
467 * The mmap_lock of the mm that belongs to the address space must be held
468 * when this function gets called.
470 * Note: Can also be called for shadow gmaps.
472 unsigned long __gmap_translate(struct gmap *gmap, unsigned long gaddr)
474 unsigned long vmaddr;
476 vmaddr = (unsigned long)
477 radix_tree_lookup(&gmap->guest_to_host, gaddr >> PMD_SHIFT);
478 /* Note: guest_to_host is empty for a shadow gmap */
479 return vmaddr ? (vmaddr | (gaddr & ~PMD_MASK)) : -EFAULT;
481 EXPORT_SYMBOL_GPL(__gmap_translate);
484 * gmap_translate - translate a guest address to a user space address
485 * @gmap: pointer to guest mapping meta data structure
486 * @gaddr: guest address
488 * Returns user space address which corresponds to the guest address or
489 * -EFAULT if no such mapping exists.
490 * This function does not establish potentially missing page table entries.
492 unsigned long gmap_translate(struct gmap *gmap, unsigned long gaddr)
496 mmap_read_lock(gmap->mm);
497 rc = __gmap_translate(gmap, gaddr);
498 mmap_read_unlock(gmap->mm);
501 EXPORT_SYMBOL_GPL(gmap_translate);
504 * gmap_unlink - disconnect a page table from the gmap shadow tables
505 * @mm: pointer to the parent mm_struct
506 * @table: pointer to the host page table
507 * @vmaddr: vm address associated with the host page table
509 void gmap_unlink(struct mm_struct *mm, unsigned long *table,
510 unsigned long vmaddr)
516 list_for_each_entry_rcu(gmap, &mm->context.gmap_list, list) {
517 flush = __gmap_unlink_by_vmaddr(gmap, vmaddr);
519 gmap_flush_tlb(gmap);
524 static void gmap_pmdp_xchg(struct gmap *gmap, pmd_t *old, pmd_t new,
525 unsigned long gaddr);
528 * __gmap_link - set up shadow page tables to connect a host to a guest address
529 * @gmap: pointer to guest mapping meta data structure
530 * @gaddr: guest address
531 * @vmaddr: vm address
533 * Returns 0 on success, -ENOMEM for out of memory conditions, and -EFAULT
534 * if the vm address is already mapped to a different guest segment.
535 * The mmap_lock of the mm that belongs to the address space must be held
536 * when this function gets called.
538 int __gmap_link(struct gmap *gmap, unsigned long gaddr, unsigned long vmaddr)
540 struct mm_struct *mm;
541 unsigned long *table;
550 BUG_ON(gmap_is_shadow(gmap));
551 /* Create higher level tables in the gmap page table */
553 if ((gmap->asce & _ASCE_TYPE_MASK) >= _ASCE_TYPE_REGION1) {
554 table += (gaddr & _REGION1_INDEX) >> _REGION1_SHIFT;
555 if ((*table & _REGION_ENTRY_INVALID) &&
556 gmap_alloc_table(gmap, table, _REGION2_ENTRY_EMPTY,
557 gaddr & _REGION1_MASK))
559 table = __va(*table & _REGION_ENTRY_ORIGIN);
561 if ((gmap->asce & _ASCE_TYPE_MASK) >= _ASCE_TYPE_REGION2) {
562 table += (gaddr & _REGION2_INDEX) >> _REGION2_SHIFT;
563 if ((*table & _REGION_ENTRY_INVALID) &&
564 gmap_alloc_table(gmap, table, _REGION3_ENTRY_EMPTY,
565 gaddr & _REGION2_MASK))
567 table = __va(*table & _REGION_ENTRY_ORIGIN);
569 if ((gmap->asce & _ASCE_TYPE_MASK) >= _ASCE_TYPE_REGION3) {
570 table += (gaddr & _REGION3_INDEX) >> _REGION3_SHIFT;
571 if ((*table & _REGION_ENTRY_INVALID) &&
572 gmap_alloc_table(gmap, table, _SEGMENT_ENTRY_EMPTY,
573 gaddr & _REGION3_MASK))
575 table = __va(*table & _REGION_ENTRY_ORIGIN);
577 table += (gaddr & _SEGMENT_INDEX) >> _SEGMENT_SHIFT;
578 /* Walk the parent mm page table */
580 pgd = pgd_offset(mm, vmaddr);
581 VM_BUG_ON(pgd_none(*pgd));
582 p4d = p4d_offset(pgd, vmaddr);
583 VM_BUG_ON(p4d_none(*p4d));
584 pud = pud_offset(p4d, vmaddr);
585 VM_BUG_ON(pud_none(*pud));
586 /* large puds cannot yet be handled */
589 pmd = pmd_offset(pud, vmaddr);
590 VM_BUG_ON(pmd_none(*pmd));
591 /* Are we allowed to use huge pages? */
592 if (pmd_large(*pmd) && !gmap->mm->context.allow_gmap_hpage_1m)
594 /* Link gmap segment table entry location to page table. */
595 rc = radix_tree_preload(GFP_KERNEL_ACCOUNT);
598 ptl = pmd_lock(mm, pmd);
599 spin_lock(&gmap->guest_table_lock);
600 if (*table == _SEGMENT_ENTRY_EMPTY) {
601 rc = radix_tree_insert(&gmap->host_to_guest,
602 vmaddr >> PMD_SHIFT, table);
604 if (pmd_large(*pmd)) {
605 *table = (pmd_val(*pmd) &
606 _SEGMENT_ENTRY_HARDWARE_BITS_LARGE)
607 | _SEGMENT_ENTRY_GMAP_UC;
609 *table = pmd_val(*pmd) &
610 _SEGMENT_ENTRY_HARDWARE_BITS;
612 } else if (*table & _SEGMENT_ENTRY_PROTECT &&
613 !(pmd_val(*pmd) & _SEGMENT_ENTRY_PROTECT)) {
614 unprot = (u64)*table;
615 unprot &= ~_SEGMENT_ENTRY_PROTECT;
616 unprot |= _SEGMENT_ENTRY_GMAP_UC;
617 gmap_pmdp_xchg(gmap, (pmd_t *)table, __pmd(unprot), gaddr);
619 spin_unlock(&gmap->guest_table_lock);
621 radix_tree_preload_end();
626 * gmap_fault - resolve a fault on a guest address
627 * @gmap: pointer to guest mapping meta data structure
628 * @gaddr: guest address
629 * @fault_flags: flags to pass down to handle_mm_fault()
631 * Returns 0 on success, -ENOMEM for out of memory conditions, and -EFAULT
632 * if the vm address is already mapped to a different guest segment.
634 int gmap_fault(struct gmap *gmap, unsigned long gaddr,
635 unsigned int fault_flags)
637 unsigned long vmaddr;
641 mmap_read_lock(gmap->mm);
645 vmaddr = __gmap_translate(gmap, gaddr);
646 if (IS_ERR_VALUE(vmaddr)) {
650 if (fixup_user_fault(gmap->mm, vmaddr, fault_flags,
656 * In the case that fixup_user_fault unlocked the mmap_lock during
657 * faultin redo __gmap_translate to not race with a map/unmap_segment.
662 rc = __gmap_link(gmap, gaddr, vmaddr);
664 mmap_read_unlock(gmap->mm);
667 EXPORT_SYMBOL_GPL(gmap_fault);
670 * this function is assumed to be called with mmap_lock held
672 void __gmap_zap(struct gmap *gmap, unsigned long gaddr)
674 struct vm_area_struct *vma;
675 unsigned long vmaddr;
679 /* Find the vm address for the guest address */
680 vmaddr = (unsigned long) radix_tree_lookup(&gmap->guest_to_host,
683 vmaddr |= gaddr & ~PMD_MASK;
685 vma = vma_lookup(gmap->mm, vmaddr);
686 if (!vma || is_vm_hugetlb_page(vma))
689 /* Get pointer to the page table entry */
690 ptep = get_locked_pte(gmap->mm, vmaddr, &ptl);
692 ptep_zap_unused(gmap->mm, vmaddr, ptep, 0);
693 pte_unmap_unlock(ptep, ptl);
697 EXPORT_SYMBOL_GPL(__gmap_zap);
699 void gmap_discard(struct gmap *gmap, unsigned long from, unsigned long to)
701 unsigned long gaddr, vmaddr, size;
702 struct vm_area_struct *vma;
704 mmap_read_lock(gmap->mm);
705 for (gaddr = from; gaddr < to;
706 gaddr = (gaddr + PMD_SIZE) & PMD_MASK) {
707 /* Find the vm address for the guest address */
708 vmaddr = (unsigned long)
709 radix_tree_lookup(&gmap->guest_to_host,
713 vmaddr |= gaddr & ~PMD_MASK;
714 /* Find vma in the parent mm */
715 vma = find_vma(gmap->mm, vmaddr);
719 * We do not discard pages that are backed by
720 * hugetlbfs, so we don't have to refault them.
722 if (is_vm_hugetlb_page(vma))
724 size = min(to - gaddr, PMD_SIZE - (gaddr & ~PMD_MASK));
725 zap_page_range_single(vma, vmaddr, size, NULL);
727 mmap_read_unlock(gmap->mm);
729 EXPORT_SYMBOL_GPL(gmap_discard);
731 static LIST_HEAD(gmap_notifier_list);
732 static DEFINE_SPINLOCK(gmap_notifier_lock);
735 * gmap_register_pte_notifier - register a pte invalidation callback
736 * @nb: pointer to the gmap notifier block
738 void gmap_register_pte_notifier(struct gmap_notifier *nb)
740 spin_lock(&gmap_notifier_lock);
741 list_add_rcu(&nb->list, &gmap_notifier_list);
742 spin_unlock(&gmap_notifier_lock);
744 EXPORT_SYMBOL_GPL(gmap_register_pte_notifier);
747 * gmap_unregister_pte_notifier - remove a pte invalidation callback
748 * @nb: pointer to the gmap notifier block
750 void gmap_unregister_pte_notifier(struct gmap_notifier *nb)
752 spin_lock(&gmap_notifier_lock);
753 list_del_rcu(&nb->list);
754 spin_unlock(&gmap_notifier_lock);
757 EXPORT_SYMBOL_GPL(gmap_unregister_pte_notifier);
760 * gmap_call_notifier - call all registered invalidation callbacks
761 * @gmap: pointer to guest mapping meta data structure
762 * @start: start virtual address in the guest address space
763 * @end: end virtual address in the guest address space
765 static void gmap_call_notifier(struct gmap *gmap, unsigned long start,
768 struct gmap_notifier *nb;
770 list_for_each_entry(nb, &gmap_notifier_list, list)
771 nb->notifier_call(gmap, start, end);
775 * gmap_table_walk - walk the gmap page tables
776 * @gmap: pointer to guest mapping meta data structure
777 * @gaddr: virtual address in the guest address space
778 * @level: page table level to stop at
780 * Returns a table entry pointer for the given guest address and @level
781 * @level=0 : returns a pointer to a page table table entry (or NULL)
782 * @level=1 : returns a pointer to a segment table entry (or NULL)
783 * @level=2 : returns a pointer to a region-3 table entry (or NULL)
784 * @level=3 : returns a pointer to a region-2 table entry (or NULL)
785 * @level=4 : returns a pointer to a region-1 table entry (or NULL)
787 * Returns NULL if the gmap page tables could not be walked to the
790 * Note: Can also be called for shadow gmaps.
792 static inline unsigned long *gmap_table_walk(struct gmap *gmap,
793 unsigned long gaddr, int level)
795 const int asce_type = gmap->asce & _ASCE_TYPE_MASK;
796 unsigned long *table = gmap->table;
798 if (gmap_is_shadow(gmap) && gmap->removed)
801 if (WARN_ON_ONCE(level > (asce_type >> 2) + 1))
804 if (asce_type != _ASCE_TYPE_REGION1 &&
805 gaddr & (-1UL << (31 + (asce_type >> 2) * 11)))
809 case _ASCE_TYPE_REGION1:
810 table += (gaddr & _REGION1_INDEX) >> _REGION1_SHIFT;
813 if (*table & _REGION_ENTRY_INVALID)
815 table = __va(*table & _REGION_ENTRY_ORIGIN);
817 case _ASCE_TYPE_REGION2:
818 table += (gaddr & _REGION2_INDEX) >> _REGION2_SHIFT;
821 if (*table & _REGION_ENTRY_INVALID)
823 table = __va(*table & _REGION_ENTRY_ORIGIN);
825 case _ASCE_TYPE_REGION3:
826 table += (gaddr & _REGION3_INDEX) >> _REGION3_SHIFT;
829 if (*table & _REGION_ENTRY_INVALID)
831 table = __va(*table & _REGION_ENTRY_ORIGIN);
833 case _ASCE_TYPE_SEGMENT:
834 table += (gaddr & _SEGMENT_INDEX) >> _SEGMENT_SHIFT;
837 if (*table & _REGION_ENTRY_INVALID)
839 table = __va(*table & _SEGMENT_ENTRY_ORIGIN);
840 table += (gaddr & _PAGE_INDEX) >> _PAGE_SHIFT;
846 * gmap_pte_op_walk - walk the gmap page table, get the page table lock
847 * and return the pte pointer
848 * @gmap: pointer to guest mapping meta data structure
849 * @gaddr: virtual address in the guest address space
850 * @ptl: pointer to the spinlock pointer
852 * Returns a pointer to the locked pte for a guest address, or NULL
854 static pte_t *gmap_pte_op_walk(struct gmap *gmap, unsigned long gaddr,
857 unsigned long *table;
859 BUG_ON(gmap_is_shadow(gmap));
860 /* Walk the gmap page table, lock and get pte pointer */
861 table = gmap_table_walk(gmap, gaddr, 1); /* get segment pointer */
862 if (!table || *table & _SEGMENT_ENTRY_INVALID)
864 return pte_alloc_map_lock(gmap->mm, (pmd_t *) table, gaddr, ptl);
868 * gmap_pte_op_fixup - force a page in and connect the gmap page table
869 * @gmap: pointer to guest mapping meta data structure
870 * @gaddr: virtual address in the guest address space
871 * @vmaddr: address in the host process address space
872 * @prot: indicates access rights: PROT_NONE, PROT_READ or PROT_WRITE
874 * Returns 0 if the caller can retry __gmap_translate (might fail again),
875 * -ENOMEM if out of memory and -EFAULT if anything goes wrong while fixing
876 * up or connecting the gmap page table.
878 static int gmap_pte_op_fixup(struct gmap *gmap, unsigned long gaddr,
879 unsigned long vmaddr, int prot)
881 struct mm_struct *mm = gmap->mm;
882 unsigned int fault_flags;
883 bool unlocked = false;
885 BUG_ON(gmap_is_shadow(gmap));
886 fault_flags = (prot == PROT_WRITE) ? FAULT_FLAG_WRITE : 0;
887 if (fixup_user_fault(mm, vmaddr, fault_flags, &unlocked))
890 /* lost mmap_lock, caller has to retry __gmap_translate */
892 /* Connect the page tables */
893 return __gmap_link(gmap, gaddr, vmaddr);
897 * gmap_pte_op_end - release the page table lock
898 * @ptl: pointer to the spinlock pointer
900 static void gmap_pte_op_end(spinlock_t *ptl)
907 * gmap_pmd_op_walk - walk the gmap tables, get the guest table lock
908 * and return the pmd pointer
909 * @gmap: pointer to guest mapping meta data structure
910 * @gaddr: virtual address in the guest address space
912 * Returns a pointer to the pmd for a guest address, or NULL
914 static inline pmd_t *gmap_pmd_op_walk(struct gmap *gmap, unsigned long gaddr)
918 BUG_ON(gmap_is_shadow(gmap));
919 pmdp = (pmd_t *) gmap_table_walk(gmap, gaddr, 1);
923 /* without huge pages, there is no need to take the table lock */
924 if (!gmap->mm->context.allow_gmap_hpage_1m)
925 return pmd_none(*pmdp) ? NULL : pmdp;
927 spin_lock(&gmap->guest_table_lock);
928 if (pmd_none(*pmdp)) {
929 spin_unlock(&gmap->guest_table_lock);
933 /* 4k page table entries are locked via the pte (pte_alloc_map_lock). */
934 if (!pmd_large(*pmdp))
935 spin_unlock(&gmap->guest_table_lock);
940 * gmap_pmd_op_end - release the guest_table_lock if needed
941 * @gmap: pointer to the guest mapping meta data structure
942 * @pmdp: pointer to the pmd
944 static inline void gmap_pmd_op_end(struct gmap *gmap, pmd_t *pmdp)
946 if (pmd_large(*pmdp))
947 spin_unlock(&gmap->guest_table_lock);
951 * gmap_protect_pmd - remove access rights to memory and set pmd notification bits
952 * @pmdp: pointer to the pmd to be protected
953 * @prot: indicates access rights: PROT_NONE, PROT_READ or PROT_WRITE
954 * @bits: notification bits to set
957 * 0 if successfully protected
958 * -EAGAIN if a fixup is needed
959 * -EINVAL if unsupported notifier bits have been specified
961 * Expected to be called with sg->mm->mmap_lock in read and
962 * guest_table_lock held.
964 static int gmap_protect_pmd(struct gmap *gmap, unsigned long gaddr,
965 pmd_t *pmdp, int prot, unsigned long bits)
967 int pmd_i = pmd_val(*pmdp) & _SEGMENT_ENTRY_INVALID;
968 int pmd_p = pmd_val(*pmdp) & _SEGMENT_ENTRY_PROTECT;
972 if ((pmd_i && (prot != PROT_NONE)) || (pmd_p && (prot == PROT_WRITE)))
975 if (prot == PROT_NONE && !pmd_i) {
976 new = set_pmd_bit(new, __pgprot(_SEGMENT_ENTRY_INVALID));
977 gmap_pmdp_xchg(gmap, pmdp, new, gaddr);
980 if (prot == PROT_READ && !pmd_p) {
981 new = clear_pmd_bit(new, __pgprot(_SEGMENT_ENTRY_INVALID));
982 new = set_pmd_bit(new, __pgprot(_SEGMENT_ENTRY_PROTECT));
983 gmap_pmdp_xchg(gmap, pmdp, new, gaddr);
986 if (bits & GMAP_NOTIFY_MPROT)
987 set_pmd(pmdp, set_pmd_bit(*pmdp, __pgprot(_SEGMENT_ENTRY_GMAP_IN)));
989 /* Shadow GMAP protection needs split PMDs */
990 if (bits & GMAP_NOTIFY_SHADOW)
997 * gmap_protect_pte - remove access rights to memory and set pgste bits
998 * @gmap: pointer to guest mapping meta data structure
999 * @gaddr: virtual address in the guest address space
1000 * @pmdp: pointer to the pmd associated with the pte
1001 * @prot: indicates access rights: PROT_NONE, PROT_READ or PROT_WRITE
1002 * @bits: notification bits to set
1004 * Returns 0 if successfully protected, -ENOMEM if out of memory and
1005 * -EAGAIN if a fixup is needed.
1007 * Expected to be called with sg->mm->mmap_lock in read
1009 static int gmap_protect_pte(struct gmap *gmap, unsigned long gaddr,
1010 pmd_t *pmdp, int prot, unsigned long bits)
1014 spinlock_t *ptl = NULL;
1015 unsigned long pbits = 0;
1017 if (pmd_val(*pmdp) & _SEGMENT_ENTRY_INVALID)
1020 ptep = pte_alloc_map_lock(gmap->mm, pmdp, gaddr, &ptl);
1024 pbits |= (bits & GMAP_NOTIFY_MPROT) ? PGSTE_IN_BIT : 0;
1025 pbits |= (bits & GMAP_NOTIFY_SHADOW) ? PGSTE_VSIE_BIT : 0;
1026 /* Protect and unlock. */
1027 rc = ptep_force_prot(gmap->mm, gaddr, ptep, prot, pbits);
1028 gmap_pte_op_end(ptl);
1033 * gmap_protect_range - remove access rights to memory and set pgste bits
1034 * @gmap: pointer to guest mapping meta data structure
1035 * @gaddr: virtual address in the guest address space
1036 * @len: size of area
1037 * @prot: indicates access rights: PROT_NONE, PROT_READ or PROT_WRITE
1038 * @bits: pgste notification bits to set
1040 * Returns 0 if successfully protected, -ENOMEM if out of memory and
1041 * -EFAULT if gaddr is invalid (or mapping for shadows is missing).
1043 * Called with sg->mm->mmap_lock in read.
1045 static int gmap_protect_range(struct gmap *gmap, unsigned long gaddr,
1046 unsigned long len, int prot, unsigned long bits)
1048 unsigned long vmaddr, dist;
1052 BUG_ON(gmap_is_shadow(gmap));
1055 pmdp = gmap_pmd_op_walk(gmap, gaddr);
1057 if (!pmd_large(*pmdp)) {
1058 rc = gmap_protect_pte(gmap, gaddr, pmdp, prot,
1065 rc = gmap_protect_pmd(gmap, gaddr, pmdp, prot,
1068 dist = HPAGE_SIZE - (gaddr & ~HPAGE_MASK);
1069 len = len < dist ? 0 : len - dist;
1070 gaddr = (gaddr & HPAGE_MASK) + HPAGE_SIZE;
1073 gmap_pmd_op_end(gmap, pmdp);
1079 /* -EAGAIN, fixup of userspace mm and gmap */
1080 vmaddr = __gmap_translate(gmap, gaddr);
1081 if (IS_ERR_VALUE(vmaddr))
1083 rc = gmap_pte_op_fixup(gmap, gaddr, vmaddr, prot);
1092 * gmap_mprotect_notify - change access rights for a range of ptes and
1093 * call the notifier if any pte changes again
1094 * @gmap: pointer to guest mapping meta data structure
1095 * @gaddr: virtual address in the guest address space
1096 * @len: size of area
1097 * @prot: indicates access rights: PROT_NONE, PROT_READ or PROT_WRITE
1099 * Returns 0 if for each page in the given range a gmap mapping exists,
1100 * the new access rights could be set and the notifier could be armed.
1101 * If the gmap mapping is missing for one or more pages -EFAULT is
1102 * returned. If no memory could be allocated -ENOMEM is returned.
1103 * This function establishes missing page table entries.
1105 int gmap_mprotect_notify(struct gmap *gmap, unsigned long gaddr,
1106 unsigned long len, int prot)
1110 if ((gaddr & ~PAGE_MASK) || (len & ~PAGE_MASK) || gmap_is_shadow(gmap))
1112 if (!MACHINE_HAS_ESOP && prot == PROT_READ)
1114 mmap_read_lock(gmap->mm);
1115 rc = gmap_protect_range(gmap, gaddr, len, prot, GMAP_NOTIFY_MPROT);
1116 mmap_read_unlock(gmap->mm);
1119 EXPORT_SYMBOL_GPL(gmap_mprotect_notify);
1122 * gmap_read_table - get an unsigned long value from a guest page table using
1123 * absolute addressing, without marking the page referenced.
1124 * @gmap: pointer to guest mapping meta data structure
1125 * @gaddr: virtual address in the guest address space
1126 * @val: pointer to the unsigned long value to return
1128 * Returns 0 if the value was read, -ENOMEM if out of memory and -EFAULT
1129 * if reading using the virtual address failed. -EINVAL if called on a gmap
1132 * Called with gmap->mm->mmap_lock in read.
1134 int gmap_read_table(struct gmap *gmap, unsigned long gaddr, unsigned long *val)
1136 unsigned long address, vmaddr;
1141 if (gmap_is_shadow(gmap))
1146 ptep = gmap_pte_op_walk(gmap, gaddr, &ptl);
1149 if (pte_present(pte) && (pte_val(pte) & _PAGE_READ)) {
1150 address = pte_val(pte) & PAGE_MASK;
1151 address += gaddr & ~PAGE_MASK;
1152 *val = *(unsigned long *)__va(address);
1153 set_pte(ptep, set_pte_bit(*ptep, __pgprot(_PAGE_YOUNG)));
1154 /* Do *NOT* clear the _PAGE_INVALID bit! */
1157 gmap_pte_op_end(ptl);
1161 vmaddr = __gmap_translate(gmap, gaddr);
1162 if (IS_ERR_VALUE(vmaddr)) {
1166 rc = gmap_pte_op_fixup(gmap, gaddr, vmaddr, PROT_READ);
1172 EXPORT_SYMBOL_GPL(gmap_read_table);
1175 * gmap_insert_rmap - add a rmap to the host_to_rmap radix tree
1176 * @sg: pointer to the shadow guest address space structure
1177 * @vmaddr: vm address associated with the rmap
1178 * @rmap: pointer to the rmap structure
1180 * Called with the sg->guest_table_lock
1182 static inline void gmap_insert_rmap(struct gmap *sg, unsigned long vmaddr,
1183 struct gmap_rmap *rmap)
1185 struct gmap_rmap *temp;
1188 BUG_ON(!gmap_is_shadow(sg));
1189 slot = radix_tree_lookup_slot(&sg->host_to_rmap, vmaddr >> PAGE_SHIFT);
1191 rmap->next = radix_tree_deref_slot_protected(slot,
1192 &sg->guest_table_lock);
1193 for (temp = rmap->next; temp; temp = temp->next) {
1194 if (temp->raddr == rmap->raddr) {
1199 radix_tree_replace_slot(&sg->host_to_rmap, slot, rmap);
1202 radix_tree_insert(&sg->host_to_rmap, vmaddr >> PAGE_SHIFT,
1208 * gmap_protect_rmap - restrict access rights to memory (RO) and create an rmap
1209 * @sg: pointer to the shadow guest address space structure
1210 * @raddr: rmap address in the shadow gmap
1211 * @paddr: address in the parent guest address space
1212 * @len: length of the memory area to protect
1214 * Returns 0 if successfully protected and the rmap was created, -ENOMEM
1215 * if out of memory and -EFAULT if paddr is invalid.
1217 static int gmap_protect_rmap(struct gmap *sg, unsigned long raddr,
1218 unsigned long paddr, unsigned long len)
1220 struct gmap *parent;
1221 struct gmap_rmap *rmap;
1222 unsigned long vmaddr;
1227 BUG_ON(!gmap_is_shadow(sg));
1228 parent = sg->parent;
1230 vmaddr = __gmap_translate(parent, paddr);
1231 if (IS_ERR_VALUE(vmaddr))
1233 rmap = kzalloc(sizeof(*rmap), GFP_KERNEL_ACCOUNT);
1236 rmap->raddr = raddr;
1237 rc = radix_tree_preload(GFP_KERNEL_ACCOUNT);
1243 ptep = gmap_pte_op_walk(parent, paddr, &ptl);
1245 spin_lock(&sg->guest_table_lock);
1246 rc = ptep_force_prot(parent->mm, paddr, ptep, PROT_READ,
1249 gmap_insert_rmap(sg, vmaddr, rmap);
1250 spin_unlock(&sg->guest_table_lock);
1251 gmap_pte_op_end(ptl);
1253 radix_tree_preload_end();
1256 rc = gmap_pte_op_fixup(parent, paddr, vmaddr, PROT_READ);
1267 #define _SHADOW_RMAP_MASK 0x7
1268 #define _SHADOW_RMAP_REGION1 0x5
1269 #define _SHADOW_RMAP_REGION2 0x4
1270 #define _SHADOW_RMAP_REGION3 0x3
1271 #define _SHADOW_RMAP_SEGMENT 0x2
1272 #define _SHADOW_RMAP_PGTABLE 0x1
1275 * gmap_idte_one - invalidate a single region or segment table entry
1276 * @asce: region or segment table *origin* + table-type bits
1277 * @vaddr: virtual address to identify the table entry to flush
1279 * The invalid bit of a single region or segment table entry is set
1280 * and the associated TLB entries depending on the entry are flushed.
1281 * The table-type of the @asce identifies the portion of the @vaddr
1282 * that is used as the invalidation index.
1284 static inline void gmap_idte_one(unsigned long asce, unsigned long vaddr)
1288 : : "a" (asce), "a" (vaddr) : "cc", "memory");
1292 * gmap_unshadow_page - remove a page from a shadow page table
1293 * @sg: pointer to the shadow guest address space structure
1294 * @raddr: rmap address in the shadow guest address space
1296 * Called with the sg->guest_table_lock
1298 static void gmap_unshadow_page(struct gmap *sg, unsigned long raddr)
1300 unsigned long *table;
1302 BUG_ON(!gmap_is_shadow(sg));
1303 table = gmap_table_walk(sg, raddr, 0); /* get page table pointer */
1304 if (!table || *table & _PAGE_INVALID)
1306 gmap_call_notifier(sg, raddr, raddr + _PAGE_SIZE - 1);
1307 ptep_unshadow_pte(sg->mm, raddr, (pte_t *) table);
1311 * __gmap_unshadow_pgt - remove all entries from a shadow page table
1312 * @sg: pointer to the shadow guest address space structure
1313 * @raddr: rmap address in the shadow guest address space
1314 * @pgt: pointer to the start of a shadow page table
1316 * Called with the sg->guest_table_lock
1318 static void __gmap_unshadow_pgt(struct gmap *sg, unsigned long raddr,
1323 BUG_ON(!gmap_is_shadow(sg));
1324 for (i = 0; i < _PAGE_ENTRIES; i++, raddr += _PAGE_SIZE)
1325 pgt[i] = _PAGE_INVALID;
1329 * gmap_unshadow_pgt - remove a shadow page table from a segment entry
1330 * @sg: pointer to the shadow guest address space structure
1331 * @raddr: address in the shadow guest address space
1333 * Called with the sg->guest_table_lock
1335 static void gmap_unshadow_pgt(struct gmap *sg, unsigned long raddr)
1338 phys_addr_t sto, pgt;
1341 BUG_ON(!gmap_is_shadow(sg));
1342 ste = gmap_table_walk(sg, raddr, 1); /* get segment pointer */
1343 if (!ste || !(*ste & _SEGMENT_ENTRY_ORIGIN))
1345 gmap_call_notifier(sg, raddr, raddr + _SEGMENT_SIZE - 1);
1346 sto = __pa(ste - ((raddr & _SEGMENT_INDEX) >> _SEGMENT_SHIFT));
1347 gmap_idte_one(sto | _ASCE_TYPE_SEGMENT, raddr);
1348 pgt = *ste & _SEGMENT_ENTRY_ORIGIN;
1349 *ste = _SEGMENT_ENTRY_EMPTY;
1350 __gmap_unshadow_pgt(sg, raddr, __va(pgt));
1351 /* Free page table */
1352 page = phys_to_page(pgt);
1353 list_del(&page->lru);
1354 page_table_free_pgste(page);
1358 * __gmap_unshadow_sgt - remove all entries from a shadow segment table
1359 * @sg: pointer to the shadow guest address space structure
1360 * @raddr: rmap address in the shadow guest address space
1361 * @sgt: pointer to the start of a shadow segment table
1363 * Called with the sg->guest_table_lock
1365 static void __gmap_unshadow_sgt(struct gmap *sg, unsigned long raddr,
1372 BUG_ON(!gmap_is_shadow(sg));
1373 for (i = 0; i < _CRST_ENTRIES; i++, raddr += _SEGMENT_SIZE) {
1374 if (!(sgt[i] & _SEGMENT_ENTRY_ORIGIN))
1376 pgt = sgt[i] & _REGION_ENTRY_ORIGIN;
1377 sgt[i] = _SEGMENT_ENTRY_EMPTY;
1378 __gmap_unshadow_pgt(sg, raddr, __va(pgt));
1379 /* Free page table */
1380 page = phys_to_page(pgt);
1381 list_del(&page->lru);
1382 page_table_free_pgste(page);
1387 * gmap_unshadow_sgt - remove a shadow segment table from a region-3 entry
1388 * @sg: pointer to the shadow guest address space structure
1389 * @raddr: rmap address in the shadow guest address space
1391 * Called with the shadow->guest_table_lock
1393 static void gmap_unshadow_sgt(struct gmap *sg, unsigned long raddr)
1395 unsigned long r3o, *r3e;
1399 BUG_ON(!gmap_is_shadow(sg));
1400 r3e = gmap_table_walk(sg, raddr, 2); /* get region-3 pointer */
1401 if (!r3e || !(*r3e & _REGION_ENTRY_ORIGIN))
1403 gmap_call_notifier(sg, raddr, raddr + _REGION3_SIZE - 1);
1404 r3o = (unsigned long) (r3e - ((raddr & _REGION3_INDEX) >> _REGION3_SHIFT));
1405 gmap_idte_one(__pa(r3o) | _ASCE_TYPE_REGION3, raddr);
1406 sgt = *r3e & _REGION_ENTRY_ORIGIN;
1407 *r3e = _REGION3_ENTRY_EMPTY;
1408 __gmap_unshadow_sgt(sg, raddr, __va(sgt));
1409 /* Free segment table */
1410 page = phys_to_page(sgt);
1411 list_del(&page->lru);
1412 __free_pages(page, CRST_ALLOC_ORDER);
1416 * __gmap_unshadow_r3t - remove all entries from a shadow region-3 table
1417 * @sg: pointer to the shadow guest address space structure
1418 * @raddr: address in the shadow guest address space
1419 * @r3t: pointer to the start of a shadow region-3 table
1421 * Called with the sg->guest_table_lock
1423 static void __gmap_unshadow_r3t(struct gmap *sg, unsigned long raddr,
1430 BUG_ON(!gmap_is_shadow(sg));
1431 for (i = 0; i < _CRST_ENTRIES; i++, raddr += _REGION3_SIZE) {
1432 if (!(r3t[i] & _REGION_ENTRY_ORIGIN))
1434 sgt = r3t[i] & _REGION_ENTRY_ORIGIN;
1435 r3t[i] = _REGION3_ENTRY_EMPTY;
1436 __gmap_unshadow_sgt(sg, raddr, __va(sgt));
1437 /* Free segment table */
1438 page = phys_to_page(sgt);
1439 list_del(&page->lru);
1440 __free_pages(page, CRST_ALLOC_ORDER);
1445 * gmap_unshadow_r3t - remove a shadow region-3 table from a region-2 entry
1446 * @sg: pointer to the shadow guest address space structure
1447 * @raddr: rmap address in the shadow guest address space
1449 * Called with the sg->guest_table_lock
1451 static void gmap_unshadow_r3t(struct gmap *sg, unsigned long raddr)
1453 unsigned long r2o, *r2e;
1457 BUG_ON(!gmap_is_shadow(sg));
1458 r2e = gmap_table_walk(sg, raddr, 3); /* get region-2 pointer */
1459 if (!r2e || !(*r2e & _REGION_ENTRY_ORIGIN))
1461 gmap_call_notifier(sg, raddr, raddr + _REGION2_SIZE - 1);
1462 r2o = (unsigned long) (r2e - ((raddr & _REGION2_INDEX) >> _REGION2_SHIFT));
1463 gmap_idte_one(__pa(r2o) | _ASCE_TYPE_REGION2, raddr);
1464 r3t = *r2e & _REGION_ENTRY_ORIGIN;
1465 *r2e = _REGION2_ENTRY_EMPTY;
1466 __gmap_unshadow_r3t(sg, raddr, __va(r3t));
1467 /* Free region 3 table */
1468 page = phys_to_page(r3t);
1469 list_del(&page->lru);
1470 __free_pages(page, CRST_ALLOC_ORDER);
1474 * __gmap_unshadow_r2t - remove all entries from a shadow region-2 table
1475 * @sg: pointer to the shadow guest address space structure
1476 * @raddr: rmap address in the shadow guest address space
1477 * @r2t: pointer to the start of a shadow region-2 table
1479 * Called with the sg->guest_table_lock
1481 static void __gmap_unshadow_r2t(struct gmap *sg, unsigned long raddr,
1488 BUG_ON(!gmap_is_shadow(sg));
1489 for (i = 0; i < _CRST_ENTRIES; i++, raddr += _REGION2_SIZE) {
1490 if (!(r2t[i] & _REGION_ENTRY_ORIGIN))
1492 r3t = r2t[i] & _REGION_ENTRY_ORIGIN;
1493 r2t[i] = _REGION2_ENTRY_EMPTY;
1494 __gmap_unshadow_r3t(sg, raddr, __va(r3t));
1495 /* Free region 3 table */
1496 page = phys_to_page(r3t);
1497 list_del(&page->lru);
1498 __free_pages(page, CRST_ALLOC_ORDER);
1503 * gmap_unshadow_r2t - remove a shadow region-2 table from a region-1 entry
1504 * @sg: pointer to the shadow guest address space structure
1505 * @raddr: rmap address in the shadow guest address space
1507 * Called with the sg->guest_table_lock
1509 static void gmap_unshadow_r2t(struct gmap *sg, unsigned long raddr)
1511 unsigned long r1o, *r1e;
1515 BUG_ON(!gmap_is_shadow(sg));
1516 r1e = gmap_table_walk(sg, raddr, 4); /* get region-1 pointer */
1517 if (!r1e || !(*r1e & _REGION_ENTRY_ORIGIN))
1519 gmap_call_notifier(sg, raddr, raddr + _REGION1_SIZE - 1);
1520 r1o = (unsigned long) (r1e - ((raddr & _REGION1_INDEX) >> _REGION1_SHIFT));
1521 gmap_idte_one(__pa(r1o) | _ASCE_TYPE_REGION1, raddr);
1522 r2t = *r1e & _REGION_ENTRY_ORIGIN;
1523 *r1e = _REGION1_ENTRY_EMPTY;
1524 __gmap_unshadow_r2t(sg, raddr, __va(r2t));
1525 /* Free region 2 table */
1526 page = phys_to_page(r2t);
1527 list_del(&page->lru);
1528 __free_pages(page, CRST_ALLOC_ORDER);
1532 * __gmap_unshadow_r1t - remove all entries from a shadow region-1 table
1533 * @sg: pointer to the shadow guest address space structure
1534 * @raddr: rmap address in the shadow guest address space
1535 * @r1t: pointer to the start of a shadow region-1 table
1537 * Called with the shadow->guest_table_lock
1539 static void __gmap_unshadow_r1t(struct gmap *sg, unsigned long raddr,
1547 BUG_ON(!gmap_is_shadow(sg));
1548 asce = __pa(r1t) | _ASCE_TYPE_REGION1;
1549 for (i = 0; i < _CRST_ENTRIES; i++, raddr += _REGION1_SIZE) {
1550 if (!(r1t[i] & _REGION_ENTRY_ORIGIN))
1552 r2t = r1t[i] & _REGION_ENTRY_ORIGIN;
1553 __gmap_unshadow_r2t(sg, raddr, __va(r2t));
1554 /* Clear entry and flush translation r1t -> r2t */
1555 gmap_idte_one(asce, raddr);
1556 r1t[i] = _REGION1_ENTRY_EMPTY;
1557 /* Free region 2 table */
1558 page = phys_to_page(r2t);
1559 list_del(&page->lru);
1560 __free_pages(page, CRST_ALLOC_ORDER);
1565 * gmap_unshadow - remove a shadow page table completely
1566 * @sg: pointer to the shadow guest address space structure
1568 * Called with sg->guest_table_lock
1570 static void gmap_unshadow(struct gmap *sg)
1572 unsigned long *table;
1574 BUG_ON(!gmap_is_shadow(sg));
1578 gmap_call_notifier(sg, 0, -1UL);
1580 table = __va(sg->asce & _ASCE_ORIGIN);
1581 switch (sg->asce & _ASCE_TYPE_MASK) {
1582 case _ASCE_TYPE_REGION1:
1583 __gmap_unshadow_r1t(sg, 0, table);
1585 case _ASCE_TYPE_REGION2:
1586 __gmap_unshadow_r2t(sg, 0, table);
1588 case _ASCE_TYPE_REGION3:
1589 __gmap_unshadow_r3t(sg, 0, table);
1591 case _ASCE_TYPE_SEGMENT:
1592 __gmap_unshadow_sgt(sg, 0, table);
1598 * gmap_find_shadow - find a specific asce in the list of shadow tables
1599 * @parent: pointer to the parent gmap
1600 * @asce: ASCE for which the shadow table is created
1601 * @edat_level: edat level to be used for the shadow translation
1603 * Returns the pointer to a gmap if a shadow table with the given asce is
1604 * already available, ERR_PTR(-EAGAIN) if another one is just being created,
1607 static struct gmap *gmap_find_shadow(struct gmap *parent, unsigned long asce,
1612 list_for_each_entry(sg, &parent->children, list) {
1613 if (sg->orig_asce != asce || sg->edat_level != edat_level ||
1616 if (!sg->initialized)
1617 return ERR_PTR(-EAGAIN);
1618 refcount_inc(&sg->ref_count);
1625 * gmap_shadow_valid - check if a shadow guest address space matches the
1626 * given properties and is still valid
1627 * @sg: pointer to the shadow guest address space structure
1628 * @asce: ASCE for which the shadow table is requested
1629 * @edat_level: edat level to be used for the shadow translation
1631 * Returns 1 if the gmap shadow is still valid and matches the given
1632 * properties, the caller can continue using it. Returns 0 otherwise, the
1633 * caller has to request a new shadow gmap in this case.
1636 int gmap_shadow_valid(struct gmap *sg, unsigned long asce, int edat_level)
1640 return sg->orig_asce == asce && sg->edat_level == edat_level;
1642 EXPORT_SYMBOL_GPL(gmap_shadow_valid);
1645 * gmap_shadow - create/find a shadow guest address space
1646 * @parent: pointer to the parent gmap
1647 * @asce: ASCE for which the shadow table is created
1648 * @edat_level: edat level to be used for the shadow translation
1650 * The pages of the top level page table referred by the asce parameter
1651 * will be set to read-only and marked in the PGSTEs of the kvm process.
1652 * The shadow table will be removed automatically on any change to the
1653 * PTE mapping for the source table.
1655 * Returns a guest address space structure, ERR_PTR(-ENOMEM) if out of memory,
1656 * ERR_PTR(-EAGAIN) if the caller has to retry and ERR_PTR(-EFAULT) if the
1657 * parent gmap table could not be protected.
1659 struct gmap *gmap_shadow(struct gmap *parent, unsigned long asce,
1662 struct gmap *sg, *new;
1663 unsigned long limit;
1666 BUG_ON(parent->mm->context.allow_gmap_hpage_1m);
1667 BUG_ON(gmap_is_shadow(parent));
1668 spin_lock(&parent->shadow_lock);
1669 sg = gmap_find_shadow(parent, asce, edat_level);
1670 spin_unlock(&parent->shadow_lock);
1673 /* Create a new shadow gmap */
1674 limit = -1UL >> (33 - (((asce & _ASCE_TYPE_MASK) >> 2) * 11));
1675 if (asce & _ASCE_REAL_SPACE)
1677 new = gmap_alloc(limit);
1679 return ERR_PTR(-ENOMEM);
1680 new->mm = parent->mm;
1681 new->parent = gmap_get(parent);
1682 new->orig_asce = asce;
1683 new->edat_level = edat_level;
1684 new->initialized = false;
1685 spin_lock(&parent->shadow_lock);
1686 /* Recheck if another CPU created the same shadow */
1687 sg = gmap_find_shadow(parent, asce, edat_level);
1689 spin_unlock(&parent->shadow_lock);
1693 if (asce & _ASCE_REAL_SPACE) {
1694 /* only allow one real-space gmap shadow */
1695 list_for_each_entry(sg, &parent->children, list) {
1696 if (sg->orig_asce & _ASCE_REAL_SPACE) {
1697 spin_lock(&sg->guest_table_lock);
1699 spin_unlock(&sg->guest_table_lock);
1700 list_del(&sg->list);
1706 refcount_set(&new->ref_count, 2);
1707 list_add(&new->list, &parent->children);
1708 if (asce & _ASCE_REAL_SPACE) {
1709 /* nothing to protect, return right away */
1710 new->initialized = true;
1711 spin_unlock(&parent->shadow_lock);
1714 spin_unlock(&parent->shadow_lock);
1715 /* protect after insertion, so it will get properly invalidated */
1716 mmap_read_lock(parent->mm);
1717 rc = gmap_protect_range(parent, asce & _ASCE_ORIGIN,
1718 ((asce & _ASCE_TABLE_LENGTH) + 1) * PAGE_SIZE,
1719 PROT_READ, GMAP_NOTIFY_SHADOW);
1720 mmap_read_unlock(parent->mm);
1721 spin_lock(&parent->shadow_lock);
1722 new->initialized = true;
1724 list_del(&new->list);
1728 spin_unlock(&parent->shadow_lock);
1731 EXPORT_SYMBOL_GPL(gmap_shadow);
1734 * gmap_shadow_r2t - create an empty shadow region 2 table
1735 * @sg: pointer to the shadow guest address space structure
1736 * @saddr: faulting address in the shadow gmap
1737 * @r2t: parent gmap address of the region 2 table to get shadowed
1738 * @fake: r2t references contiguous guest memory block, not a r2t
1740 * The r2t parameter specifies the address of the source table. The
1741 * four pages of the source table are made read-only in the parent gmap
1742 * address space. A write to the source table area @r2t will automatically
1743 * remove the shadow r2 table and all of its decendents.
1745 * Returns 0 if successfully shadowed or already shadowed, -EAGAIN if the
1746 * shadow table structure is incomplete, -ENOMEM if out of memory and
1747 * -EFAULT if an address in the parent gmap could not be resolved.
1749 * Called with sg->mm->mmap_lock in read.
1751 int gmap_shadow_r2t(struct gmap *sg, unsigned long saddr, unsigned long r2t,
1754 unsigned long raddr, origin, offset, len;
1755 unsigned long *table;
1760 BUG_ON(!gmap_is_shadow(sg));
1761 /* Allocate a shadow region second table */
1762 page = alloc_pages(GFP_KERNEL_ACCOUNT, CRST_ALLOC_ORDER);
1765 page->index = r2t & _REGION_ENTRY_ORIGIN;
1767 page->index |= GMAP_SHADOW_FAKE_TABLE;
1768 s_r2t = page_to_phys(page);
1769 /* Install shadow region second table */
1770 spin_lock(&sg->guest_table_lock);
1771 table = gmap_table_walk(sg, saddr, 4); /* get region-1 pointer */
1773 rc = -EAGAIN; /* Race with unshadow */
1776 if (!(*table & _REGION_ENTRY_INVALID)) {
1777 rc = 0; /* Already established */
1779 } else if (*table & _REGION_ENTRY_ORIGIN) {
1780 rc = -EAGAIN; /* Race with shadow */
1783 crst_table_init(__va(s_r2t), _REGION2_ENTRY_EMPTY);
1784 /* mark as invalid as long as the parent table is not protected */
1785 *table = s_r2t | _REGION_ENTRY_LENGTH |
1786 _REGION_ENTRY_TYPE_R1 | _REGION_ENTRY_INVALID;
1787 if (sg->edat_level >= 1)
1788 *table |= (r2t & _REGION_ENTRY_PROTECT);
1789 list_add(&page->lru, &sg->crst_list);
1791 /* nothing to protect for fake tables */
1792 *table &= ~_REGION_ENTRY_INVALID;
1793 spin_unlock(&sg->guest_table_lock);
1796 spin_unlock(&sg->guest_table_lock);
1797 /* Make r2t read-only in parent gmap page table */
1798 raddr = (saddr & _REGION1_MASK) | _SHADOW_RMAP_REGION1;
1799 origin = r2t & _REGION_ENTRY_ORIGIN;
1800 offset = ((r2t & _REGION_ENTRY_OFFSET) >> 6) * PAGE_SIZE;
1801 len = ((r2t & _REGION_ENTRY_LENGTH) + 1) * PAGE_SIZE - offset;
1802 rc = gmap_protect_rmap(sg, raddr, origin + offset, len);
1803 spin_lock(&sg->guest_table_lock);
1805 table = gmap_table_walk(sg, saddr, 4);
1806 if (!table || (*table & _REGION_ENTRY_ORIGIN) != s_r2t)
1807 rc = -EAGAIN; /* Race with unshadow */
1809 *table &= ~_REGION_ENTRY_INVALID;
1811 gmap_unshadow_r2t(sg, raddr);
1813 spin_unlock(&sg->guest_table_lock);
1816 spin_unlock(&sg->guest_table_lock);
1817 __free_pages(page, CRST_ALLOC_ORDER);
1820 EXPORT_SYMBOL_GPL(gmap_shadow_r2t);
1823 * gmap_shadow_r3t - create a shadow region 3 table
1824 * @sg: pointer to the shadow guest address space structure
1825 * @saddr: faulting address in the shadow gmap
1826 * @r3t: parent gmap address of the region 3 table to get shadowed
1827 * @fake: r3t references contiguous guest memory block, not a r3t
1829 * Returns 0 if successfully shadowed or already shadowed, -EAGAIN if the
1830 * shadow table structure is incomplete, -ENOMEM if out of memory and
1831 * -EFAULT if an address in the parent gmap could not be resolved.
1833 * Called with sg->mm->mmap_lock in read.
1835 int gmap_shadow_r3t(struct gmap *sg, unsigned long saddr, unsigned long r3t,
1838 unsigned long raddr, origin, offset, len;
1839 unsigned long *table;
1844 BUG_ON(!gmap_is_shadow(sg));
1845 /* Allocate a shadow region second table */
1846 page = alloc_pages(GFP_KERNEL_ACCOUNT, CRST_ALLOC_ORDER);
1849 page->index = r3t & _REGION_ENTRY_ORIGIN;
1851 page->index |= GMAP_SHADOW_FAKE_TABLE;
1852 s_r3t = page_to_phys(page);
1853 /* Install shadow region second table */
1854 spin_lock(&sg->guest_table_lock);
1855 table = gmap_table_walk(sg, saddr, 3); /* get region-2 pointer */
1857 rc = -EAGAIN; /* Race with unshadow */
1860 if (!(*table & _REGION_ENTRY_INVALID)) {
1861 rc = 0; /* Already established */
1863 } else if (*table & _REGION_ENTRY_ORIGIN) {
1864 rc = -EAGAIN; /* Race with shadow */
1867 crst_table_init(__va(s_r3t), _REGION3_ENTRY_EMPTY);
1868 /* mark as invalid as long as the parent table is not protected */
1869 *table = s_r3t | _REGION_ENTRY_LENGTH |
1870 _REGION_ENTRY_TYPE_R2 | _REGION_ENTRY_INVALID;
1871 if (sg->edat_level >= 1)
1872 *table |= (r3t & _REGION_ENTRY_PROTECT);
1873 list_add(&page->lru, &sg->crst_list);
1875 /* nothing to protect for fake tables */
1876 *table &= ~_REGION_ENTRY_INVALID;
1877 spin_unlock(&sg->guest_table_lock);
1880 spin_unlock(&sg->guest_table_lock);
1881 /* Make r3t read-only in parent gmap page table */
1882 raddr = (saddr & _REGION2_MASK) | _SHADOW_RMAP_REGION2;
1883 origin = r3t & _REGION_ENTRY_ORIGIN;
1884 offset = ((r3t & _REGION_ENTRY_OFFSET) >> 6) * PAGE_SIZE;
1885 len = ((r3t & _REGION_ENTRY_LENGTH) + 1) * PAGE_SIZE - offset;
1886 rc = gmap_protect_rmap(sg, raddr, origin + offset, len);
1887 spin_lock(&sg->guest_table_lock);
1889 table = gmap_table_walk(sg, saddr, 3);
1890 if (!table || (*table & _REGION_ENTRY_ORIGIN) != s_r3t)
1891 rc = -EAGAIN; /* Race with unshadow */
1893 *table &= ~_REGION_ENTRY_INVALID;
1895 gmap_unshadow_r3t(sg, raddr);
1897 spin_unlock(&sg->guest_table_lock);
1900 spin_unlock(&sg->guest_table_lock);
1901 __free_pages(page, CRST_ALLOC_ORDER);
1904 EXPORT_SYMBOL_GPL(gmap_shadow_r3t);
1907 * gmap_shadow_sgt - create a shadow segment table
1908 * @sg: pointer to the shadow guest address space structure
1909 * @saddr: faulting address in the shadow gmap
1910 * @sgt: parent gmap address of the segment table to get shadowed
1911 * @fake: sgt references contiguous guest memory block, not a sgt
1913 * Returns: 0 if successfully shadowed or already shadowed, -EAGAIN if the
1914 * shadow table structure is incomplete, -ENOMEM if out of memory and
1915 * -EFAULT if an address in the parent gmap could not be resolved.
1917 * Called with sg->mm->mmap_lock in read.
1919 int gmap_shadow_sgt(struct gmap *sg, unsigned long saddr, unsigned long sgt,
1922 unsigned long raddr, origin, offset, len;
1923 unsigned long *table;
1928 BUG_ON(!gmap_is_shadow(sg) || (sgt & _REGION3_ENTRY_LARGE));
1929 /* Allocate a shadow segment table */
1930 page = alloc_pages(GFP_KERNEL_ACCOUNT, CRST_ALLOC_ORDER);
1933 page->index = sgt & _REGION_ENTRY_ORIGIN;
1935 page->index |= GMAP_SHADOW_FAKE_TABLE;
1936 s_sgt = page_to_phys(page);
1937 /* Install shadow region second table */
1938 spin_lock(&sg->guest_table_lock);
1939 table = gmap_table_walk(sg, saddr, 2); /* get region-3 pointer */
1941 rc = -EAGAIN; /* Race with unshadow */
1944 if (!(*table & _REGION_ENTRY_INVALID)) {
1945 rc = 0; /* Already established */
1947 } else if (*table & _REGION_ENTRY_ORIGIN) {
1948 rc = -EAGAIN; /* Race with shadow */
1951 crst_table_init(__va(s_sgt), _SEGMENT_ENTRY_EMPTY);
1952 /* mark as invalid as long as the parent table is not protected */
1953 *table = s_sgt | _REGION_ENTRY_LENGTH |
1954 _REGION_ENTRY_TYPE_R3 | _REGION_ENTRY_INVALID;
1955 if (sg->edat_level >= 1)
1956 *table |= sgt & _REGION_ENTRY_PROTECT;
1957 list_add(&page->lru, &sg->crst_list);
1959 /* nothing to protect for fake tables */
1960 *table &= ~_REGION_ENTRY_INVALID;
1961 spin_unlock(&sg->guest_table_lock);
1964 spin_unlock(&sg->guest_table_lock);
1965 /* Make sgt read-only in parent gmap page table */
1966 raddr = (saddr & _REGION3_MASK) | _SHADOW_RMAP_REGION3;
1967 origin = sgt & _REGION_ENTRY_ORIGIN;
1968 offset = ((sgt & _REGION_ENTRY_OFFSET) >> 6) * PAGE_SIZE;
1969 len = ((sgt & _REGION_ENTRY_LENGTH) + 1) * PAGE_SIZE - offset;
1970 rc = gmap_protect_rmap(sg, raddr, origin + offset, len);
1971 spin_lock(&sg->guest_table_lock);
1973 table = gmap_table_walk(sg, saddr, 2);
1974 if (!table || (*table & _REGION_ENTRY_ORIGIN) != s_sgt)
1975 rc = -EAGAIN; /* Race with unshadow */
1977 *table &= ~_REGION_ENTRY_INVALID;
1979 gmap_unshadow_sgt(sg, raddr);
1981 spin_unlock(&sg->guest_table_lock);
1984 spin_unlock(&sg->guest_table_lock);
1985 __free_pages(page, CRST_ALLOC_ORDER);
1988 EXPORT_SYMBOL_GPL(gmap_shadow_sgt);
1991 * gmap_shadow_pgt_lookup - find a shadow page table
1992 * @sg: pointer to the shadow guest address space structure
1993 * @saddr: the address in the shadow aguest address space
1994 * @pgt: parent gmap address of the page table to get shadowed
1995 * @dat_protection: if the pgtable is marked as protected by dat
1996 * @fake: pgt references contiguous guest memory block, not a pgtable
1998 * Returns 0 if the shadow page table was found and -EAGAIN if the page
1999 * table was not found.
2001 * Called with sg->mm->mmap_lock in read.
2003 int gmap_shadow_pgt_lookup(struct gmap *sg, unsigned long saddr,
2004 unsigned long *pgt, int *dat_protection,
2007 unsigned long *table;
2011 BUG_ON(!gmap_is_shadow(sg));
2012 spin_lock(&sg->guest_table_lock);
2013 table = gmap_table_walk(sg, saddr, 1); /* get segment pointer */
2014 if (table && !(*table & _SEGMENT_ENTRY_INVALID)) {
2015 /* Shadow page tables are full pages (pte+pgste) */
2016 page = pfn_to_page(*table >> PAGE_SHIFT);
2017 *pgt = page->index & ~GMAP_SHADOW_FAKE_TABLE;
2018 *dat_protection = !!(*table & _SEGMENT_ENTRY_PROTECT);
2019 *fake = !!(page->index & GMAP_SHADOW_FAKE_TABLE);
2024 spin_unlock(&sg->guest_table_lock);
2028 EXPORT_SYMBOL_GPL(gmap_shadow_pgt_lookup);
2031 * gmap_shadow_pgt - instantiate a shadow page table
2032 * @sg: pointer to the shadow guest address space structure
2033 * @saddr: faulting address in the shadow gmap
2034 * @pgt: parent gmap address of the page table to get shadowed
2035 * @fake: pgt references contiguous guest memory block, not a pgtable
2037 * Returns 0 if successfully shadowed or already shadowed, -EAGAIN if the
2038 * shadow table structure is incomplete, -ENOMEM if out of memory,
2039 * -EFAULT if an address in the parent gmap could not be resolved and
2041 * Called with gmap->mm->mmap_lock in read
2043 int gmap_shadow_pgt(struct gmap *sg, unsigned long saddr, unsigned long pgt,
2046 unsigned long raddr, origin;
2047 unsigned long *table;
2052 BUG_ON(!gmap_is_shadow(sg) || (pgt & _SEGMENT_ENTRY_LARGE));
2053 /* Allocate a shadow page table */
2054 page = page_table_alloc_pgste(sg->mm);
2057 page->index = pgt & _SEGMENT_ENTRY_ORIGIN;
2059 page->index |= GMAP_SHADOW_FAKE_TABLE;
2060 s_pgt = page_to_phys(page);
2061 /* Install shadow page table */
2062 spin_lock(&sg->guest_table_lock);
2063 table = gmap_table_walk(sg, saddr, 1); /* get segment pointer */
2065 rc = -EAGAIN; /* Race with unshadow */
2068 if (!(*table & _SEGMENT_ENTRY_INVALID)) {
2069 rc = 0; /* Already established */
2071 } else if (*table & _SEGMENT_ENTRY_ORIGIN) {
2072 rc = -EAGAIN; /* Race with shadow */
2075 /* mark as invalid as long as the parent table is not protected */
2076 *table = (unsigned long) s_pgt | _SEGMENT_ENTRY |
2077 (pgt & _SEGMENT_ENTRY_PROTECT) | _SEGMENT_ENTRY_INVALID;
2078 list_add(&page->lru, &sg->pt_list);
2080 /* nothing to protect for fake tables */
2081 *table &= ~_SEGMENT_ENTRY_INVALID;
2082 spin_unlock(&sg->guest_table_lock);
2085 spin_unlock(&sg->guest_table_lock);
2086 /* Make pgt read-only in parent gmap page table (not the pgste) */
2087 raddr = (saddr & _SEGMENT_MASK) | _SHADOW_RMAP_SEGMENT;
2088 origin = pgt & _SEGMENT_ENTRY_ORIGIN & PAGE_MASK;
2089 rc = gmap_protect_rmap(sg, raddr, origin, PAGE_SIZE);
2090 spin_lock(&sg->guest_table_lock);
2092 table = gmap_table_walk(sg, saddr, 1);
2093 if (!table || (*table & _SEGMENT_ENTRY_ORIGIN) != s_pgt)
2094 rc = -EAGAIN; /* Race with unshadow */
2096 *table &= ~_SEGMENT_ENTRY_INVALID;
2098 gmap_unshadow_pgt(sg, raddr);
2100 spin_unlock(&sg->guest_table_lock);
2103 spin_unlock(&sg->guest_table_lock);
2104 page_table_free_pgste(page);
2108 EXPORT_SYMBOL_GPL(gmap_shadow_pgt);
2111 * gmap_shadow_page - create a shadow page mapping
2112 * @sg: pointer to the shadow guest address space structure
2113 * @saddr: faulting address in the shadow gmap
2114 * @pte: pte in parent gmap address space to get shadowed
2116 * Returns 0 if successfully shadowed or already shadowed, -EAGAIN if the
2117 * shadow table structure is incomplete, -ENOMEM if out of memory and
2118 * -EFAULT if an address in the parent gmap could not be resolved.
2120 * Called with sg->mm->mmap_lock in read.
2122 int gmap_shadow_page(struct gmap *sg, unsigned long saddr, pte_t pte)
2124 struct gmap *parent;
2125 struct gmap_rmap *rmap;
2126 unsigned long vmaddr, paddr;
2128 pte_t *sptep, *tptep;
2132 BUG_ON(!gmap_is_shadow(sg));
2133 parent = sg->parent;
2134 prot = (pte_val(pte) & _PAGE_PROTECT) ? PROT_READ : PROT_WRITE;
2136 rmap = kzalloc(sizeof(*rmap), GFP_KERNEL_ACCOUNT);
2139 rmap->raddr = (saddr & PAGE_MASK) | _SHADOW_RMAP_PGTABLE;
2142 paddr = pte_val(pte) & PAGE_MASK;
2143 vmaddr = __gmap_translate(parent, paddr);
2144 if (IS_ERR_VALUE(vmaddr)) {
2148 rc = radix_tree_preload(GFP_KERNEL_ACCOUNT);
2152 sptep = gmap_pte_op_walk(parent, paddr, &ptl);
2154 spin_lock(&sg->guest_table_lock);
2155 /* Get page table pointer */
2156 tptep = (pte_t *) gmap_table_walk(sg, saddr, 0);
2158 spin_unlock(&sg->guest_table_lock);
2159 gmap_pte_op_end(ptl);
2160 radix_tree_preload_end();
2163 rc = ptep_shadow_pte(sg->mm, saddr, sptep, tptep, pte);
2165 /* Success and a new mapping */
2166 gmap_insert_rmap(sg, vmaddr, rmap);
2170 gmap_pte_op_end(ptl);
2171 spin_unlock(&sg->guest_table_lock);
2173 radix_tree_preload_end();
2176 rc = gmap_pte_op_fixup(parent, paddr, vmaddr, prot);
2183 EXPORT_SYMBOL_GPL(gmap_shadow_page);
2186 * gmap_shadow_notify - handle notifications for shadow gmap
2188 * Called with sg->parent->shadow_lock.
2190 static void gmap_shadow_notify(struct gmap *sg, unsigned long vmaddr,
2191 unsigned long gaddr)
2193 struct gmap_rmap *rmap, *rnext, *head;
2194 unsigned long start, end, bits, raddr;
2196 BUG_ON(!gmap_is_shadow(sg));
2198 spin_lock(&sg->guest_table_lock);
2200 spin_unlock(&sg->guest_table_lock);
2203 /* Check for top level table */
2204 start = sg->orig_asce & _ASCE_ORIGIN;
2205 end = start + ((sg->orig_asce & _ASCE_TABLE_LENGTH) + 1) * PAGE_SIZE;
2206 if (!(sg->orig_asce & _ASCE_REAL_SPACE) && gaddr >= start &&
2208 /* The complete shadow table has to go */
2210 spin_unlock(&sg->guest_table_lock);
2211 list_del(&sg->list);
2215 /* Remove the page table tree from on specific entry */
2216 head = radix_tree_delete(&sg->host_to_rmap, vmaddr >> PAGE_SHIFT);
2217 gmap_for_each_rmap_safe(rmap, rnext, head) {
2218 bits = rmap->raddr & _SHADOW_RMAP_MASK;
2219 raddr = rmap->raddr ^ bits;
2221 case _SHADOW_RMAP_REGION1:
2222 gmap_unshadow_r2t(sg, raddr);
2224 case _SHADOW_RMAP_REGION2:
2225 gmap_unshadow_r3t(sg, raddr);
2227 case _SHADOW_RMAP_REGION3:
2228 gmap_unshadow_sgt(sg, raddr);
2230 case _SHADOW_RMAP_SEGMENT:
2231 gmap_unshadow_pgt(sg, raddr);
2233 case _SHADOW_RMAP_PGTABLE:
2234 gmap_unshadow_page(sg, raddr);
2239 spin_unlock(&sg->guest_table_lock);
2243 * ptep_notify - call all invalidation callbacks for a specific pte.
2244 * @mm: pointer to the process mm_struct
2245 * @vmaddr: virtual address in the process address space
2246 * @pte: pointer to the page table entry
2247 * @bits: bits from the pgste that caused the notify call
2249 * This function is assumed to be called with the page table lock held
2250 * for the pte to notify.
2252 void ptep_notify(struct mm_struct *mm, unsigned long vmaddr,
2253 pte_t *pte, unsigned long bits)
2255 unsigned long offset, gaddr = 0;
2256 unsigned long *table;
2257 struct gmap *gmap, *sg, *next;
2259 offset = ((unsigned long) pte) & (255 * sizeof(pte_t));
2260 offset = offset * (PAGE_SIZE / sizeof(pte_t));
2262 list_for_each_entry_rcu(gmap, &mm->context.gmap_list, list) {
2263 spin_lock(&gmap->guest_table_lock);
2264 table = radix_tree_lookup(&gmap->host_to_guest,
2265 vmaddr >> PMD_SHIFT);
2267 gaddr = __gmap_segment_gaddr(table) + offset;
2268 spin_unlock(&gmap->guest_table_lock);
2272 if (!list_empty(&gmap->children) && (bits & PGSTE_VSIE_BIT)) {
2273 spin_lock(&gmap->shadow_lock);
2274 list_for_each_entry_safe(sg, next,
2275 &gmap->children, list)
2276 gmap_shadow_notify(sg, vmaddr, gaddr);
2277 spin_unlock(&gmap->shadow_lock);
2279 if (bits & PGSTE_IN_BIT)
2280 gmap_call_notifier(gmap, gaddr, gaddr + PAGE_SIZE - 1);
2284 EXPORT_SYMBOL_GPL(ptep_notify);
2286 static void pmdp_notify_gmap(struct gmap *gmap, pmd_t *pmdp,
2287 unsigned long gaddr)
2289 set_pmd(pmdp, clear_pmd_bit(*pmdp, __pgprot(_SEGMENT_ENTRY_GMAP_IN)));
2290 gmap_call_notifier(gmap, gaddr, gaddr + HPAGE_SIZE - 1);
2294 * gmap_pmdp_xchg - exchange a gmap pmd with another
2295 * @gmap: pointer to the guest address space structure
2296 * @pmdp: pointer to the pmd entry
2297 * @new: replacement entry
2298 * @gaddr: the affected guest address
2300 * This function is assumed to be called with the guest_table_lock
2303 static void gmap_pmdp_xchg(struct gmap *gmap, pmd_t *pmdp, pmd_t new,
2304 unsigned long gaddr)
2306 gaddr &= HPAGE_MASK;
2307 pmdp_notify_gmap(gmap, pmdp, gaddr);
2308 new = clear_pmd_bit(new, __pgprot(_SEGMENT_ENTRY_GMAP_IN));
2309 if (MACHINE_HAS_TLB_GUEST)
2310 __pmdp_idte(gaddr, (pmd_t *)pmdp, IDTE_GUEST_ASCE, gmap->asce,
2312 else if (MACHINE_HAS_IDTE)
2313 __pmdp_idte(gaddr, (pmd_t *)pmdp, 0, 0, IDTE_GLOBAL);
2319 static void gmap_pmdp_clear(struct mm_struct *mm, unsigned long vmaddr,
2324 unsigned long gaddr;
2327 list_for_each_entry_rcu(gmap, &mm->context.gmap_list, list) {
2328 spin_lock(&gmap->guest_table_lock);
2329 pmdp = (pmd_t *)radix_tree_delete(&gmap->host_to_guest,
2330 vmaddr >> PMD_SHIFT);
2332 gaddr = __gmap_segment_gaddr((unsigned long *)pmdp);
2333 pmdp_notify_gmap(gmap, pmdp, gaddr);
2334 WARN_ON(pmd_val(*pmdp) & ~(_SEGMENT_ENTRY_HARDWARE_BITS_LARGE |
2335 _SEGMENT_ENTRY_GMAP_UC));
2338 set_pmd(pmdp, __pmd(_SEGMENT_ENTRY_EMPTY));
2340 spin_unlock(&gmap->guest_table_lock);
2346 * gmap_pmdp_invalidate - invalidate all affected guest pmd entries without
2348 * @mm: pointer to the process mm_struct
2349 * @vmaddr: virtual address in the process address space
2351 void gmap_pmdp_invalidate(struct mm_struct *mm, unsigned long vmaddr)
2353 gmap_pmdp_clear(mm, vmaddr, 0);
2355 EXPORT_SYMBOL_GPL(gmap_pmdp_invalidate);
2358 * gmap_pmdp_csp - csp all affected guest pmd entries
2359 * @mm: pointer to the process mm_struct
2360 * @vmaddr: virtual address in the process address space
2362 void gmap_pmdp_csp(struct mm_struct *mm, unsigned long vmaddr)
2364 gmap_pmdp_clear(mm, vmaddr, 1);
2366 EXPORT_SYMBOL_GPL(gmap_pmdp_csp);
2369 * gmap_pmdp_idte_local - invalidate and clear a guest pmd entry
2370 * @mm: pointer to the process mm_struct
2371 * @vmaddr: virtual address in the process address space
2373 void gmap_pmdp_idte_local(struct mm_struct *mm, unsigned long vmaddr)
2375 unsigned long *entry, gaddr;
2380 list_for_each_entry_rcu(gmap, &mm->context.gmap_list, list) {
2381 spin_lock(&gmap->guest_table_lock);
2382 entry = radix_tree_delete(&gmap->host_to_guest,
2383 vmaddr >> PMD_SHIFT);
2385 pmdp = (pmd_t *)entry;
2386 gaddr = __gmap_segment_gaddr(entry);
2387 pmdp_notify_gmap(gmap, pmdp, gaddr);
2388 WARN_ON(*entry & ~(_SEGMENT_ENTRY_HARDWARE_BITS_LARGE |
2389 _SEGMENT_ENTRY_GMAP_UC));
2390 if (MACHINE_HAS_TLB_GUEST)
2391 __pmdp_idte(gaddr, pmdp, IDTE_GUEST_ASCE,
2392 gmap->asce, IDTE_LOCAL);
2393 else if (MACHINE_HAS_IDTE)
2394 __pmdp_idte(gaddr, pmdp, 0, 0, IDTE_LOCAL);
2395 *entry = _SEGMENT_ENTRY_EMPTY;
2397 spin_unlock(&gmap->guest_table_lock);
2401 EXPORT_SYMBOL_GPL(gmap_pmdp_idte_local);
2404 * gmap_pmdp_idte_global - invalidate and clear a guest pmd entry
2405 * @mm: pointer to the process mm_struct
2406 * @vmaddr: virtual address in the process address space
2408 void gmap_pmdp_idte_global(struct mm_struct *mm, unsigned long vmaddr)
2410 unsigned long *entry, gaddr;
2415 list_for_each_entry_rcu(gmap, &mm->context.gmap_list, list) {
2416 spin_lock(&gmap->guest_table_lock);
2417 entry = radix_tree_delete(&gmap->host_to_guest,
2418 vmaddr >> PMD_SHIFT);
2420 pmdp = (pmd_t *)entry;
2421 gaddr = __gmap_segment_gaddr(entry);
2422 pmdp_notify_gmap(gmap, pmdp, gaddr);
2423 WARN_ON(*entry & ~(_SEGMENT_ENTRY_HARDWARE_BITS_LARGE |
2424 _SEGMENT_ENTRY_GMAP_UC));
2425 if (MACHINE_HAS_TLB_GUEST)
2426 __pmdp_idte(gaddr, pmdp, IDTE_GUEST_ASCE,
2427 gmap->asce, IDTE_GLOBAL);
2428 else if (MACHINE_HAS_IDTE)
2429 __pmdp_idte(gaddr, pmdp, 0, 0, IDTE_GLOBAL);
2432 *entry = _SEGMENT_ENTRY_EMPTY;
2434 spin_unlock(&gmap->guest_table_lock);
2438 EXPORT_SYMBOL_GPL(gmap_pmdp_idte_global);
2441 * gmap_test_and_clear_dirty_pmd - test and reset segment dirty status
2442 * @gmap: pointer to guest address space
2443 * @pmdp: pointer to the pmd to be tested
2444 * @gaddr: virtual address in the guest address space
2446 * This function is assumed to be called with the guest_table_lock
2449 static bool gmap_test_and_clear_dirty_pmd(struct gmap *gmap, pmd_t *pmdp,
2450 unsigned long gaddr)
2452 if (pmd_val(*pmdp) & _SEGMENT_ENTRY_INVALID)
2455 /* Already protected memory, which did not change is clean */
2456 if (pmd_val(*pmdp) & _SEGMENT_ENTRY_PROTECT &&
2457 !(pmd_val(*pmdp) & _SEGMENT_ENTRY_GMAP_UC))
2460 /* Clear UC indication and reset protection */
2461 set_pmd(pmdp, clear_pmd_bit(*pmdp, __pgprot(_SEGMENT_ENTRY_GMAP_UC)));
2462 gmap_protect_pmd(gmap, gaddr, pmdp, PROT_READ, 0);
2467 * gmap_sync_dirty_log_pmd - set bitmap based on dirty status of segment
2468 * @gmap: pointer to guest address space
2469 * @bitmap: dirty bitmap for this pmd
2470 * @gaddr: virtual address in the guest address space
2471 * @vmaddr: virtual address in the host address space
2473 * This function is assumed to be called with the guest_table_lock
2476 void gmap_sync_dirty_log_pmd(struct gmap *gmap, unsigned long bitmap[4],
2477 unsigned long gaddr, unsigned long vmaddr)
2484 pmdp = gmap_pmd_op_walk(gmap, gaddr);
2488 if (pmd_large(*pmdp)) {
2489 if (gmap_test_and_clear_dirty_pmd(gmap, pmdp, gaddr))
2490 bitmap_fill(bitmap, _PAGE_ENTRIES);
2492 for (i = 0; i < _PAGE_ENTRIES; i++, vmaddr += PAGE_SIZE) {
2493 ptep = pte_alloc_map_lock(gmap->mm, pmdp, vmaddr, &ptl);
2496 if (ptep_test_and_clear_uc(gmap->mm, vmaddr, ptep))
2501 gmap_pmd_op_end(gmap, pmdp);
2503 EXPORT_SYMBOL_GPL(gmap_sync_dirty_log_pmd);
2505 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
2506 static int thp_split_walk_pmd_entry(pmd_t *pmd, unsigned long addr,
2507 unsigned long end, struct mm_walk *walk)
2509 struct vm_area_struct *vma = walk->vma;
2511 split_huge_pmd(vma, pmd, addr);
2515 static const struct mm_walk_ops thp_split_walk_ops = {
2516 .pmd_entry = thp_split_walk_pmd_entry,
2519 static inline void thp_split_mm(struct mm_struct *mm)
2521 struct vm_area_struct *vma;
2522 VMA_ITERATOR(vmi, mm, 0);
2524 for_each_vma(vmi, vma) {
2525 vm_flags_mod(vma, VM_NOHUGEPAGE, VM_HUGEPAGE);
2526 walk_page_vma(vma, &thp_split_walk_ops, NULL);
2528 mm->def_flags |= VM_NOHUGEPAGE;
2531 static inline void thp_split_mm(struct mm_struct *mm)
2534 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
2537 * Remove all empty zero pages from the mapping for lazy refaulting
2538 * - This must be called after mm->context.has_pgste is set, to avoid
2539 * future creation of zero pages
2540 * - This must be called after THP was enabled
2542 static int __zap_zero_pages(pmd_t *pmd, unsigned long start,
2543 unsigned long end, struct mm_walk *walk)
2547 for (addr = start; addr != end; addr += PAGE_SIZE) {
2551 ptep = pte_offset_map_lock(walk->mm, pmd, addr, &ptl);
2552 if (is_zero_pfn(pte_pfn(*ptep)))
2553 ptep_xchg_direct(walk->mm, addr, ptep, __pte(_PAGE_INVALID));
2554 pte_unmap_unlock(ptep, ptl);
2559 static const struct mm_walk_ops zap_zero_walk_ops = {
2560 .pmd_entry = __zap_zero_pages,
2564 * switch on pgstes for its userspace process (for kvm)
2566 int s390_enable_sie(void)
2568 struct mm_struct *mm = current->mm;
2570 /* Do we have pgstes? if yes, we are done */
2571 if (mm_has_pgste(mm))
2573 /* Fail if the page tables are 2K */
2574 if (!mm_alloc_pgste(mm))
2576 mmap_write_lock(mm);
2577 mm->context.has_pgste = 1;
2578 /* split thp mappings and disable thp for future mappings */
2580 walk_page_range(mm, 0, TASK_SIZE, &zap_zero_walk_ops, NULL);
2581 mmap_write_unlock(mm);
2584 EXPORT_SYMBOL_GPL(s390_enable_sie);
2586 int gmap_mark_unmergeable(void)
2588 struct mm_struct *mm = current->mm;
2589 struct vm_area_struct *vma;
2590 unsigned long vm_flags;
2592 VMA_ITERATOR(vmi, mm, 0);
2595 * Make sure to disable KSM (if enabled for the whole process or
2596 * individual VMAs). Note that nothing currently hinders user space
2597 * from re-enabling it.
2599 clear_bit(MMF_VM_MERGE_ANY, &mm->flags);
2601 for_each_vma(vmi, vma) {
2602 /* Copy vm_flags to avoid partial modifications in ksm_madvise */
2603 vm_flags = vma->vm_flags;
2604 ret = ksm_madvise(vma, vma->vm_start, vma->vm_end,
2605 MADV_UNMERGEABLE, &vm_flags);
2608 vm_flags_reset(vma, vm_flags);
2610 mm->def_flags &= ~VM_MERGEABLE;
2613 EXPORT_SYMBOL_GPL(gmap_mark_unmergeable);
2616 * Enable storage key handling from now on and initialize the storage
2617 * keys with the default key.
2619 static int __s390_enable_skey_pte(pte_t *pte, unsigned long addr,
2620 unsigned long next, struct mm_walk *walk)
2622 /* Clear storage key */
2623 ptep_zap_key(walk->mm, addr, pte);
2628 * Give a chance to schedule after setting a key to 256 pages.
2629 * We only hold the mm lock, which is a rwsem and the kvm srcu.
2632 static int __s390_enable_skey_pmd(pmd_t *pmd, unsigned long addr,
2633 unsigned long next, struct mm_walk *walk)
2639 static int __s390_enable_skey_hugetlb(pte_t *pte, unsigned long addr,
2640 unsigned long hmask, unsigned long next,
2641 struct mm_walk *walk)
2643 pmd_t *pmd = (pmd_t *)pte;
2644 unsigned long start, end;
2645 struct page *page = pmd_page(*pmd);
2648 * The write check makes sure we do not set a key on shared
2649 * memory. This is needed as the walker does not differentiate
2650 * between actual guest memory and the process executable or
2653 if (pmd_val(*pmd) & _SEGMENT_ENTRY_INVALID ||
2654 !(pmd_val(*pmd) & _SEGMENT_ENTRY_WRITE))
2657 start = pmd_val(*pmd) & HPAGE_MASK;
2658 end = start + HPAGE_SIZE - 1;
2659 __storage_key_init_range(start, end);
2660 set_bit(PG_arch_1, &page->flags);
2665 static const struct mm_walk_ops enable_skey_walk_ops = {
2666 .hugetlb_entry = __s390_enable_skey_hugetlb,
2667 .pte_entry = __s390_enable_skey_pte,
2668 .pmd_entry = __s390_enable_skey_pmd,
2671 int s390_enable_skey(void)
2673 struct mm_struct *mm = current->mm;
2676 mmap_write_lock(mm);
2677 if (mm_uses_skeys(mm))
2680 mm->context.uses_skeys = 1;
2681 rc = gmap_mark_unmergeable();
2683 mm->context.uses_skeys = 0;
2686 walk_page_range(mm, 0, TASK_SIZE, &enable_skey_walk_ops, NULL);
2689 mmap_write_unlock(mm);
2692 EXPORT_SYMBOL_GPL(s390_enable_skey);
2695 * Reset CMMA state, make all pages stable again.
2697 static int __s390_reset_cmma(pte_t *pte, unsigned long addr,
2698 unsigned long next, struct mm_walk *walk)
2700 ptep_zap_unused(walk->mm, addr, pte, 1);
2704 static const struct mm_walk_ops reset_cmma_walk_ops = {
2705 .pte_entry = __s390_reset_cmma,
2708 void s390_reset_cmma(struct mm_struct *mm)
2710 mmap_write_lock(mm);
2711 walk_page_range(mm, 0, TASK_SIZE, &reset_cmma_walk_ops, NULL);
2712 mmap_write_unlock(mm);
2714 EXPORT_SYMBOL_GPL(s390_reset_cmma);
2716 #define GATHER_GET_PAGES 32
2718 struct reset_walk_state {
2720 unsigned long count;
2721 unsigned long pfns[GATHER_GET_PAGES];
2724 static int s390_gather_pages(pte_t *ptep, unsigned long addr,
2725 unsigned long next, struct mm_walk *walk)
2727 struct reset_walk_state *p = walk->private;
2728 pte_t pte = READ_ONCE(*ptep);
2730 if (pte_present(pte)) {
2731 /* we have a reference from the mapping, take an extra one */
2732 get_page(phys_to_page(pte_val(pte)));
2733 p->pfns[p->count] = phys_to_pfn(pte_val(pte));
2737 return p->count >= GATHER_GET_PAGES;
2740 static const struct mm_walk_ops gather_pages_ops = {
2741 .pte_entry = s390_gather_pages,
2745 * Call the Destroy secure page UVC on each page in the given array of PFNs.
2746 * Each page needs to have an extra reference, which will be released here.
2748 void s390_uv_destroy_pfns(unsigned long count, unsigned long *pfns)
2752 for (i = 0; i < count; i++) {
2753 /* we always have an extra reference */
2754 uv_destroy_owned_page(pfn_to_phys(pfns[i]));
2755 /* get rid of the extra reference */
2756 put_page(pfn_to_page(pfns[i]));
2760 EXPORT_SYMBOL_GPL(s390_uv_destroy_pfns);
2763 * __s390_uv_destroy_range - Call the destroy secure page UVC on each page
2764 * in the given range of the given address space.
2765 * @mm: the mm to operate on
2766 * @start: the start of the range
2767 * @end: the end of the range
2768 * @interruptible: if not 0, stop when a fatal signal is received
2770 * Walk the given range of the given address space and call the destroy
2771 * secure page UVC on each page. Optionally exit early if a fatal signal is
2774 * Return: 0 on success, -EINTR if the function stopped before completing
2776 int __s390_uv_destroy_range(struct mm_struct *mm, unsigned long start,
2777 unsigned long end, bool interruptible)
2779 struct reset_walk_state state = { .next = start };
2785 r = walk_page_range(mm, state.next, end, &gather_pages_ops, &state);
2786 mmap_read_unlock(mm);
2788 s390_uv_destroy_pfns(state.count, state.pfns);
2789 if (interruptible && fatal_signal_pending(current))
2794 EXPORT_SYMBOL_GPL(__s390_uv_destroy_range);
2797 * s390_unlist_old_asce - Remove the topmost level of page tables from the
2798 * list of page tables of the gmap.
2799 * @gmap: the gmap whose table is to be removed
2801 * On s390x, KVM keeps a list of all pages containing the page tables of the
2802 * gmap (the CRST list). This list is used at tear down time to free all
2803 * pages that are now not needed anymore.
2805 * This function removes the topmost page of the tree (the one pointed to by
2806 * the ASCE) from the CRST list.
2808 * This means that it will not be freed when the VM is torn down, and needs
2809 * to be handled separately by the caller, unless a leak is actually
2810 * intended. Notice that this function will only remove the page from the
2811 * list, the page will still be used as a top level page table (and ASCE).
2813 void s390_unlist_old_asce(struct gmap *gmap)
2817 old = virt_to_page(gmap->table);
2818 spin_lock(&gmap->guest_table_lock);
2819 list_del(&old->lru);
2821 * Sometimes the topmost page might need to be "removed" multiple
2822 * times, for example if the VM is rebooted into secure mode several
2823 * times concurrently, or if s390_replace_asce fails after calling
2824 * s390_remove_old_asce and is attempted again later. In that case
2825 * the old asce has been removed from the list, and therefore it
2826 * will not be freed when the VM terminates, but the ASCE is still
2827 * in use and still pointed to.
2828 * A subsequent call to replace_asce will follow the pointer and try
2829 * to remove the same page from the list again.
2830 * Therefore it's necessary that the page of the ASCE has valid
2831 * pointers, so list_del can work (and do nothing) without
2832 * dereferencing stale or invalid pointers.
2834 INIT_LIST_HEAD(&old->lru);
2835 spin_unlock(&gmap->guest_table_lock);
2837 EXPORT_SYMBOL_GPL(s390_unlist_old_asce);
2840 * s390_replace_asce - Try to replace the current ASCE of a gmap with a copy
2841 * @gmap: the gmap whose ASCE needs to be replaced
2843 * If the allocation of the new top level page table fails, the ASCE is not
2845 * In any case, the old ASCE is always removed from the gmap CRST list.
2846 * Therefore the caller has to make sure to save a pointer to it
2847 * beforehand, unless a leak is actually intended.
2849 int s390_replace_asce(struct gmap *gmap)
2855 s390_unlist_old_asce(gmap);
2857 page = alloc_pages(GFP_KERNEL_ACCOUNT, CRST_ALLOC_ORDER);
2860 table = page_to_virt(page);
2861 memcpy(table, gmap->table, 1UL << (CRST_ALLOC_ORDER + PAGE_SHIFT));
2864 * The caller has to deal with the old ASCE, but here we make sure
2865 * the new one is properly added to the CRST list, so that
2866 * it will be freed when the VM is torn down.
2868 spin_lock(&gmap->guest_table_lock);
2869 list_add(&page->lru, &gmap->crst_list);
2870 spin_unlock(&gmap->guest_table_lock);
2872 /* Set new table origin while preserving existing ASCE control bits */
2873 asce = (gmap->asce & ~_ASCE_ORIGIN) | __pa(table);
2874 WRITE_ONCE(gmap->asce, asce);
2875 WRITE_ONCE(gmap->mm->context.gmap_asce, asce);
2876 WRITE_ONCE(gmap->table, table);
2880 EXPORT_SYMBOL_GPL(s390_replace_asce);