1 // SPDX-License-Identifier: GPL-2.0
3 * KVM guest address space mapping code
5 * Copyright IBM Corp. 2007, 2020
11 #include <linux/kernel.h>
12 #include <linux/pagewalk.h>
13 #include <linux/swap.h>
14 #include <linux/smp.h>
15 #include <linux/spinlock.h>
16 #include <linux/slab.h>
17 #include <linux/swapops.h>
18 #include <linux/ksm.h>
19 #include <linux/mman.h>
20 #include <linux/pgtable.h>
21 #include <asm/page-states.h>
22 #include <asm/pgalloc.h>
27 #define GMAP_SHADOW_FAKE_TABLE 1ULL
29 static struct page *gmap_alloc_crst(void)
33 page = alloc_pages(GFP_KERNEL_ACCOUNT, CRST_ALLOC_ORDER);
36 __arch_set_page_dat(page_to_virt(page), 1UL << CRST_ALLOC_ORDER);
41 * gmap_alloc - allocate and initialize a guest address space
42 * @limit: maximum address of the gmap address space
44 * Returns a guest address space structure.
46 static struct gmap *gmap_alloc(unsigned long limit)
51 unsigned long etype, atype;
53 if (limit < _REGION3_SIZE) {
54 limit = _REGION3_SIZE - 1;
55 atype = _ASCE_TYPE_SEGMENT;
56 etype = _SEGMENT_ENTRY_EMPTY;
57 } else if (limit < _REGION2_SIZE) {
58 limit = _REGION2_SIZE - 1;
59 atype = _ASCE_TYPE_REGION3;
60 etype = _REGION3_ENTRY_EMPTY;
61 } else if (limit < _REGION1_SIZE) {
62 limit = _REGION1_SIZE - 1;
63 atype = _ASCE_TYPE_REGION2;
64 etype = _REGION2_ENTRY_EMPTY;
67 atype = _ASCE_TYPE_REGION1;
68 etype = _REGION1_ENTRY_EMPTY;
70 gmap = kzalloc(sizeof(struct gmap), GFP_KERNEL_ACCOUNT);
73 INIT_LIST_HEAD(&gmap->crst_list);
74 INIT_LIST_HEAD(&gmap->children);
75 INIT_LIST_HEAD(&gmap->pt_list);
76 INIT_RADIX_TREE(&gmap->guest_to_host, GFP_KERNEL_ACCOUNT);
77 INIT_RADIX_TREE(&gmap->host_to_guest, GFP_ATOMIC | __GFP_ACCOUNT);
78 INIT_RADIX_TREE(&gmap->host_to_rmap, GFP_ATOMIC | __GFP_ACCOUNT);
79 spin_lock_init(&gmap->guest_table_lock);
80 spin_lock_init(&gmap->shadow_lock);
81 refcount_set(&gmap->ref_count, 1);
82 page = gmap_alloc_crst();
86 list_add(&page->lru, &gmap->crst_list);
87 table = page_to_virt(page);
88 crst_table_init(table, etype);
90 gmap->asce = atype | _ASCE_TABLE_LENGTH |
91 _ASCE_USER_BITS | __pa(table);
92 gmap->asce_end = limit;
102 * gmap_create - create a guest address space
103 * @mm: pointer to the parent mm_struct
104 * @limit: maximum size of the gmap address space
106 * Returns a guest address space structure.
108 struct gmap *gmap_create(struct mm_struct *mm, unsigned long limit)
111 unsigned long gmap_asce;
113 gmap = gmap_alloc(limit);
117 spin_lock(&mm->context.lock);
118 list_add_rcu(&gmap->list, &mm->context.gmap_list);
119 if (list_is_singular(&mm->context.gmap_list))
120 gmap_asce = gmap->asce;
123 WRITE_ONCE(mm->context.gmap_asce, gmap_asce);
124 spin_unlock(&mm->context.lock);
127 EXPORT_SYMBOL_GPL(gmap_create);
129 static void gmap_flush_tlb(struct gmap *gmap)
131 if (MACHINE_HAS_IDTE)
132 __tlb_flush_idte(gmap->asce);
134 __tlb_flush_global();
137 static void gmap_radix_tree_free(struct radix_tree_root *root)
139 struct radix_tree_iter iter;
140 unsigned long indices[16];
145 /* A radix tree is freed by deleting all of its entries */
149 radix_tree_for_each_slot(slot, root, &iter, index) {
150 indices[nr] = iter.index;
154 for (i = 0; i < nr; i++) {
156 radix_tree_delete(root, index);
161 static void gmap_rmap_radix_tree_free(struct radix_tree_root *root)
163 struct gmap_rmap *rmap, *rnext, *head;
164 struct radix_tree_iter iter;
165 unsigned long indices[16];
170 /* A radix tree is freed by deleting all of its entries */
174 radix_tree_for_each_slot(slot, root, &iter, index) {
175 indices[nr] = iter.index;
179 for (i = 0; i < nr; i++) {
181 head = radix_tree_delete(root, index);
182 gmap_for_each_rmap_safe(rmap, rnext, head)
189 * gmap_free - free a guest address space
190 * @gmap: pointer to the guest address space structure
192 * No locks required. There are no references to this gmap anymore.
194 static void gmap_free(struct gmap *gmap)
196 struct page *page, *next;
198 /* Flush tlb of all gmaps (if not already done for shadows) */
199 if (!(gmap_is_shadow(gmap) && gmap->removed))
200 gmap_flush_tlb(gmap);
201 /* Free all segment & region tables. */
202 list_for_each_entry_safe(page, next, &gmap->crst_list, lru)
203 __free_pages(page, CRST_ALLOC_ORDER);
204 gmap_radix_tree_free(&gmap->guest_to_host);
205 gmap_radix_tree_free(&gmap->host_to_guest);
207 /* Free additional data for a shadow gmap */
208 if (gmap_is_shadow(gmap)) {
209 struct ptdesc *ptdesc, *n;
211 /* Free all page tables. */
212 list_for_each_entry_safe(ptdesc, n, &gmap->pt_list, pt_list)
213 page_table_free_pgste(ptdesc);
214 gmap_rmap_radix_tree_free(&gmap->host_to_rmap);
215 /* Release reference to the parent */
216 gmap_put(gmap->parent);
223 * gmap_get - increase reference counter for guest address space
224 * @gmap: pointer to the guest address space structure
226 * Returns the gmap pointer
228 struct gmap *gmap_get(struct gmap *gmap)
230 refcount_inc(&gmap->ref_count);
233 EXPORT_SYMBOL_GPL(gmap_get);
236 * gmap_put - decrease reference counter for guest address space
237 * @gmap: pointer to the guest address space structure
239 * If the reference counter reaches zero the guest address space is freed.
241 void gmap_put(struct gmap *gmap)
243 if (refcount_dec_and_test(&gmap->ref_count))
246 EXPORT_SYMBOL_GPL(gmap_put);
249 * gmap_remove - remove a guest address space but do not free it yet
250 * @gmap: pointer to the guest address space structure
252 void gmap_remove(struct gmap *gmap)
254 struct gmap *sg, *next;
255 unsigned long gmap_asce;
257 /* Remove all shadow gmaps linked to this gmap */
258 if (!list_empty(&gmap->children)) {
259 spin_lock(&gmap->shadow_lock);
260 list_for_each_entry_safe(sg, next, &gmap->children, list) {
264 spin_unlock(&gmap->shadow_lock);
266 /* Remove gmap from the pre-mm list */
267 spin_lock(&gmap->mm->context.lock);
268 list_del_rcu(&gmap->list);
269 if (list_empty(&gmap->mm->context.gmap_list))
271 else if (list_is_singular(&gmap->mm->context.gmap_list))
272 gmap_asce = list_first_entry(&gmap->mm->context.gmap_list,
273 struct gmap, list)->asce;
276 WRITE_ONCE(gmap->mm->context.gmap_asce, gmap_asce);
277 spin_unlock(&gmap->mm->context.lock);
282 EXPORT_SYMBOL_GPL(gmap_remove);
285 * gmap_alloc_table is assumed to be called with mmap_lock held
287 static int gmap_alloc_table(struct gmap *gmap, unsigned long *table,
288 unsigned long init, unsigned long gaddr)
293 /* since we dont free the gmap table until gmap_free we can unlock */
294 page = gmap_alloc_crst();
297 new = page_to_virt(page);
298 crst_table_init(new, init);
299 spin_lock(&gmap->guest_table_lock);
300 if (*table & _REGION_ENTRY_INVALID) {
301 list_add(&page->lru, &gmap->crst_list);
302 *table = __pa(new) | _REGION_ENTRY_LENGTH |
303 (*table & _REGION_ENTRY_TYPE_MASK);
307 spin_unlock(&gmap->guest_table_lock);
309 __free_pages(page, CRST_ALLOC_ORDER);
314 * __gmap_segment_gaddr - find virtual address from segment pointer
315 * @entry: pointer to a segment table entry in the guest address space
317 * Returns the virtual address in the guest address space for the segment
319 static unsigned long __gmap_segment_gaddr(unsigned long *entry)
322 unsigned long offset;
324 offset = (unsigned long) entry / sizeof(unsigned long);
325 offset = (offset & (PTRS_PER_PMD - 1)) * PMD_SIZE;
326 page = pmd_pgtable_page((pmd_t *) entry);
327 return page->index + offset;
331 * __gmap_unlink_by_vmaddr - unlink a single segment via a host address
332 * @gmap: pointer to the guest address space structure
333 * @vmaddr: address in the host process address space
335 * Returns 1 if a TLB flush is required
337 static int __gmap_unlink_by_vmaddr(struct gmap *gmap, unsigned long vmaddr)
339 unsigned long *entry;
342 BUG_ON(gmap_is_shadow(gmap));
343 spin_lock(&gmap->guest_table_lock);
344 entry = radix_tree_delete(&gmap->host_to_guest, vmaddr >> PMD_SHIFT);
346 flush = (*entry != _SEGMENT_ENTRY_EMPTY);
347 *entry = _SEGMENT_ENTRY_EMPTY;
349 spin_unlock(&gmap->guest_table_lock);
354 * __gmap_unmap_by_gaddr - unmap a single segment via a guest address
355 * @gmap: pointer to the guest address space structure
356 * @gaddr: address in the guest address space
358 * Returns 1 if a TLB flush is required
360 static int __gmap_unmap_by_gaddr(struct gmap *gmap, unsigned long gaddr)
362 unsigned long vmaddr;
364 vmaddr = (unsigned long) radix_tree_delete(&gmap->guest_to_host,
366 return vmaddr ? __gmap_unlink_by_vmaddr(gmap, vmaddr) : 0;
370 * gmap_unmap_segment - unmap segment from the guest address space
371 * @gmap: pointer to the guest address space structure
372 * @to: address in the guest address space
373 * @len: length of the memory area to unmap
375 * Returns 0 if the unmap succeeded, -EINVAL if not.
377 int gmap_unmap_segment(struct gmap *gmap, unsigned long to, unsigned long len)
382 BUG_ON(gmap_is_shadow(gmap));
383 if ((to | len) & (PMD_SIZE - 1))
385 if (len == 0 || to + len < to)
389 mmap_write_lock(gmap->mm);
390 for (off = 0; off < len; off += PMD_SIZE)
391 flush |= __gmap_unmap_by_gaddr(gmap, to + off);
392 mmap_write_unlock(gmap->mm);
394 gmap_flush_tlb(gmap);
397 EXPORT_SYMBOL_GPL(gmap_unmap_segment);
400 * gmap_map_segment - map a segment to the guest address space
401 * @gmap: pointer to the guest address space structure
402 * @from: source address in the parent address space
403 * @to: target address in the guest address space
404 * @len: length of the memory area to map
406 * Returns 0 if the mmap succeeded, -EINVAL or -ENOMEM if not.
408 int gmap_map_segment(struct gmap *gmap, unsigned long from,
409 unsigned long to, unsigned long len)
414 BUG_ON(gmap_is_shadow(gmap));
415 if ((from | to | len) & (PMD_SIZE - 1))
417 if (len == 0 || from + len < from || to + len < to ||
418 from + len - 1 > TASK_SIZE_MAX || to + len - 1 > gmap->asce_end)
422 mmap_write_lock(gmap->mm);
423 for (off = 0; off < len; off += PMD_SIZE) {
424 /* Remove old translation */
425 flush |= __gmap_unmap_by_gaddr(gmap, to + off);
426 /* Store new translation */
427 if (radix_tree_insert(&gmap->guest_to_host,
428 (to + off) >> PMD_SHIFT,
429 (void *) from + off))
432 mmap_write_unlock(gmap->mm);
434 gmap_flush_tlb(gmap);
437 gmap_unmap_segment(gmap, to, len);
440 EXPORT_SYMBOL_GPL(gmap_map_segment);
443 * __gmap_translate - translate a guest address to a user space address
444 * @gmap: pointer to guest mapping meta data structure
445 * @gaddr: guest address
447 * Returns user space address which corresponds to the guest address or
448 * -EFAULT if no such mapping exists.
449 * This function does not establish potentially missing page table entries.
450 * The mmap_lock of the mm that belongs to the address space must be held
451 * when this function gets called.
453 * Note: Can also be called for shadow gmaps.
455 unsigned long __gmap_translate(struct gmap *gmap, unsigned long gaddr)
457 unsigned long vmaddr;
459 vmaddr = (unsigned long)
460 radix_tree_lookup(&gmap->guest_to_host, gaddr >> PMD_SHIFT);
461 /* Note: guest_to_host is empty for a shadow gmap */
462 return vmaddr ? (vmaddr | (gaddr & ~PMD_MASK)) : -EFAULT;
464 EXPORT_SYMBOL_GPL(__gmap_translate);
467 * gmap_translate - translate a guest address to a user space address
468 * @gmap: pointer to guest mapping meta data structure
469 * @gaddr: guest address
471 * Returns user space address which corresponds to the guest address or
472 * -EFAULT if no such mapping exists.
473 * This function does not establish potentially missing page table entries.
475 unsigned long gmap_translate(struct gmap *gmap, unsigned long gaddr)
479 mmap_read_lock(gmap->mm);
480 rc = __gmap_translate(gmap, gaddr);
481 mmap_read_unlock(gmap->mm);
484 EXPORT_SYMBOL_GPL(gmap_translate);
487 * gmap_unlink - disconnect a page table from the gmap shadow tables
488 * @mm: pointer to the parent mm_struct
489 * @table: pointer to the host page table
490 * @vmaddr: vm address associated with the host page table
492 void gmap_unlink(struct mm_struct *mm, unsigned long *table,
493 unsigned long vmaddr)
499 list_for_each_entry_rcu(gmap, &mm->context.gmap_list, list) {
500 flush = __gmap_unlink_by_vmaddr(gmap, vmaddr);
502 gmap_flush_tlb(gmap);
507 static void gmap_pmdp_xchg(struct gmap *gmap, pmd_t *old, pmd_t new,
508 unsigned long gaddr);
511 * __gmap_link - set up shadow page tables to connect a host to a guest address
512 * @gmap: pointer to guest mapping meta data structure
513 * @gaddr: guest address
514 * @vmaddr: vm address
516 * Returns 0 on success, -ENOMEM for out of memory conditions, and -EFAULT
517 * if the vm address is already mapped to a different guest segment.
518 * The mmap_lock of the mm that belongs to the address space must be held
519 * when this function gets called.
521 int __gmap_link(struct gmap *gmap, unsigned long gaddr, unsigned long vmaddr)
523 struct mm_struct *mm;
524 unsigned long *table;
533 BUG_ON(gmap_is_shadow(gmap));
534 /* Create higher level tables in the gmap page table */
536 if ((gmap->asce & _ASCE_TYPE_MASK) >= _ASCE_TYPE_REGION1) {
537 table += (gaddr & _REGION1_INDEX) >> _REGION1_SHIFT;
538 if ((*table & _REGION_ENTRY_INVALID) &&
539 gmap_alloc_table(gmap, table, _REGION2_ENTRY_EMPTY,
540 gaddr & _REGION1_MASK))
542 table = __va(*table & _REGION_ENTRY_ORIGIN);
544 if ((gmap->asce & _ASCE_TYPE_MASK) >= _ASCE_TYPE_REGION2) {
545 table += (gaddr & _REGION2_INDEX) >> _REGION2_SHIFT;
546 if ((*table & _REGION_ENTRY_INVALID) &&
547 gmap_alloc_table(gmap, table, _REGION3_ENTRY_EMPTY,
548 gaddr & _REGION2_MASK))
550 table = __va(*table & _REGION_ENTRY_ORIGIN);
552 if ((gmap->asce & _ASCE_TYPE_MASK) >= _ASCE_TYPE_REGION3) {
553 table += (gaddr & _REGION3_INDEX) >> _REGION3_SHIFT;
554 if ((*table & _REGION_ENTRY_INVALID) &&
555 gmap_alloc_table(gmap, table, _SEGMENT_ENTRY_EMPTY,
556 gaddr & _REGION3_MASK))
558 table = __va(*table & _REGION_ENTRY_ORIGIN);
560 table += (gaddr & _SEGMENT_INDEX) >> _SEGMENT_SHIFT;
561 /* Walk the parent mm page table */
563 pgd = pgd_offset(mm, vmaddr);
564 VM_BUG_ON(pgd_none(*pgd));
565 p4d = p4d_offset(pgd, vmaddr);
566 VM_BUG_ON(p4d_none(*p4d));
567 pud = pud_offset(p4d, vmaddr);
568 VM_BUG_ON(pud_none(*pud));
569 /* large puds cannot yet be handled */
572 pmd = pmd_offset(pud, vmaddr);
573 VM_BUG_ON(pmd_none(*pmd));
574 /* Are we allowed to use huge pages? */
575 if (pmd_leaf(*pmd) && !gmap->mm->context.allow_gmap_hpage_1m)
577 /* Link gmap segment table entry location to page table. */
578 rc = radix_tree_preload(GFP_KERNEL_ACCOUNT);
581 ptl = pmd_lock(mm, pmd);
582 spin_lock(&gmap->guest_table_lock);
583 if (*table == _SEGMENT_ENTRY_EMPTY) {
584 rc = radix_tree_insert(&gmap->host_to_guest,
585 vmaddr >> PMD_SHIFT, table);
587 if (pmd_leaf(*pmd)) {
588 *table = (pmd_val(*pmd) &
589 _SEGMENT_ENTRY_HARDWARE_BITS_LARGE)
590 | _SEGMENT_ENTRY_GMAP_UC
593 *table = pmd_val(*pmd) &
594 _SEGMENT_ENTRY_HARDWARE_BITS;
596 } else if (*table & _SEGMENT_ENTRY_PROTECT &&
597 !(pmd_val(*pmd) & _SEGMENT_ENTRY_PROTECT)) {
598 unprot = (u64)*table;
599 unprot &= ~_SEGMENT_ENTRY_PROTECT;
600 unprot |= _SEGMENT_ENTRY_GMAP_UC;
601 gmap_pmdp_xchg(gmap, (pmd_t *)table, __pmd(unprot), gaddr);
603 spin_unlock(&gmap->guest_table_lock);
605 radix_tree_preload_end();
610 * fixup_user_fault_nowait - manually resolve a user page fault without waiting
611 * @mm: mm_struct of target mm
612 * @address: user address
613 * @fault_flags:flags to pass down to handle_mm_fault()
614 * @unlocked: did we unlock the mmap_lock while retrying
616 * This function behaves similarly to fixup_user_fault(), but it guarantees
617 * that the fault will be resolved without waiting. The function might drop
618 * and re-acquire the mm lock, in which case @unlocked will be set to true.
620 * The guarantee is that the fault is handled without waiting, but the
621 * function itself might sleep, due to the lock.
623 * Context: Needs to be called with mm->mmap_lock held in read mode, and will
624 * return with the lock held in read mode; @unlocked will indicate whether
625 * the lock has been dropped and re-acquired. This is the same behaviour as
626 * fixup_user_fault().
628 * Return: 0 on success, -EAGAIN if the fault cannot be resolved without
629 * waiting, -EFAULT if the fault cannot be resolved, -ENOMEM if out of
632 static int fixup_user_fault_nowait(struct mm_struct *mm, unsigned long address,
633 unsigned int fault_flags, bool *unlocked)
635 struct vm_area_struct *vma;
636 unsigned int test_flags;
640 fault_flags |= FAULT_FLAG_ALLOW_RETRY | FAULT_FLAG_RETRY_NOWAIT;
641 test_flags = fault_flags & FAULT_FLAG_WRITE ? VM_WRITE : VM_READ;
643 vma = find_vma(mm, address);
644 if (unlikely(!vma || address < vma->vm_start))
646 if (unlikely(!(vma->vm_flags & test_flags)))
649 fault = handle_mm_fault(vma, address, fault_flags, NULL);
650 /* the mm lock has been dropped, take it again */
651 if (fault & VM_FAULT_COMPLETED) {
656 /* the mm lock has not been dropped */
657 if (fault & VM_FAULT_ERROR) {
658 rc = vm_fault_to_errno(fault, 0);
662 /* the mm lock has not been dropped because of FAULT_FLAG_RETRY_NOWAIT */
663 if (fault & VM_FAULT_RETRY)
665 /* nothing needed to be done and the mm lock has not been dropped */
670 * __gmap_fault - resolve a fault on a guest address
671 * @gmap: pointer to guest mapping meta data structure
672 * @gaddr: guest address
673 * @fault_flags: flags to pass down to handle_mm_fault()
675 * Context: Needs to be called with mm->mmap_lock held in read mode. Might
676 * drop and re-acquire the lock. Will always return with the lock held.
678 static int __gmap_fault(struct gmap *gmap, unsigned long gaddr, unsigned int fault_flags)
680 unsigned long vmaddr;
687 vmaddr = __gmap_translate(gmap, gaddr);
688 if (IS_ERR_VALUE(vmaddr))
691 if (fault_flags & FAULT_FLAG_RETRY_NOWAIT)
692 rc = fixup_user_fault_nowait(gmap->mm, vmaddr, fault_flags, &unlocked);
694 rc = fixup_user_fault(gmap->mm, vmaddr, fault_flags, &unlocked);
698 * In the case that fixup_user_fault unlocked the mmap_lock during
699 * fault-in, redo __gmap_translate() to avoid racing with a
701 * In particular, __gmap_translate(), fixup_user_fault{,_nowait}(),
702 * and __gmap_link() must all be called atomically in one go; if the
703 * lock had been dropped in between, a retry is needed.
708 return __gmap_link(gmap, gaddr, vmaddr);
712 * gmap_fault - resolve a fault on a guest address
713 * @gmap: pointer to guest mapping meta data structure
714 * @gaddr: guest address
715 * @fault_flags: flags to pass down to handle_mm_fault()
717 * Returns 0 on success, -ENOMEM for out of memory conditions, -EFAULT if the
718 * vm address is already mapped to a different guest segment, and -EAGAIN if
719 * FAULT_FLAG_RETRY_NOWAIT was specified and the fault could not be processed
722 int gmap_fault(struct gmap *gmap, unsigned long gaddr, unsigned int fault_flags)
726 mmap_read_lock(gmap->mm);
727 rc = __gmap_fault(gmap, gaddr, fault_flags);
728 mmap_read_unlock(gmap->mm);
731 EXPORT_SYMBOL_GPL(gmap_fault);
734 * this function is assumed to be called with mmap_lock held
736 void __gmap_zap(struct gmap *gmap, unsigned long gaddr)
738 struct vm_area_struct *vma;
739 unsigned long vmaddr;
743 /* Find the vm address for the guest address */
744 vmaddr = (unsigned long) radix_tree_lookup(&gmap->guest_to_host,
747 vmaddr |= gaddr & ~PMD_MASK;
749 vma = vma_lookup(gmap->mm, vmaddr);
750 if (!vma || is_vm_hugetlb_page(vma))
753 /* Get pointer to the page table entry */
754 ptep = get_locked_pte(gmap->mm, vmaddr, &ptl);
756 ptep_zap_unused(gmap->mm, vmaddr, ptep, 0);
757 pte_unmap_unlock(ptep, ptl);
761 EXPORT_SYMBOL_GPL(__gmap_zap);
763 void gmap_discard(struct gmap *gmap, unsigned long from, unsigned long to)
765 unsigned long gaddr, vmaddr, size;
766 struct vm_area_struct *vma;
768 mmap_read_lock(gmap->mm);
769 for (gaddr = from; gaddr < to;
770 gaddr = (gaddr + PMD_SIZE) & PMD_MASK) {
771 /* Find the vm address for the guest address */
772 vmaddr = (unsigned long)
773 radix_tree_lookup(&gmap->guest_to_host,
777 vmaddr |= gaddr & ~PMD_MASK;
778 /* Find vma in the parent mm */
779 vma = find_vma(gmap->mm, vmaddr);
783 * We do not discard pages that are backed by
784 * hugetlbfs, so we don't have to refault them.
786 if (is_vm_hugetlb_page(vma))
788 size = min(to - gaddr, PMD_SIZE - (gaddr & ~PMD_MASK));
789 zap_page_range_single(vma, vmaddr, size, NULL);
791 mmap_read_unlock(gmap->mm);
793 EXPORT_SYMBOL_GPL(gmap_discard);
795 static LIST_HEAD(gmap_notifier_list);
796 static DEFINE_SPINLOCK(gmap_notifier_lock);
799 * gmap_register_pte_notifier - register a pte invalidation callback
800 * @nb: pointer to the gmap notifier block
802 void gmap_register_pte_notifier(struct gmap_notifier *nb)
804 spin_lock(&gmap_notifier_lock);
805 list_add_rcu(&nb->list, &gmap_notifier_list);
806 spin_unlock(&gmap_notifier_lock);
808 EXPORT_SYMBOL_GPL(gmap_register_pte_notifier);
811 * gmap_unregister_pte_notifier - remove a pte invalidation callback
812 * @nb: pointer to the gmap notifier block
814 void gmap_unregister_pte_notifier(struct gmap_notifier *nb)
816 spin_lock(&gmap_notifier_lock);
817 list_del_rcu(&nb->list);
818 spin_unlock(&gmap_notifier_lock);
821 EXPORT_SYMBOL_GPL(gmap_unregister_pte_notifier);
824 * gmap_call_notifier - call all registered invalidation callbacks
825 * @gmap: pointer to guest mapping meta data structure
826 * @start: start virtual address in the guest address space
827 * @end: end virtual address in the guest address space
829 static void gmap_call_notifier(struct gmap *gmap, unsigned long start,
832 struct gmap_notifier *nb;
834 list_for_each_entry(nb, &gmap_notifier_list, list)
835 nb->notifier_call(gmap, start, end);
839 * gmap_table_walk - walk the gmap page tables
840 * @gmap: pointer to guest mapping meta data structure
841 * @gaddr: virtual address in the guest address space
842 * @level: page table level to stop at
844 * Returns a table entry pointer for the given guest address and @level
845 * @level=0 : returns a pointer to a page table table entry (or NULL)
846 * @level=1 : returns a pointer to a segment table entry (or NULL)
847 * @level=2 : returns a pointer to a region-3 table entry (or NULL)
848 * @level=3 : returns a pointer to a region-2 table entry (or NULL)
849 * @level=4 : returns a pointer to a region-1 table entry (or NULL)
851 * Returns NULL if the gmap page tables could not be walked to the
854 * Note: Can also be called for shadow gmaps.
856 static inline unsigned long *gmap_table_walk(struct gmap *gmap,
857 unsigned long gaddr, int level)
859 const int asce_type = gmap->asce & _ASCE_TYPE_MASK;
860 unsigned long *table = gmap->table;
862 if (gmap_is_shadow(gmap) && gmap->removed)
865 if (WARN_ON_ONCE(level > (asce_type >> 2) + 1))
868 if (asce_type != _ASCE_TYPE_REGION1 &&
869 gaddr & (-1UL << (31 + (asce_type >> 2) * 11)))
873 case _ASCE_TYPE_REGION1:
874 table += (gaddr & _REGION1_INDEX) >> _REGION1_SHIFT;
877 if (*table & _REGION_ENTRY_INVALID)
879 table = __va(*table & _REGION_ENTRY_ORIGIN);
881 case _ASCE_TYPE_REGION2:
882 table += (gaddr & _REGION2_INDEX) >> _REGION2_SHIFT;
885 if (*table & _REGION_ENTRY_INVALID)
887 table = __va(*table & _REGION_ENTRY_ORIGIN);
889 case _ASCE_TYPE_REGION3:
890 table += (gaddr & _REGION3_INDEX) >> _REGION3_SHIFT;
893 if (*table & _REGION_ENTRY_INVALID)
895 table = __va(*table & _REGION_ENTRY_ORIGIN);
897 case _ASCE_TYPE_SEGMENT:
898 table += (gaddr & _SEGMENT_INDEX) >> _SEGMENT_SHIFT;
901 if (*table & _REGION_ENTRY_INVALID)
903 table = __va(*table & _SEGMENT_ENTRY_ORIGIN);
904 table += (gaddr & _PAGE_INDEX) >> PAGE_SHIFT;
910 * gmap_pte_op_walk - walk the gmap page table, get the page table lock
911 * and return the pte pointer
912 * @gmap: pointer to guest mapping meta data structure
913 * @gaddr: virtual address in the guest address space
914 * @ptl: pointer to the spinlock pointer
916 * Returns a pointer to the locked pte for a guest address, or NULL
918 static pte_t *gmap_pte_op_walk(struct gmap *gmap, unsigned long gaddr,
921 unsigned long *table;
923 BUG_ON(gmap_is_shadow(gmap));
924 /* Walk the gmap page table, lock and get pte pointer */
925 table = gmap_table_walk(gmap, gaddr, 1); /* get segment pointer */
926 if (!table || *table & _SEGMENT_ENTRY_INVALID)
928 return pte_alloc_map_lock(gmap->mm, (pmd_t *) table, gaddr, ptl);
932 * gmap_pte_op_fixup - force a page in and connect the gmap page table
933 * @gmap: pointer to guest mapping meta data structure
934 * @gaddr: virtual address in the guest address space
935 * @vmaddr: address in the host process address space
936 * @prot: indicates access rights: PROT_NONE, PROT_READ or PROT_WRITE
938 * Returns 0 if the caller can retry __gmap_translate (might fail again),
939 * -ENOMEM if out of memory and -EFAULT if anything goes wrong while fixing
940 * up or connecting the gmap page table.
942 static int gmap_pte_op_fixup(struct gmap *gmap, unsigned long gaddr,
943 unsigned long vmaddr, int prot)
945 struct mm_struct *mm = gmap->mm;
946 unsigned int fault_flags;
947 bool unlocked = false;
949 BUG_ON(gmap_is_shadow(gmap));
950 fault_flags = (prot == PROT_WRITE) ? FAULT_FLAG_WRITE : 0;
951 if (fixup_user_fault(mm, vmaddr, fault_flags, &unlocked))
954 /* lost mmap_lock, caller has to retry __gmap_translate */
956 /* Connect the page tables */
957 return __gmap_link(gmap, gaddr, vmaddr);
961 * gmap_pte_op_end - release the page table lock
962 * @ptep: pointer to the locked pte
963 * @ptl: pointer to the page table spinlock
965 static void gmap_pte_op_end(pte_t *ptep, spinlock_t *ptl)
967 pte_unmap_unlock(ptep, ptl);
971 * gmap_pmd_op_walk - walk the gmap tables, get the guest table lock
972 * and return the pmd pointer
973 * @gmap: pointer to guest mapping meta data structure
974 * @gaddr: virtual address in the guest address space
976 * Returns a pointer to the pmd for a guest address, or NULL
978 static inline pmd_t *gmap_pmd_op_walk(struct gmap *gmap, unsigned long gaddr)
982 BUG_ON(gmap_is_shadow(gmap));
983 pmdp = (pmd_t *) gmap_table_walk(gmap, gaddr, 1);
987 /* without huge pages, there is no need to take the table lock */
988 if (!gmap->mm->context.allow_gmap_hpage_1m)
989 return pmd_none(*pmdp) ? NULL : pmdp;
991 spin_lock(&gmap->guest_table_lock);
992 if (pmd_none(*pmdp)) {
993 spin_unlock(&gmap->guest_table_lock);
997 /* 4k page table entries are locked via the pte (pte_alloc_map_lock). */
998 if (!pmd_leaf(*pmdp))
999 spin_unlock(&gmap->guest_table_lock);
1004 * gmap_pmd_op_end - release the guest_table_lock if needed
1005 * @gmap: pointer to the guest mapping meta data structure
1006 * @pmdp: pointer to the pmd
1008 static inline void gmap_pmd_op_end(struct gmap *gmap, pmd_t *pmdp)
1010 if (pmd_leaf(*pmdp))
1011 spin_unlock(&gmap->guest_table_lock);
1015 * gmap_protect_pmd - remove access rights to memory and set pmd notification bits
1016 * @pmdp: pointer to the pmd to be protected
1017 * @prot: indicates access rights: PROT_NONE, PROT_READ or PROT_WRITE
1018 * @bits: notification bits to set
1021 * 0 if successfully protected
1022 * -EAGAIN if a fixup is needed
1023 * -EINVAL if unsupported notifier bits have been specified
1025 * Expected to be called with sg->mm->mmap_lock in read and
1026 * guest_table_lock held.
1028 static int gmap_protect_pmd(struct gmap *gmap, unsigned long gaddr,
1029 pmd_t *pmdp, int prot, unsigned long bits)
1031 int pmd_i = pmd_val(*pmdp) & _SEGMENT_ENTRY_INVALID;
1032 int pmd_p = pmd_val(*pmdp) & _SEGMENT_ENTRY_PROTECT;
1036 if ((pmd_i && (prot != PROT_NONE)) || (pmd_p && (prot == PROT_WRITE)))
1039 if (prot == PROT_NONE && !pmd_i) {
1040 new = set_pmd_bit(new, __pgprot(_SEGMENT_ENTRY_INVALID));
1041 gmap_pmdp_xchg(gmap, pmdp, new, gaddr);
1044 if (prot == PROT_READ && !pmd_p) {
1045 new = clear_pmd_bit(new, __pgprot(_SEGMENT_ENTRY_INVALID));
1046 new = set_pmd_bit(new, __pgprot(_SEGMENT_ENTRY_PROTECT));
1047 gmap_pmdp_xchg(gmap, pmdp, new, gaddr);
1050 if (bits & GMAP_NOTIFY_MPROT)
1051 set_pmd(pmdp, set_pmd_bit(*pmdp, __pgprot(_SEGMENT_ENTRY_GMAP_IN)));
1053 /* Shadow GMAP protection needs split PMDs */
1054 if (bits & GMAP_NOTIFY_SHADOW)
1061 * gmap_protect_pte - remove access rights to memory and set pgste bits
1062 * @gmap: pointer to guest mapping meta data structure
1063 * @gaddr: virtual address in the guest address space
1064 * @pmdp: pointer to the pmd associated with the pte
1065 * @prot: indicates access rights: PROT_NONE, PROT_READ or PROT_WRITE
1066 * @bits: notification bits to set
1068 * Returns 0 if successfully protected, -ENOMEM if out of memory and
1069 * -EAGAIN if a fixup is needed.
1071 * Expected to be called with sg->mm->mmap_lock in read
1073 static int gmap_protect_pte(struct gmap *gmap, unsigned long gaddr,
1074 pmd_t *pmdp, int prot, unsigned long bits)
1079 unsigned long pbits = 0;
1081 if (pmd_val(*pmdp) & _SEGMENT_ENTRY_INVALID)
1084 ptep = pte_alloc_map_lock(gmap->mm, pmdp, gaddr, &ptl);
1088 pbits |= (bits & GMAP_NOTIFY_MPROT) ? PGSTE_IN_BIT : 0;
1089 pbits |= (bits & GMAP_NOTIFY_SHADOW) ? PGSTE_VSIE_BIT : 0;
1090 /* Protect and unlock. */
1091 rc = ptep_force_prot(gmap->mm, gaddr, ptep, prot, pbits);
1092 gmap_pte_op_end(ptep, ptl);
1097 * gmap_protect_range - remove access rights to memory and set pgste bits
1098 * @gmap: pointer to guest mapping meta data structure
1099 * @gaddr: virtual address in the guest address space
1100 * @len: size of area
1101 * @prot: indicates access rights: PROT_NONE, PROT_READ or PROT_WRITE
1102 * @bits: pgste notification bits to set
1104 * Returns 0 if successfully protected, -ENOMEM if out of memory and
1105 * -EFAULT if gaddr is invalid (or mapping for shadows is missing).
1107 * Called with sg->mm->mmap_lock in read.
1109 static int gmap_protect_range(struct gmap *gmap, unsigned long gaddr,
1110 unsigned long len, int prot, unsigned long bits)
1112 unsigned long vmaddr, dist;
1116 BUG_ON(gmap_is_shadow(gmap));
1119 pmdp = gmap_pmd_op_walk(gmap, gaddr);
1121 if (!pmd_leaf(*pmdp)) {
1122 rc = gmap_protect_pte(gmap, gaddr, pmdp, prot,
1129 rc = gmap_protect_pmd(gmap, gaddr, pmdp, prot,
1132 dist = HPAGE_SIZE - (gaddr & ~HPAGE_MASK);
1133 len = len < dist ? 0 : len - dist;
1134 gaddr = (gaddr & HPAGE_MASK) + HPAGE_SIZE;
1137 gmap_pmd_op_end(gmap, pmdp);
1143 /* -EAGAIN, fixup of userspace mm and gmap */
1144 vmaddr = __gmap_translate(gmap, gaddr);
1145 if (IS_ERR_VALUE(vmaddr))
1147 rc = gmap_pte_op_fixup(gmap, gaddr, vmaddr, prot);
1156 * gmap_mprotect_notify - change access rights for a range of ptes and
1157 * call the notifier if any pte changes again
1158 * @gmap: pointer to guest mapping meta data structure
1159 * @gaddr: virtual address in the guest address space
1160 * @len: size of area
1161 * @prot: indicates access rights: PROT_NONE, PROT_READ or PROT_WRITE
1163 * Returns 0 if for each page in the given range a gmap mapping exists,
1164 * the new access rights could be set and the notifier could be armed.
1165 * If the gmap mapping is missing for one or more pages -EFAULT is
1166 * returned. If no memory could be allocated -ENOMEM is returned.
1167 * This function establishes missing page table entries.
1169 int gmap_mprotect_notify(struct gmap *gmap, unsigned long gaddr,
1170 unsigned long len, int prot)
1174 if ((gaddr & ~PAGE_MASK) || (len & ~PAGE_MASK) || gmap_is_shadow(gmap))
1176 if (!MACHINE_HAS_ESOP && prot == PROT_READ)
1178 mmap_read_lock(gmap->mm);
1179 rc = gmap_protect_range(gmap, gaddr, len, prot, GMAP_NOTIFY_MPROT);
1180 mmap_read_unlock(gmap->mm);
1183 EXPORT_SYMBOL_GPL(gmap_mprotect_notify);
1186 * gmap_read_table - get an unsigned long value from a guest page table using
1187 * absolute addressing, without marking the page referenced.
1188 * @gmap: pointer to guest mapping meta data structure
1189 * @gaddr: virtual address in the guest address space
1190 * @val: pointer to the unsigned long value to return
1192 * Returns 0 if the value was read, -ENOMEM if out of memory and -EFAULT
1193 * if reading using the virtual address failed. -EINVAL if called on a gmap
1196 * Called with gmap->mm->mmap_lock in read.
1198 int gmap_read_table(struct gmap *gmap, unsigned long gaddr, unsigned long *val)
1200 unsigned long address, vmaddr;
1205 if (gmap_is_shadow(gmap))
1210 ptep = gmap_pte_op_walk(gmap, gaddr, &ptl);
1213 if (pte_present(pte) && (pte_val(pte) & _PAGE_READ)) {
1214 address = pte_val(pte) & PAGE_MASK;
1215 address += gaddr & ~PAGE_MASK;
1216 *val = *(unsigned long *)__va(address);
1217 set_pte(ptep, set_pte_bit(*ptep, __pgprot(_PAGE_YOUNG)));
1218 /* Do *NOT* clear the _PAGE_INVALID bit! */
1221 gmap_pte_op_end(ptep, ptl);
1225 vmaddr = __gmap_translate(gmap, gaddr);
1226 if (IS_ERR_VALUE(vmaddr)) {
1230 rc = gmap_pte_op_fixup(gmap, gaddr, vmaddr, PROT_READ);
1236 EXPORT_SYMBOL_GPL(gmap_read_table);
1239 * gmap_insert_rmap - add a rmap to the host_to_rmap radix tree
1240 * @sg: pointer to the shadow guest address space structure
1241 * @vmaddr: vm address associated with the rmap
1242 * @rmap: pointer to the rmap structure
1244 * Called with the sg->guest_table_lock
1246 static inline void gmap_insert_rmap(struct gmap *sg, unsigned long vmaddr,
1247 struct gmap_rmap *rmap)
1249 struct gmap_rmap *temp;
1252 BUG_ON(!gmap_is_shadow(sg));
1253 slot = radix_tree_lookup_slot(&sg->host_to_rmap, vmaddr >> PAGE_SHIFT);
1255 rmap->next = radix_tree_deref_slot_protected(slot,
1256 &sg->guest_table_lock);
1257 for (temp = rmap->next; temp; temp = temp->next) {
1258 if (temp->raddr == rmap->raddr) {
1263 radix_tree_replace_slot(&sg->host_to_rmap, slot, rmap);
1266 radix_tree_insert(&sg->host_to_rmap, vmaddr >> PAGE_SHIFT,
1272 * gmap_protect_rmap - restrict access rights to memory (RO) and create an rmap
1273 * @sg: pointer to the shadow guest address space structure
1274 * @raddr: rmap address in the shadow gmap
1275 * @paddr: address in the parent guest address space
1276 * @len: length of the memory area to protect
1278 * Returns 0 if successfully protected and the rmap was created, -ENOMEM
1279 * if out of memory and -EFAULT if paddr is invalid.
1281 static int gmap_protect_rmap(struct gmap *sg, unsigned long raddr,
1282 unsigned long paddr, unsigned long len)
1284 struct gmap *parent;
1285 struct gmap_rmap *rmap;
1286 unsigned long vmaddr;
1291 BUG_ON(!gmap_is_shadow(sg));
1292 parent = sg->parent;
1294 vmaddr = __gmap_translate(parent, paddr);
1295 if (IS_ERR_VALUE(vmaddr))
1297 rmap = kzalloc(sizeof(*rmap), GFP_KERNEL_ACCOUNT);
1300 rmap->raddr = raddr;
1301 rc = radix_tree_preload(GFP_KERNEL_ACCOUNT);
1307 ptep = gmap_pte_op_walk(parent, paddr, &ptl);
1309 spin_lock(&sg->guest_table_lock);
1310 rc = ptep_force_prot(parent->mm, paddr, ptep, PROT_READ,
1313 gmap_insert_rmap(sg, vmaddr, rmap);
1314 spin_unlock(&sg->guest_table_lock);
1315 gmap_pte_op_end(ptep, ptl);
1317 radix_tree_preload_end();
1320 rc = gmap_pte_op_fixup(parent, paddr, vmaddr, PROT_READ);
1331 #define _SHADOW_RMAP_MASK 0x7
1332 #define _SHADOW_RMAP_REGION1 0x5
1333 #define _SHADOW_RMAP_REGION2 0x4
1334 #define _SHADOW_RMAP_REGION3 0x3
1335 #define _SHADOW_RMAP_SEGMENT 0x2
1336 #define _SHADOW_RMAP_PGTABLE 0x1
1339 * gmap_idte_one - invalidate a single region or segment table entry
1340 * @asce: region or segment table *origin* + table-type bits
1341 * @vaddr: virtual address to identify the table entry to flush
1343 * The invalid bit of a single region or segment table entry is set
1344 * and the associated TLB entries depending on the entry are flushed.
1345 * The table-type of the @asce identifies the portion of the @vaddr
1346 * that is used as the invalidation index.
1348 static inline void gmap_idte_one(unsigned long asce, unsigned long vaddr)
1352 : : "a" (asce), "a" (vaddr) : "cc", "memory");
1356 * gmap_unshadow_page - remove a page from a shadow page table
1357 * @sg: pointer to the shadow guest address space structure
1358 * @raddr: rmap address in the shadow guest address space
1360 * Called with the sg->guest_table_lock
1362 static void gmap_unshadow_page(struct gmap *sg, unsigned long raddr)
1364 unsigned long *table;
1366 BUG_ON(!gmap_is_shadow(sg));
1367 table = gmap_table_walk(sg, raddr, 0); /* get page table pointer */
1368 if (!table || *table & _PAGE_INVALID)
1370 gmap_call_notifier(sg, raddr, raddr + PAGE_SIZE - 1);
1371 ptep_unshadow_pte(sg->mm, raddr, (pte_t *) table);
1375 * __gmap_unshadow_pgt - remove all entries from a shadow page table
1376 * @sg: pointer to the shadow guest address space structure
1377 * @raddr: rmap address in the shadow guest address space
1378 * @pgt: pointer to the start of a shadow page table
1380 * Called with the sg->guest_table_lock
1382 static void __gmap_unshadow_pgt(struct gmap *sg, unsigned long raddr,
1387 BUG_ON(!gmap_is_shadow(sg));
1388 for (i = 0; i < _PAGE_ENTRIES; i++, raddr += PAGE_SIZE)
1389 pgt[i] = _PAGE_INVALID;
1393 * gmap_unshadow_pgt - remove a shadow page table from a segment entry
1394 * @sg: pointer to the shadow guest address space structure
1395 * @raddr: address in the shadow guest address space
1397 * Called with the sg->guest_table_lock
1399 static void gmap_unshadow_pgt(struct gmap *sg, unsigned long raddr)
1402 phys_addr_t sto, pgt;
1403 struct ptdesc *ptdesc;
1405 BUG_ON(!gmap_is_shadow(sg));
1406 ste = gmap_table_walk(sg, raddr, 1); /* get segment pointer */
1407 if (!ste || !(*ste & _SEGMENT_ENTRY_ORIGIN))
1409 gmap_call_notifier(sg, raddr, raddr + _SEGMENT_SIZE - 1);
1410 sto = __pa(ste - ((raddr & _SEGMENT_INDEX) >> _SEGMENT_SHIFT));
1411 gmap_idte_one(sto | _ASCE_TYPE_SEGMENT, raddr);
1412 pgt = *ste & _SEGMENT_ENTRY_ORIGIN;
1413 *ste = _SEGMENT_ENTRY_EMPTY;
1414 __gmap_unshadow_pgt(sg, raddr, __va(pgt));
1415 /* Free page table */
1416 ptdesc = page_ptdesc(phys_to_page(pgt));
1417 list_del(&ptdesc->pt_list);
1418 page_table_free_pgste(ptdesc);
1422 * __gmap_unshadow_sgt - remove all entries from a shadow segment table
1423 * @sg: pointer to the shadow guest address space structure
1424 * @raddr: rmap address in the shadow guest address space
1425 * @sgt: pointer to the start of a shadow segment table
1427 * Called with the sg->guest_table_lock
1429 static void __gmap_unshadow_sgt(struct gmap *sg, unsigned long raddr,
1432 struct ptdesc *ptdesc;
1436 BUG_ON(!gmap_is_shadow(sg));
1437 for (i = 0; i < _CRST_ENTRIES; i++, raddr += _SEGMENT_SIZE) {
1438 if (!(sgt[i] & _SEGMENT_ENTRY_ORIGIN))
1440 pgt = sgt[i] & _REGION_ENTRY_ORIGIN;
1441 sgt[i] = _SEGMENT_ENTRY_EMPTY;
1442 __gmap_unshadow_pgt(sg, raddr, __va(pgt));
1443 /* Free page table */
1444 ptdesc = page_ptdesc(phys_to_page(pgt));
1445 list_del(&ptdesc->pt_list);
1446 page_table_free_pgste(ptdesc);
1451 * gmap_unshadow_sgt - remove a shadow segment table from a region-3 entry
1452 * @sg: pointer to the shadow guest address space structure
1453 * @raddr: rmap address in the shadow guest address space
1455 * Called with the shadow->guest_table_lock
1457 static void gmap_unshadow_sgt(struct gmap *sg, unsigned long raddr)
1459 unsigned long r3o, *r3e;
1463 BUG_ON(!gmap_is_shadow(sg));
1464 r3e = gmap_table_walk(sg, raddr, 2); /* get region-3 pointer */
1465 if (!r3e || !(*r3e & _REGION_ENTRY_ORIGIN))
1467 gmap_call_notifier(sg, raddr, raddr + _REGION3_SIZE - 1);
1468 r3o = (unsigned long) (r3e - ((raddr & _REGION3_INDEX) >> _REGION3_SHIFT));
1469 gmap_idte_one(__pa(r3o) | _ASCE_TYPE_REGION3, raddr);
1470 sgt = *r3e & _REGION_ENTRY_ORIGIN;
1471 *r3e = _REGION3_ENTRY_EMPTY;
1472 __gmap_unshadow_sgt(sg, raddr, __va(sgt));
1473 /* Free segment table */
1474 page = phys_to_page(sgt);
1475 list_del(&page->lru);
1476 __free_pages(page, CRST_ALLOC_ORDER);
1480 * __gmap_unshadow_r3t - remove all entries from a shadow region-3 table
1481 * @sg: pointer to the shadow guest address space structure
1482 * @raddr: address in the shadow guest address space
1483 * @r3t: pointer to the start of a shadow region-3 table
1485 * Called with the sg->guest_table_lock
1487 static void __gmap_unshadow_r3t(struct gmap *sg, unsigned long raddr,
1494 BUG_ON(!gmap_is_shadow(sg));
1495 for (i = 0; i < _CRST_ENTRIES; i++, raddr += _REGION3_SIZE) {
1496 if (!(r3t[i] & _REGION_ENTRY_ORIGIN))
1498 sgt = r3t[i] & _REGION_ENTRY_ORIGIN;
1499 r3t[i] = _REGION3_ENTRY_EMPTY;
1500 __gmap_unshadow_sgt(sg, raddr, __va(sgt));
1501 /* Free segment table */
1502 page = phys_to_page(sgt);
1503 list_del(&page->lru);
1504 __free_pages(page, CRST_ALLOC_ORDER);
1509 * gmap_unshadow_r3t - remove a shadow region-3 table from a region-2 entry
1510 * @sg: pointer to the shadow guest address space structure
1511 * @raddr: rmap address in the shadow guest address space
1513 * Called with the sg->guest_table_lock
1515 static void gmap_unshadow_r3t(struct gmap *sg, unsigned long raddr)
1517 unsigned long r2o, *r2e;
1521 BUG_ON(!gmap_is_shadow(sg));
1522 r2e = gmap_table_walk(sg, raddr, 3); /* get region-2 pointer */
1523 if (!r2e || !(*r2e & _REGION_ENTRY_ORIGIN))
1525 gmap_call_notifier(sg, raddr, raddr + _REGION2_SIZE - 1);
1526 r2o = (unsigned long) (r2e - ((raddr & _REGION2_INDEX) >> _REGION2_SHIFT));
1527 gmap_idte_one(__pa(r2o) | _ASCE_TYPE_REGION2, raddr);
1528 r3t = *r2e & _REGION_ENTRY_ORIGIN;
1529 *r2e = _REGION2_ENTRY_EMPTY;
1530 __gmap_unshadow_r3t(sg, raddr, __va(r3t));
1531 /* Free region 3 table */
1532 page = phys_to_page(r3t);
1533 list_del(&page->lru);
1534 __free_pages(page, CRST_ALLOC_ORDER);
1538 * __gmap_unshadow_r2t - remove all entries from a shadow region-2 table
1539 * @sg: pointer to the shadow guest address space structure
1540 * @raddr: rmap address in the shadow guest address space
1541 * @r2t: pointer to the start of a shadow region-2 table
1543 * Called with the sg->guest_table_lock
1545 static void __gmap_unshadow_r2t(struct gmap *sg, unsigned long raddr,
1552 BUG_ON(!gmap_is_shadow(sg));
1553 for (i = 0; i < _CRST_ENTRIES; i++, raddr += _REGION2_SIZE) {
1554 if (!(r2t[i] & _REGION_ENTRY_ORIGIN))
1556 r3t = r2t[i] & _REGION_ENTRY_ORIGIN;
1557 r2t[i] = _REGION2_ENTRY_EMPTY;
1558 __gmap_unshadow_r3t(sg, raddr, __va(r3t));
1559 /* Free region 3 table */
1560 page = phys_to_page(r3t);
1561 list_del(&page->lru);
1562 __free_pages(page, CRST_ALLOC_ORDER);
1567 * gmap_unshadow_r2t - remove a shadow region-2 table from a region-1 entry
1568 * @sg: pointer to the shadow guest address space structure
1569 * @raddr: rmap address in the shadow guest address space
1571 * Called with the sg->guest_table_lock
1573 static void gmap_unshadow_r2t(struct gmap *sg, unsigned long raddr)
1575 unsigned long r1o, *r1e;
1579 BUG_ON(!gmap_is_shadow(sg));
1580 r1e = gmap_table_walk(sg, raddr, 4); /* get region-1 pointer */
1581 if (!r1e || !(*r1e & _REGION_ENTRY_ORIGIN))
1583 gmap_call_notifier(sg, raddr, raddr + _REGION1_SIZE - 1);
1584 r1o = (unsigned long) (r1e - ((raddr & _REGION1_INDEX) >> _REGION1_SHIFT));
1585 gmap_idte_one(__pa(r1o) | _ASCE_TYPE_REGION1, raddr);
1586 r2t = *r1e & _REGION_ENTRY_ORIGIN;
1587 *r1e = _REGION1_ENTRY_EMPTY;
1588 __gmap_unshadow_r2t(sg, raddr, __va(r2t));
1589 /* Free region 2 table */
1590 page = phys_to_page(r2t);
1591 list_del(&page->lru);
1592 __free_pages(page, CRST_ALLOC_ORDER);
1596 * __gmap_unshadow_r1t - remove all entries from a shadow region-1 table
1597 * @sg: pointer to the shadow guest address space structure
1598 * @raddr: rmap address in the shadow guest address space
1599 * @r1t: pointer to the start of a shadow region-1 table
1601 * Called with the shadow->guest_table_lock
1603 static void __gmap_unshadow_r1t(struct gmap *sg, unsigned long raddr,
1611 BUG_ON(!gmap_is_shadow(sg));
1612 asce = __pa(r1t) | _ASCE_TYPE_REGION1;
1613 for (i = 0; i < _CRST_ENTRIES; i++, raddr += _REGION1_SIZE) {
1614 if (!(r1t[i] & _REGION_ENTRY_ORIGIN))
1616 r2t = r1t[i] & _REGION_ENTRY_ORIGIN;
1617 __gmap_unshadow_r2t(sg, raddr, __va(r2t));
1618 /* Clear entry and flush translation r1t -> r2t */
1619 gmap_idte_one(asce, raddr);
1620 r1t[i] = _REGION1_ENTRY_EMPTY;
1621 /* Free region 2 table */
1622 page = phys_to_page(r2t);
1623 list_del(&page->lru);
1624 __free_pages(page, CRST_ALLOC_ORDER);
1629 * gmap_unshadow - remove a shadow page table completely
1630 * @sg: pointer to the shadow guest address space structure
1632 * Called with sg->guest_table_lock
1634 static void gmap_unshadow(struct gmap *sg)
1636 unsigned long *table;
1638 BUG_ON(!gmap_is_shadow(sg));
1642 gmap_call_notifier(sg, 0, -1UL);
1644 table = __va(sg->asce & _ASCE_ORIGIN);
1645 switch (sg->asce & _ASCE_TYPE_MASK) {
1646 case _ASCE_TYPE_REGION1:
1647 __gmap_unshadow_r1t(sg, 0, table);
1649 case _ASCE_TYPE_REGION2:
1650 __gmap_unshadow_r2t(sg, 0, table);
1652 case _ASCE_TYPE_REGION3:
1653 __gmap_unshadow_r3t(sg, 0, table);
1655 case _ASCE_TYPE_SEGMENT:
1656 __gmap_unshadow_sgt(sg, 0, table);
1662 * gmap_find_shadow - find a specific asce in the list of shadow tables
1663 * @parent: pointer to the parent gmap
1664 * @asce: ASCE for which the shadow table is created
1665 * @edat_level: edat level to be used for the shadow translation
1667 * Returns the pointer to a gmap if a shadow table with the given asce is
1668 * already available, ERR_PTR(-EAGAIN) if another one is just being created,
1671 static struct gmap *gmap_find_shadow(struct gmap *parent, unsigned long asce,
1676 list_for_each_entry(sg, &parent->children, list) {
1677 if (sg->orig_asce != asce || sg->edat_level != edat_level ||
1680 if (!sg->initialized)
1681 return ERR_PTR(-EAGAIN);
1682 refcount_inc(&sg->ref_count);
1689 * gmap_shadow_valid - check if a shadow guest address space matches the
1690 * given properties and is still valid
1691 * @sg: pointer to the shadow guest address space structure
1692 * @asce: ASCE for which the shadow table is requested
1693 * @edat_level: edat level to be used for the shadow translation
1695 * Returns 1 if the gmap shadow is still valid and matches the given
1696 * properties, the caller can continue using it. Returns 0 otherwise, the
1697 * caller has to request a new shadow gmap in this case.
1700 int gmap_shadow_valid(struct gmap *sg, unsigned long asce, int edat_level)
1704 return sg->orig_asce == asce && sg->edat_level == edat_level;
1706 EXPORT_SYMBOL_GPL(gmap_shadow_valid);
1709 * gmap_shadow - create/find a shadow guest address space
1710 * @parent: pointer to the parent gmap
1711 * @asce: ASCE for which the shadow table is created
1712 * @edat_level: edat level to be used for the shadow translation
1714 * The pages of the top level page table referred by the asce parameter
1715 * will be set to read-only and marked in the PGSTEs of the kvm process.
1716 * The shadow table will be removed automatically on any change to the
1717 * PTE mapping for the source table.
1719 * Returns a guest address space structure, ERR_PTR(-ENOMEM) if out of memory,
1720 * ERR_PTR(-EAGAIN) if the caller has to retry and ERR_PTR(-EFAULT) if the
1721 * parent gmap table could not be protected.
1723 struct gmap *gmap_shadow(struct gmap *parent, unsigned long asce,
1726 struct gmap *sg, *new;
1727 unsigned long limit;
1730 BUG_ON(parent->mm->context.allow_gmap_hpage_1m);
1731 BUG_ON(gmap_is_shadow(parent));
1732 spin_lock(&parent->shadow_lock);
1733 sg = gmap_find_shadow(parent, asce, edat_level);
1734 spin_unlock(&parent->shadow_lock);
1737 /* Create a new shadow gmap */
1738 limit = -1UL >> (33 - (((asce & _ASCE_TYPE_MASK) >> 2) * 11));
1739 if (asce & _ASCE_REAL_SPACE)
1741 new = gmap_alloc(limit);
1743 return ERR_PTR(-ENOMEM);
1744 new->mm = parent->mm;
1745 new->parent = gmap_get(parent);
1746 new->private = parent->private;
1747 new->orig_asce = asce;
1748 new->edat_level = edat_level;
1749 new->initialized = false;
1750 spin_lock(&parent->shadow_lock);
1751 /* Recheck if another CPU created the same shadow */
1752 sg = gmap_find_shadow(parent, asce, edat_level);
1754 spin_unlock(&parent->shadow_lock);
1758 if (asce & _ASCE_REAL_SPACE) {
1759 /* only allow one real-space gmap shadow */
1760 list_for_each_entry(sg, &parent->children, list) {
1761 if (sg->orig_asce & _ASCE_REAL_SPACE) {
1762 spin_lock(&sg->guest_table_lock);
1764 spin_unlock(&sg->guest_table_lock);
1765 list_del(&sg->list);
1771 refcount_set(&new->ref_count, 2);
1772 list_add(&new->list, &parent->children);
1773 if (asce & _ASCE_REAL_SPACE) {
1774 /* nothing to protect, return right away */
1775 new->initialized = true;
1776 spin_unlock(&parent->shadow_lock);
1779 spin_unlock(&parent->shadow_lock);
1780 /* protect after insertion, so it will get properly invalidated */
1781 mmap_read_lock(parent->mm);
1782 rc = gmap_protect_range(parent, asce & _ASCE_ORIGIN,
1783 ((asce & _ASCE_TABLE_LENGTH) + 1) * PAGE_SIZE,
1784 PROT_READ, GMAP_NOTIFY_SHADOW);
1785 mmap_read_unlock(parent->mm);
1786 spin_lock(&parent->shadow_lock);
1787 new->initialized = true;
1789 list_del(&new->list);
1793 spin_unlock(&parent->shadow_lock);
1796 EXPORT_SYMBOL_GPL(gmap_shadow);
1799 * gmap_shadow_r2t - create an empty shadow region 2 table
1800 * @sg: pointer to the shadow guest address space structure
1801 * @saddr: faulting address in the shadow gmap
1802 * @r2t: parent gmap address of the region 2 table to get shadowed
1803 * @fake: r2t references contiguous guest memory block, not a r2t
1805 * The r2t parameter specifies the address of the source table. The
1806 * four pages of the source table are made read-only in the parent gmap
1807 * address space. A write to the source table area @r2t will automatically
1808 * remove the shadow r2 table and all of its descendants.
1810 * Returns 0 if successfully shadowed or already shadowed, -EAGAIN if the
1811 * shadow table structure is incomplete, -ENOMEM if out of memory and
1812 * -EFAULT if an address in the parent gmap could not be resolved.
1814 * Called with sg->mm->mmap_lock in read.
1816 int gmap_shadow_r2t(struct gmap *sg, unsigned long saddr, unsigned long r2t,
1819 unsigned long raddr, origin, offset, len;
1820 unsigned long *table;
1825 BUG_ON(!gmap_is_shadow(sg));
1826 /* Allocate a shadow region second table */
1827 page = gmap_alloc_crst();
1830 page->index = r2t & _REGION_ENTRY_ORIGIN;
1832 page->index |= GMAP_SHADOW_FAKE_TABLE;
1833 s_r2t = page_to_phys(page);
1834 /* Install shadow region second table */
1835 spin_lock(&sg->guest_table_lock);
1836 table = gmap_table_walk(sg, saddr, 4); /* get region-1 pointer */
1838 rc = -EAGAIN; /* Race with unshadow */
1841 if (!(*table & _REGION_ENTRY_INVALID)) {
1842 rc = 0; /* Already established */
1844 } else if (*table & _REGION_ENTRY_ORIGIN) {
1845 rc = -EAGAIN; /* Race with shadow */
1848 crst_table_init(__va(s_r2t), _REGION2_ENTRY_EMPTY);
1849 /* mark as invalid as long as the parent table is not protected */
1850 *table = s_r2t | _REGION_ENTRY_LENGTH |
1851 _REGION_ENTRY_TYPE_R1 | _REGION_ENTRY_INVALID;
1852 if (sg->edat_level >= 1)
1853 *table |= (r2t & _REGION_ENTRY_PROTECT);
1854 list_add(&page->lru, &sg->crst_list);
1856 /* nothing to protect for fake tables */
1857 *table &= ~_REGION_ENTRY_INVALID;
1858 spin_unlock(&sg->guest_table_lock);
1861 spin_unlock(&sg->guest_table_lock);
1862 /* Make r2t read-only in parent gmap page table */
1863 raddr = (saddr & _REGION1_MASK) | _SHADOW_RMAP_REGION1;
1864 origin = r2t & _REGION_ENTRY_ORIGIN;
1865 offset = ((r2t & _REGION_ENTRY_OFFSET) >> 6) * PAGE_SIZE;
1866 len = ((r2t & _REGION_ENTRY_LENGTH) + 1) * PAGE_SIZE - offset;
1867 rc = gmap_protect_rmap(sg, raddr, origin + offset, len);
1868 spin_lock(&sg->guest_table_lock);
1870 table = gmap_table_walk(sg, saddr, 4);
1871 if (!table || (*table & _REGION_ENTRY_ORIGIN) != s_r2t)
1872 rc = -EAGAIN; /* Race with unshadow */
1874 *table &= ~_REGION_ENTRY_INVALID;
1876 gmap_unshadow_r2t(sg, raddr);
1878 spin_unlock(&sg->guest_table_lock);
1881 spin_unlock(&sg->guest_table_lock);
1882 __free_pages(page, CRST_ALLOC_ORDER);
1885 EXPORT_SYMBOL_GPL(gmap_shadow_r2t);
1888 * gmap_shadow_r3t - create a shadow region 3 table
1889 * @sg: pointer to the shadow guest address space structure
1890 * @saddr: faulting address in the shadow gmap
1891 * @r3t: parent gmap address of the region 3 table to get shadowed
1892 * @fake: r3t references contiguous guest memory block, not a r3t
1894 * Returns 0 if successfully shadowed or already shadowed, -EAGAIN if the
1895 * shadow table structure is incomplete, -ENOMEM if out of memory and
1896 * -EFAULT if an address in the parent gmap could not be resolved.
1898 * Called with sg->mm->mmap_lock in read.
1900 int gmap_shadow_r3t(struct gmap *sg, unsigned long saddr, unsigned long r3t,
1903 unsigned long raddr, origin, offset, len;
1904 unsigned long *table;
1909 BUG_ON(!gmap_is_shadow(sg));
1910 /* Allocate a shadow region second table */
1911 page = gmap_alloc_crst();
1914 page->index = r3t & _REGION_ENTRY_ORIGIN;
1916 page->index |= GMAP_SHADOW_FAKE_TABLE;
1917 s_r3t = page_to_phys(page);
1918 /* Install shadow region second table */
1919 spin_lock(&sg->guest_table_lock);
1920 table = gmap_table_walk(sg, saddr, 3); /* get region-2 pointer */
1922 rc = -EAGAIN; /* Race with unshadow */
1925 if (!(*table & _REGION_ENTRY_INVALID)) {
1926 rc = 0; /* Already established */
1928 } else if (*table & _REGION_ENTRY_ORIGIN) {
1929 rc = -EAGAIN; /* Race with shadow */
1932 crst_table_init(__va(s_r3t), _REGION3_ENTRY_EMPTY);
1933 /* mark as invalid as long as the parent table is not protected */
1934 *table = s_r3t | _REGION_ENTRY_LENGTH |
1935 _REGION_ENTRY_TYPE_R2 | _REGION_ENTRY_INVALID;
1936 if (sg->edat_level >= 1)
1937 *table |= (r3t & _REGION_ENTRY_PROTECT);
1938 list_add(&page->lru, &sg->crst_list);
1940 /* nothing to protect for fake tables */
1941 *table &= ~_REGION_ENTRY_INVALID;
1942 spin_unlock(&sg->guest_table_lock);
1945 spin_unlock(&sg->guest_table_lock);
1946 /* Make r3t read-only in parent gmap page table */
1947 raddr = (saddr & _REGION2_MASK) | _SHADOW_RMAP_REGION2;
1948 origin = r3t & _REGION_ENTRY_ORIGIN;
1949 offset = ((r3t & _REGION_ENTRY_OFFSET) >> 6) * PAGE_SIZE;
1950 len = ((r3t & _REGION_ENTRY_LENGTH) + 1) * PAGE_SIZE - offset;
1951 rc = gmap_protect_rmap(sg, raddr, origin + offset, len);
1952 spin_lock(&sg->guest_table_lock);
1954 table = gmap_table_walk(sg, saddr, 3);
1955 if (!table || (*table & _REGION_ENTRY_ORIGIN) != s_r3t)
1956 rc = -EAGAIN; /* Race with unshadow */
1958 *table &= ~_REGION_ENTRY_INVALID;
1960 gmap_unshadow_r3t(sg, raddr);
1962 spin_unlock(&sg->guest_table_lock);
1965 spin_unlock(&sg->guest_table_lock);
1966 __free_pages(page, CRST_ALLOC_ORDER);
1969 EXPORT_SYMBOL_GPL(gmap_shadow_r3t);
1972 * gmap_shadow_sgt - create a shadow segment table
1973 * @sg: pointer to the shadow guest address space structure
1974 * @saddr: faulting address in the shadow gmap
1975 * @sgt: parent gmap address of the segment table to get shadowed
1976 * @fake: sgt references contiguous guest memory block, not a sgt
1978 * Returns: 0 if successfully shadowed or already shadowed, -EAGAIN if the
1979 * shadow table structure is incomplete, -ENOMEM if out of memory and
1980 * -EFAULT if an address in the parent gmap could not be resolved.
1982 * Called with sg->mm->mmap_lock in read.
1984 int gmap_shadow_sgt(struct gmap *sg, unsigned long saddr, unsigned long sgt,
1987 unsigned long raddr, origin, offset, len;
1988 unsigned long *table;
1993 BUG_ON(!gmap_is_shadow(sg) || (sgt & _REGION3_ENTRY_LARGE));
1994 /* Allocate a shadow segment table */
1995 page = gmap_alloc_crst();
1998 page->index = sgt & _REGION_ENTRY_ORIGIN;
2000 page->index |= GMAP_SHADOW_FAKE_TABLE;
2001 s_sgt = page_to_phys(page);
2002 /* Install shadow region second table */
2003 spin_lock(&sg->guest_table_lock);
2004 table = gmap_table_walk(sg, saddr, 2); /* get region-3 pointer */
2006 rc = -EAGAIN; /* Race with unshadow */
2009 if (!(*table & _REGION_ENTRY_INVALID)) {
2010 rc = 0; /* Already established */
2012 } else if (*table & _REGION_ENTRY_ORIGIN) {
2013 rc = -EAGAIN; /* Race with shadow */
2016 crst_table_init(__va(s_sgt), _SEGMENT_ENTRY_EMPTY);
2017 /* mark as invalid as long as the parent table is not protected */
2018 *table = s_sgt | _REGION_ENTRY_LENGTH |
2019 _REGION_ENTRY_TYPE_R3 | _REGION_ENTRY_INVALID;
2020 if (sg->edat_level >= 1)
2021 *table |= sgt & _REGION_ENTRY_PROTECT;
2022 list_add(&page->lru, &sg->crst_list);
2024 /* nothing to protect for fake tables */
2025 *table &= ~_REGION_ENTRY_INVALID;
2026 spin_unlock(&sg->guest_table_lock);
2029 spin_unlock(&sg->guest_table_lock);
2030 /* Make sgt read-only in parent gmap page table */
2031 raddr = (saddr & _REGION3_MASK) | _SHADOW_RMAP_REGION3;
2032 origin = sgt & _REGION_ENTRY_ORIGIN;
2033 offset = ((sgt & _REGION_ENTRY_OFFSET) >> 6) * PAGE_SIZE;
2034 len = ((sgt & _REGION_ENTRY_LENGTH) + 1) * PAGE_SIZE - offset;
2035 rc = gmap_protect_rmap(sg, raddr, origin + offset, len);
2036 spin_lock(&sg->guest_table_lock);
2038 table = gmap_table_walk(sg, saddr, 2);
2039 if (!table || (*table & _REGION_ENTRY_ORIGIN) != s_sgt)
2040 rc = -EAGAIN; /* Race with unshadow */
2042 *table &= ~_REGION_ENTRY_INVALID;
2044 gmap_unshadow_sgt(sg, raddr);
2046 spin_unlock(&sg->guest_table_lock);
2049 spin_unlock(&sg->guest_table_lock);
2050 __free_pages(page, CRST_ALLOC_ORDER);
2053 EXPORT_SYMBOL_GPL(gmap_shadow_sgt);
2056 * gmap_shadow_pgt_lookup - find a shadow page table
2057 * @sg: pointer to the shadow guest address space structure
2058 * @saddr: the address in the shadow aguest address space
2059 * @pgt: parent gmap address of the page table to get shadowed
2060 * @dat_protection: if the pgtable is marked as protected by dat
2061 * @fake: pgt references contiguous guest memory block, not a pgtable
2063 * Returns 0 if the shadow page table was found and -EAGAIN if the page
2064 * table was not found.
2066 * Called with sg->mm->mmap_lock in read.
2068 int gmap_shadow_pgt_lookup(struct gmap *sg, unsigned long saddr,
2069 unsigned long *pgt, int *dat_protection,
2072 unsigned long *table;
2076 BUG_ON(!gmap_is_shadow(sg));
2077 spin_lock(&sg->guest_table_lock);
2078 table = gmap_table_walk(sg, saddr, 1); /* get segment pointer */
2079 if (table && !(*table & _SEGMENT_ENTRY_INVALID)) {
2080 /* Shadow page tables are full pages (pte+pgste) */
2081 page = pfn_to_page(*table >> PAGE_SHIFT);
2082 *pgt = page->index & ~GMAP_SHADOW_FAKE_TABLE;
2083 *dat_protection = !!(*table & _SEGMENT_ENTRY_PROTECT);
2084 *fake = !!(page->index & GMAP_SHADOW_FAKE_TABLE);
2089 spin_unlock(&sg->guest_table_lock);
2093 EXPORT_SYMBOL_GPL(gmap_shadow_pgt_lookup);
2096 * gmap_shadow_pgt - instantiate a shadow page table
2097 * @sg: pointer to the shadow guest address space structure
2098 * @saddr: faulting address in the shadow gmap
2099 * @pgt: parent gmap address of the page table to get shadowed
2100 * @fake: pgt references contiguous guest memory block, not a pgtable
2102 * Returns 0 if successfully shadowed or already shadowed, -EAGAIN if the
2103 * shadow table structure is incomplete, -ENOMEM if out of memory,
2104 * -EFAULT if an address in the parent gmap could not be resolved and
2106 * Called with gmap->mm->mmap_lock in read
2108 int gmap_shadow_pgt(struct gmap *sg, unsigned long saddr, unsigned long pgt,
2111 unsigned long raddr, origin;
2112 unsigned long *table;
2113 struct ptdesc *ptdesc;
2117 BUG_ON(!gmap_is_shadow(sg) || (pgt & _SEGMENT_ENTRY_LARGE));
2118 /* Allocate a shadow page table */
2119 ptdesc = page_table_alloc_pgste(sg->mm);
2122 ptdesc->pt_index = pgt & _SEGMENT_ENTRY_ORIGIN;
2124 ptdesc->pt_index |= GMAP_SHADOW_FAKE_TABLE;
2125 s_pgt = page_to_phys(ptdesc_page(ptdesc));
2126 /* Install shadow page table */
2127 spin_lock(&sg->guest_table_lock);
2128 table = gmap_table_walk(sg, saddr, 1); /* get segment pointer */
2130 rc = -EAGAIN; /* Race with unshadow */
2133 if (!(*table & _SEGMENT_ENTRY_INVALID)) {
2134 rc = 0; /* Already established */
2136 } else if (*table & _SEGMENT_ENTRY_ORIGIN) {
2137 rc = -EAGAIN; /* Race with shadow */
2140 /* mark as invalid as long as the parent table is not protected */
2141 *table = (unsigned long) s_pgt | _SEGMENT_ENTRY |
2142 (pgt & _SEGMENT_ENTRY_PROTECT) | _SEGMENT_ENTRY_INVALID;
2143 list_add(&ptdesc->pt_list, &sg->pt_list);
2145 /* nothing to protect for fake tables */
2146 *table &= ~_SEGMENT_ENTRY_INVALID;
2147 spin_unlock(&sg->guest_table_lock);
2150 spin_unlock(&sg->guest_table_lock);
2151 /* Make pgt read-only in parent gmap page table (not the pgste) */
2152 raddr = (saddr & _SEGMENT_MASK) | _SHADOW_RMAP_SEGMENT;
2153 origin = pgt & _SEGMENT_ENTRY_ORIGIN & PAGE_MASK;
2154 rc = gmap_protect_rmap(sg, raddr, origin, PAGE_SIZE);
2155 spin_lock(&sg->guest_table_lock);
2157 table = gmap_table_walk(sg, saddr, 1);
2158 if (!table || (*table & _SEGMENT_ENTRY_ORIGIN) != s_pgt)
2159 rc = -EAGAIN; /* Race with unshadow */
2161 *table &= ~_SEGMENT_ENTRY_INVALID;
2163 gmap_unshadow_pgt(sg, raddr);
2165 spin_unlock(&sg->guest_table_lock);
2168 spin_unlock(&sg->guest_table_lock);
2169 page_table_free_pgste(ptdesc);
2173 EXPORT_SYMBOL_GPL(gmap_shadow_pgt);
2176 * gmap_shadow_page - create a shadow page mapping
2177 * @sg: pointer to the shadow guest address space structure
2178 * @saddr: faulting address in the shadow gmap
2179 * @pte: pte in parent gmap address space to get shadowed
2181 * Returns 0 if successfully shadowed or already shadowed, -EAGAIN if the
2182 * shadow table structure is incomplete, -ENOMEM if out of memory and
2183 * -EFAULT if an address in the parent gmap could not be resolved.
2185 * Called with sg->mm->mmap_lock in read.
2187 int gmap_shadow_page(struct gmap *sg, unsigned long saddr, pte_t pte)
2189 struct gmap *parent;
2190 struct gmap_rmap *rmap;
2191 unsigned long vmaddr, paddr;
2193 pte_t *sptep, *tptep;
2197 BUG_ON(!gmap_is_shadow(sg));
2198 parent = sg->parent;
2199 prot = (pte_val(pte) & _PAGE_PROTECT) ? PROT_READ : PROT_WRITE;
2201 rmap = kzalloc(sizeof(*rmap), GFP_KERNEL_ACCOUNT);
2204 rmap->raddr = (saddr & PAGE_MASK) | _SHADOW_RMAP_PGTABLE;
2207 paddr = pte_val(pte) & PAGE_MASK;
2208 vmaddr = __gmap_translate(parent, paddr);
2209 if (IS_ERR_VALUE(vmaddr)) {
2213 rc = radix_tree_preload(GFP_KERNEL_ACCOUNT);
2217 sptep = gmap_pte_op_walk(parent, paddr, &ptl);
2219 spin_lock(&sg->guest_table_lock);
2220 /* Get page table pointer */
2221 tptep = (pte_t *) gmap_table_walk(sg, saddr, 0);
2223 spin_unlock(&sg->guest_table_lock);
2224 gmap_pte_op_end(sptep, ptl);
2225 radix_tree_preload_end();
2228 rc = ptep_shadow_pte(sg->mm, saddr, sptep, tptep, pte);
2230 /* Success and a new mapping */
2231 gmap_insert_rmap(sg, vmaddr, rmap);
2235 gmap_pte_op_end(sptep, ptl);
2236 spin_unlock(&sg->guest_table_lock);
2238 radix_tree_preload_end();
2241 rc = gmap_pte_op_fixup(parent, paddr, vmaddr, prot);
2248 EXPORT_SYMBOL_GPL(gmap_shadow_page);
2251 * gmap_shadow_notify - handle notifications for shadow gmap
2253 * Called with sg->parent->shadow_lock.
2255 static void gmap_shadow_notify(struct gmap *sg, unsigned long vmaddr,
2256 unsigned long gaddr)
2258 struct gmap_rmap *rmap, *rnext, *head;
2259 unsigned long start, end, bits, raddr;
2261 BUG_ON(!gmap_is_shadow(sg));
2263 spin_lock(&sg->guest_table_lock);
2265 spin_unlock(&sg->guest_table_lock);
2268 /* Check for top level table */
2269 start = sg->orig_asce & _ASCE_ORIGIN;
2270 end = start + ((sg->orig_asce & _ASCE_TABLE_LENGTH) + 1) * PAGE_SIZE;
2271 if (!(sg->orig_asce & _ASCE_REAL_SPACE) && gaddr >= start &&
2273 /* The complete shadow table has to go */
2275 spin_unlock(&sg->guest_table_lock);
2276 list_del(&sg->list);
2280 /* Remove the page table tree from on specific entry */
2281 head = radix_tree_delete(&sg->host_to_rmap, vmaddr >> PAGE_SHIFT);
2282 gmap_for_each_rmap_safe(rmap, rnext, head) {
2283 bits = rmap->raddr & _SHADOW_RMAP_MASK;
2284 raddr = rmap->raddr ^ bits;
2286 case _SHADOW_RMAP_REGION1:
2287 gmap_unshadow_r2t(sg, raddr);
2289 case _SHADOW_RMAP_REGION2:
2290 gmap_unshadow_r3t(sg, raddr);
2292 case _SHADOW_RMAP_REGION3:
2293 gmap_unshadow_sgt(sg, raddr);
2295 case _SHADOW_RMAP_SEGMENT:
2296 gmap_unshadow_pgt(sg, raddr);
2298 case _SHADOW_RMAP_PGTABLE:
2299 gmap_unshadow_page(sg, raddr);
2304 spin_unlock(&sg->guest_table_lock);
2308 * ptep_notify - call all invalidation callbacks for a specific pte.
2309 * @mm: pointer to the process mm_struct
2310 * @vmaddr: virtual address in the process address space
2311 * @pte: pointer to the page table entry
2312 * @bits: bits from the pgste that caused the notify call
2314 * This function is assumed to be called with the page table lock held
2315 * for the pte to notify.
2317 void ptep_notify(struct mm_struct *mm, unsigned long vmaddr,
2318 pte_t *pte, unsigned long bits)
2320 unsigned long offset, gaddr = 0;
2321 unsigned long *table;
2322 struct gmap *gmap, *sg, *next;
2324 offset = ((unsigned long) pte) & (255 * sizeof(pte_t));
2325 offset = offset * (PAGE_SIZE / sizeof(pte_t));
2327 list_for_each_entry_rcu(gmap, &mm->context.gmap_list, list) {
2328 spin_lock(&gmap->guest_table_lock);
2329 table = radix_tree_lookup(&gmap->host_to_guest,
2330 vmaddr >> PMD_SHIFT);
2332 gaddr = __gmap_segment_gaddr(table) + offset;
2333 spin_unlock(&gmap->guest_table_lock);
2337 if (!list_empty(&gmap->children) && (bits & PGSTE_VSIE_BIT)) {
2338 spin_lock(&gmap->shadow_lock);
2339 list_for_each_entry_safe(sg, next,
2340 &gmap->children, list)
2341 gmap_shadow_notify(sg, vmaddr, gaddr);
2342 spin_unlock(&gmap->shadow_lock);
2344 if (bits & PGSTE_IN_BIT)
2345 gmap_call_notifier(gmap, gaddr, gaddr + PAGE_SIZE - 1);
2349 EXPORT_SYMBOL_GPL(ptep_notify);
2351 static void pmdp_notify_gmap(struct gmap *gmap, pmd_t *pmdp,
2352 unsigned long gaddr)
2354 set_pmd(pmdp, clear_pmd_bit(*pmdp, __pgprot(_SEGMENT_ENTRY_GMAP_IN)));
2355 gmap_call_notifier(gmap, gaddr, gaddr + HPAGE_SIZE - 1);
2359 * gmap_pmdp_xchg - exchange a gmap pmd with another
2360 * @gmap: pointer to the guest address space structure
2361 * @pmdp: pointer to the pmd entry
2362 * @new: replacement entry
2363 * @gaddr: the affected guest address
2365 * This function is assumed to be called with the guest_table_lock
2368 static void gmap_pmdp_xchg(struct gmap *gmap, pmd_t *pmdp, pmd_t new,
2369 unsigned long gaddr)
2371 gaddr &= HPAGE_MASK;
2372 pmdp_notify_gmap(gmap, pmdp, gaddr);
2373 new = clear_pmd_bit(new, __pgprot(_SEGMENT_ENTRY_GMAP_IN));
2374 if (MACHINE_HAS_TLB_GUEST)
2375 __pmdp_idte(gaddr, (pmd_t *)pmdp, IDTE_GUEST_ASCE, gmap->asce,
2377 else if (MACHINE_HAS_IDTE)
2378 __pmdp_idte(gaddr, (pmd_t *)pmdp, 0, 0, IDTE_GLOBAL);
2384 static void gmap_pmdp_clear(struct mm_struct *mm, unsigned long vmaddr,
2389 unsigned long gaddr;
2392 list_for_each_entry_rcu(gmap, &mm->context.gmap_list, list) {
2393 spin_lock(&gmap->guest_table_lock);
2394 pmdp = (pmd_t *)radix_tree_delete(&gmap->host_to_guest,
2395 vmaddr >> PMD_SHIFT);
2397 gaddr = __gmap_segment_gaddr((unsigned long *)pmdp);
2398 pmdp_notify_gmap(gmap, pmdp, gaddr);
2399 WARN_ON(pmd_val(*pmdp) & ~(_SEGMENT_ENTRY_HARDWARE_BITS_LARGE |
2400 _SEGMENT_ENTRY_GMAP_UC |
2404 set_pmd(pmdp, __pmd(_SEGMENT_ENTRY_EMPTY));
2406 spin_unlock(&gmap->guest_table_lock);
2412 * gmap_pmdp_invalidate - invalidate all affected guest pmd entries without
2414 * @mm: pointer to the process mm_struct
2415 * @vmaddr: virtual address in the process address space
2417 void gmap_pmdp_invalidate(struct mm_struct *mm, unsigned long vmaddr)
2419 gmap_pmdp_clear(mm, vmaddr, 0);
2421 EXPORT_SYMBOL_GPL(gmap_pmdp_invalidate);
2424 * gmap_pmdp_csp - csp all affected guest pmd entries
2425 * @mm: pointer to the process mm_struct
2426 * @vmaddr: virtual address in the process address space
2428 void gmap_pmdp_csp(struct mm_struct *mm, unsigned long vmaddr)
2430 gmap_pmdp_clear(mm, vmaddr, 1);
2432 EXPORT_SYMBOL_GPL(gmap_pmdp_csp);
2435 * gmap_pmdp_idte_local - invalidate and clear a guest pmd entry
2436 * @mm: pointer to the process mm_struct
2437 * @vmaddr: virtual address in the process address space
2439 void gmap_pmdp_idte_local(struct mm_struct *mm, unsigned long vmaddr)
2441 unsigned long *entry, gaddr;
2446 list_for_each_entry_rcu(gmap, &mm->context.gmap_list, list) {
2447 spin_lock(&gmap->guest_table_lock);
2448 entry = radix_tree_delete(&gmap->host_to_guest,
2449 vmaddr >> PMD_SHIFT);
2451 pmdp = (pmd_t *)entry;
2452 gaddr = __gmap_segment_gaddr(entry);
2453 pmdp_notify_gmap(gmap, pmdp, gaddr);
2454 WARN_ON(*entry & ~(_SEGMENT_ENTRY_HARDWARE_BITS_LARGE |
2455 _SEGMENT_ENTRY_GMAP_UC |
2457 if (MACHINE_HAS_TLB_GUEST)
2458 __pmdp_idte(gaddr, pmdp, IDTE_GUEST_ASCE,
2459 gmap->asce, IDTE_LOCAL);
2460 else if (MACHINE_HAS_IDTE)
2461 __pmdp_idte(gaddr, pmdp, 0, 0, IDTE_LOCAL);
2462 *entry = _SEGMENT_ENTRY_EMPTY;
2464 spin_unlock(&gmap->guest_table_lock);
2468 EXPORT_SYMBOL_GPL(gmap_pmdp_idte_local);
2471 * gmap_pmdp_idte_global - invalidate and clear a guest pmd entry
2472 * @mm: pointer to the process mm_struct
2473 * @vmaddr: virtual address in the process address space
2475 void gmap_pmdp_idte_global(struct mm_struct *mm, unsigned long vmaddr)
2477 unsigned long *entry, gaddr;
2482 list_for_each_entry_rcu(gmap, &mm->context.gmap_list, list) {
2483 spin_lock(&gmap->guest_table_lock);
2484 entry = radix_tree_delete(&gmap->host_to_guest,
2485 vmaddr >> PMD_SHIFT);
2487 pmdp = (pmd_t *)entry;
2488 gaddr = __gmap_segment_gaddr(entry);
2489 pmdp_notify_gmap(gmap, pmdp, gaddr);
2490 WARN_ON(*entry & ~(_SEGMENT_ENTRY_HARDWARE_BITS_LARGE |
2491 _SEGMENT_ENTRY_GMAP_UC |
2493 if (MACHINE_HAS_TLB_GUEST)
2494 __pmdp_idte(gaddr, pmdp, IDTE_GUEST_ASCE,
2495 gmap->asce, IDTE_GLOBAL);
2496 else if (MACHINE_HAS_IDTE)
2497 __pmdp_idte(gaddr, pmdp, 0, 0, IDTE_GLOBAL);
2500 *entry = _SEGMENT_ENTRY_EMPTY;
2502 spin_unlock(&gmap->guest_table_lock);
2506 EXPORT_SYMBOL_GPL(gmap_pmdp_idte_global);
2509 * gmap_test_and_clear_dirty_pmd - test and reset segment dirty status
2510 * @gmap: pointer to guest address space
2511 * @pmdp: pointer to the pmd to be tested
2512 * @gaddr: virtual address in the guest address space
2514 * This function is assumed to be called with the guest_table_lock
2517 static bool gmap_test_and_clear_dirty_pmd(struct gmap *gmap, pmd_t *pmdp,
2518 unsigned long gaddr)
2520 if (pmd_val(*pmdp) & _SEGMENT_ENTRY_INVALID)
2523 /* Already protected memory, which did not change is clean */
2524 if (pmd_val(*pmdp) & _SEGMENT_ENTRY_PROTECT &&
2525 !(pmd_val(*pmdp) & _SEGMENT_ENTRY_GMAP_UC))
2528 /* Clear UC indication and reset protection */
2529 set_pmd(pmdp, clear_pmd_bit(*pmdp, __pgprot(_SEGMENT_ENTRY_GMAP_UC)));
2530 gmap_protect_pmd(gmap, gaddr, pmdp, PROT_READ, 0);
2535 * gmap_sync_dirty_log_pmd - set bitmap based on dirty status of segment
2536 * @gmap: pointer to guest address space
2537 * @bitmap: dirty bitmap for this pmd
2538 * @gaddr: virtual address in the guest address space
2539 * @vmaddr: virtual address in the host address space
2541 * This function is assumed to be called with the guest_table_lock
2544 void gmap_sync_dirty_log_pmd(struct gmap *gmap, unsigned long bitmap[4],
2545 unsigned long gaddr, unsigned long vmaddr)
2552 pmdp = gmap_pmd_op_walk(gmap, gaddr);
2556 if (pmd_leaf(*pmdp)) {
2557 if (gmap_test_and_clear_dirty_pmd(gmap, pmdp, gaddr))
2558 bitmap_fill(bitmap, _PAGE_ENTRIES);
2560 for (i = 0; i < _PAGE_ENTRIES; i++, vmaddr += PAGE_SIZE) {
2561 ptep = pte_alloc_map_lock(gmap->mm, pmdp, vmaddr, &ptl);
2564 if (ptep_test_and_clear_uc(gmap->mm, vmaddr, ptep))
2566 pte_unmap_unlock(ptep, ptl);
2569 gmap_pmd_op_end(gmap, pmdp);
2571 EXPORT_SYMBOL_GPL(gmap_sync_dirty_log_pmd);
2573 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
2574 static int thp_split_walk_pmd_entry(pmd_t *pmd, unsigned long addr,
2575 unsigned long end, struct mm_walk *walk)
2577 struct vm_area_struct *vma = walk->vma;
2579 split_huge_pmd(vma, pmd, addr);
2583 static const struct mm_walk_ops thp_split_walk_ops = {
2584 .pmd_entry = thp_split_walk_pmd_entry,
2585 .walk_lock = PGWALK_WRLOCK_VERIFY,
2588 static inline void thp_split_mm(struct mm_struct *mm)
2590 struct vm_area_struct *vma;
2591 VMA_ITERATOR(vmi, mm, 0);
2593 for_each_vma(vmi, vma) {
2594 vm_flags_mod(vma, VM_NOHUGEPAGE, VM_HUGEPAGE);
2595 walk_page_vma(vma, &thp_split_walk_ops, NULL);
2597 mm->def_flags |= VM_NOHUGEPAGE;
2600 static inline void thp_split_mm(struct mm_struct *mm)
2603 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
2606 * switch on pgstes for its userspace process (for kvm)
2608 int s390_enable_sie(void)
2610 struct mm_struct *mm = current->mm;
2612 /* Do we have pgstes? if yes, we are done */
2613 if (mm_has_pgste(mm))
2615 /* Fail if the page tables are 2K */
2616 if (!mm_alloc_pgste(mm))
2618 mmap_write_lock(mm);
2619 mm->context.has_pgste = 1;
2620 /* split thp mappings and disable thp for future mappings */
2622 mmap_write_unlock(mm);
2625 EXPORT_SYMBOL_GPL(s390_enable_sie);
2627 static int find_zeropage_pte_entry(pte_t *pte, unsigned long addr,
2628 unsigned long end, struct mm_walk *walk)
2630 unsigned long *found_addr = walk->private;
2632 /* Return 1 of the page is a zeropage. */
2633 if (is_zero_pfn(pte_pfn(*pte))) {
2635 * Shared zeropage in e.g., a FS DAX mapping? We cannot do the
2636 * right thing and likely don't care: FAULT_FLAG_UNSHARE
2637 * currently only works in COW mappings, which is also where
2638 * mm_forbids_zeropage() is checked.
2640 if (!is_cow_mapping(walk->vma->vm_flags))
2649 static const struct mm_walk_ops find_zeropage_ops = {
2650 .pte_entry = find_zeropage_pte_entry,
2651 .walk_lock = PGWALK_WRLOCK,
2655 * Unshare all shared zeropages, replacing them by anonymous pages. Note that
2656 * we cannot simply zap all shared zeropages, because this could later
2657 * trigger unexpected userfaultfd missing events.
2659 * This must be called after mm->context.allow_cow_sharing was
2660 * set to 0, to avoid future mappings of shared zeropages.
2662 * mm contracts with s390, that even if mm were to remove a page table,
2663 * and racing with walk_page_range_vma() calling pte_offset_map_lock()
2664 * would fail, it will never insert a page table containing empty zero
2665 * pages once mm_forbids_zeropage(mm) i.e.
2666 * mm->context.allow_cow_sharing is set to 0.
2668 static int __s390_unshare_zeropages(struct mm_struct *mm)
2670 struct vm_area_struct *vma;
2671 VMA_ITERATOR(vmi, mm, 0);
2676 for_each_vma(vmi, vma) {
2678 * We could only look at COW mappings, but it's more future
2679 * proof to catch unexpected zeropages in other mappings and
2682 if ((vma->vm_flags & VM_PFNMAP) || is_vm_hugetlb_page(vma))
2684 addr = vma->vm_start;
2687 rc = walk_page_range_vma(vma, addr, vma->vm_end,
2688 &find_zeropage_ops, &addr);
2694 /* addr was updated by find_zeropage_pte_entry() */
2695 fault = handle_mm_fault(vma, addr,
2696 FAULT_FLAG_UNSHARE | FAULT_FLAG_REMOTE,
2698 if (fault & VM_FAULT_OOM)
2701 * See break_ksm(): even after handle_mm_fault() returned 0, we
2702 * must start the lookup from the current address, because
2703 * handle_mm_fault() may back out if there's any difficulty.
2705 * VM_FAULT_SIGBUS and VM_FAULT_SIGSEGV are unexpected but
2706 * maybe they could trigger in the future on concurrent
2707 * truncation. In that case, the shared zeropage would be gone
2708 * and we can simply retry and make progress.
2717 static int __s390_disable_cow_sharing(struct mm_struct *mm)
2721 if (!mm->context.allow_cow_sharing)
2724 mm->context.allow_cow_sharing = 0;
2726 /* Replace all shared zeropages by anonymous pages. */
2727 rc = __s390_unshare_zeropages(mm);
2729 * Make sure to disable KSM (if enabled for the whole process or
2730 * individual VMAs). Note that nothing currently hinders user space
2731 * from re-enabling it.
2734 rc = ksm_disable(mm);
2736 mm->context.allow_cow_sharing = 1;
2741 * Disable most COW-sharing of memory pages for the whole process:
2742 * (1) Disable KSM and unmerge/unshare any KSM pages.
2743 * (2) Disallow shared zeropages and unshare any zerpages that are mapped.
2745 * Not that we currently don't bother with COW-shared pages that are shared
2746 * with parent/child processes due to fork().
2748 int s390_disable_cow_sharing(void)
2752 mmap_write_lock(current->mm);
2753 rc = __s390_disable_cow_sharing(current->mm);
2754 mmap_write_unlock(current->mm);
2757 EXPORT_SYMBOL_GPL(s390_disable_cow_sharing);
2760 * Enable storage key handling from now on and initialize the storage
2761 * keys with the default key.
2763 static int __s390_enable_skey_pte(pte_t *pte, unsigned long addr,
2764 unsigned long next, struct mm_walk *walk)
2766 /* Clear storage key */
2767 ptep_zap_key(walk->mm, addr, pte);
2772 * Give a chance to schedule after setting a key to 256 pages.
2773 * We only hold the mm lock, which is a rwsem and the kvm srcu.
2776 static int __s390_enable_skey_pmd(pmd_t *pmd, unsigned long addr,
2777 unsigned long next, struct mm_walk *walk)
2783 static int __s390_enable_skey_hugetlb(pte_t *pte, unsigned long addr,
2784 unsigned long hmask, unsigned long next,
2785 struct mm_walk *walk)
2787 pmd_t *pmd = (pmd_t *)pte;
2788 unsigned long start, end;
2789 struct folio *folio = page_folio(pmd_page(*pmd));
2792 * The write check makes sure we do not set a key on shared
2793 * memory. This is needed as the walker does not differentiate
2794 * between actual guest memory and the process executable or
2797 if (pmd_val(*pmd) & _SEGMENT_ENTRY_INVALID ||
2798 !(pmd_val(*pmd) & _SEGMENT_ENTRY_WRITE))
2801 start = pmd_val(*pmd) & HPAGE_MASK;
2802 end = start + HPAGE_SIZE;
2803 __storage_key_init_range(start, end);
2804 set_bit(PG_arch_1, &folio->flags);
2809 static const struct mm_walk_ops enable_skey_walk_ops = {
2810 .hugetlb_entry = __s390_enable_skey_hugetlb,
2811 .pte_entry = __s390_enable_skey_pte,
2812 .pmd_entry = __s390_enable_skey_pmd,
2813 .walk_lock = PGWALK_WRLOCK,
2816 int s390_enable_skey(void)
2818 struct mm_struct *mm = current->mm;
2821 mmap_write_lock(mm);
2822 if (mm_uses_skeys(mm))
2825 mm->context.uses_skeys = 1;
2826 rc = __s390_disable_cow_sharing(mm);
2828 mm->context.uses_skeys = 0;
2831 walk_page_range(mm, 0, TASK_SIZE, &enable_skey_walk_ops, NULL);
2834 mmap_write_unlock(mm);
2837 EXPORT_SYMBOL_GPL(s390_enable_skey);
2840 * Reset CMMA state, make all pages stable again.
2842 static int __s390_reset_cmma(pte_t *pte, unsigned long addr,
2843 unsigned long next, struct mm_walk *walk)
2845 ptep_zap_unused(walk->mm, addr, pte, 1);
2849 static const struct mm_walk_ops reset_cmma_walk_ops = {
2850 .pte_entry = __s390_reset_cmma,
2851 .walk_lock = PGWALK_WRLOCK,
2854 void s390_reset_cmma(struct mm_struct *mm)
2856 mmap_write_lock(mm);
2857 walk_page_range(mm, 0, TASK_SIZE, &reset_cmma_walk_ops, NULL);
2858 mmap_write_unlock(mm);
2860 EXPORT_SYMBOL_GPL(s390_reset_cmma);
2862 #define GATHER_GET_PAGES 32
2864 struct reset_walk_state {
2866 unsigned long count;
2867 unsigned long pfns[GATHER_GET_PAGES];
2870 static int s390_gather_pages(pte_t *ptep, unsigned long addr,
2871 unsigned long next, struct mm_walk *walk)
2873 struct reset_walk_state *p = walk->private;
2874 pte_t pte = READ_ONCE(*ptep);
2876 if (pte_present(pte)) {
2877 /* we have a reference from the mapping, take an extra one */
2878 get_page(phys_to_page(pte_val(pte)));
2879 p->pfns[p->count] = phys_to_pfn(pte_val(pte));
2883 return p->count >= GATHER_GET_PAGES;
2886 static const struct mm_walk_ops gather_pages_ops = {
2887 .pte_entry = s390_gather_pages,
2888 .walk_lock = PGWALK_RDLOCK,
2892 * Call the Destroy secure page UVC on each page in the given array of PFNs.
2893 * Each page needs to have an extra reference, which will be released here.
2895 void s390_uv_destroy_pfns(unsigned long count, unsigned long *pfns)
2897 struct folio *folio;
2900 for (i = 0; i < count; i++) {
2901 folio = pfn_folio(pfns[i]);
2902 /* we always have an extra reference */
2903 uv_destroy_folio(folio);
2904 /* get rid of the extra reference */
2909 EXPORT_SYMBOL_GPL(s390_uv_destroy_pfns);
2912 * __s390_uv_destroy_range - Call the destroy secure page UVC on each page
2913 * in the given range of the given address space.
2914 * @mm: the mm to operate on
2915 * @start: the start of the range
2916 * @end: the end of the range
2917 * @interruptible: if not 0, stop when a fatal signal is received
2919 * Walk the given range of the given address space and call the destroy
2920 * secure page UVC on each page. Optionally exit early if a fatal signal is
2923 * Return: 0 on success, -EINTR if the function stopped before completing
2925 int __s390_uv_destroy_range(struct mm_struct *mm, unsigned long start,
2926 unsigned long end, bool interruptible)
2928 struct reset_walk_state state = { .next = start };
2934 r = walk_page_range(mm, state.next, end, &gather_pages_ops, &state);
2935 mmap_read_unlock(mm);
2937 s390_uv_destroy_pfns(state.count, state.pfns);
2938 if (interruptible && fatal_signal_pending(current))
2943 EXPORT_SYMBOL_GPL(__s390_uv_destroy_range);
2946 * s390_unlist_old_asce - Remove the topmost level of page tables from the
2947 * list of page tables of the gmap.
2948 * @gmap: the gmap whose table is to be removed
2950 * On s390x, KVM keeps a list of all pages containing the page tables of the
2951 * gmap (the CRST list). This list is used at tear down time to free all
2952 * pages that are now not needed anymore.
2954 * This function removes the topmost page of the tree (the one pointed to by
2955 * the ASCE) from the CRST list.
2957 * This means that it will not be freed when the VM is torn down, and needs
2958 * to be handled separately by the caller, unless a leak is actually
2959 * intended. Notice that this function will only remove the page from the
2960 * list, the page will still be used as a top level page table (and ASCE).
2962 void s390_unlist_old_asce(struct gmap *gmap)
2966 old = virt_to_page(gmap->table);
2967 spin_lock(&gmap->guest_table_lock);
2968 list_del(&old->lru);
2970 * Sometimes the topmost page might need to be "removed" multiple
2971 * times, for example if the VM is rebooted into secure mode several
2972 * times concurrently, or if s390_replace_asce fails after calling
2973 * s390_remove_old_asce and is attempted again later. In that case
2974 * the old asce has been removed from the list, and therefore it
2975 * will not be freed when the VM terminates, but the ASCE is still
2976 * in use and still pointed to.
2977 * A subsequent call to replace_asce will follow the pointer and try
2978 * to remove the same page from the list again.
2979 * Therefore it's necessary that the page of the ASCE has valid
2980 * pointers, so list_del can work (and do nothing) without
2981 * dereferencing stale or invalid pointers.
2983 INIT_LIST_HEAD(&old->lru);
2984 spin_unlock(&gmap->guest_table_lock);
2986 EXPORT_SYMBOL_GPL(s390_unlist_old_asce);
2989 * s390_replace_asce - Try to replace the current ASCE of a gmap with a copy
2990 * @gmap: the gmap whose ASCE needs to be replaced
2992 * If the ASCE is a SEGMENT type then this function will return -EINVAL,
2993 * otherwise the pointers in the host_to_guest radix tree will keep pointing
2994 * to the wrong pages, causing use-after-free and memory corruption.
2995 * If the allocation of the new top level page table fails, the ASCE is not
2997 * In any case, the old ASCE is always removed from the gmap CRST list.
2998 * Therefore the caller has to make sure to save a pointer to it
2999 * beforehand, unless a leak is actually intended.
3001 int s390_replace_asce(struct gmap *gmap)
3007 s390_unlist_old_asce(gmap);
3009 /* Replacing segment type ASCEs would cause serious issues */
3010 if ((gmap->asce & _ASCE_TYPE_MASK) == _ASCE_TYPE_SEGMENT)
3013 page = gmap_alloc_crst();
3017 table = page_to_virt(page);
3018 memcpy(table, gmap->table, 1UL << (CRST_ALLOC_ORDER + PAGE_SHIFT));
3021 * The caller has to deal with the old ASCE, but here we make sure
3022 * the new one is properly added to the CRST list, so that
3023 * it will be freed when the VM is torn down.
3025 spin_lock(&gmap->guest_table_lock);
3026 list_add(&page->lru, &gmap->crst_list);
3027 spin_unlock(&gmap->guest_table_lock);
3029 /* Set new table origin while preserving existing ASCE control bits */
3030 asce = (gmap->asce & ~_ASCE_ORIGIN) | __pa(table);
3031 WRITE_ONCE(gmap->asce, asce);
3032 WRITE_ONCE(gmap->mm->context.gmap_asce, asce);
3033 WRITE_ONCE(gmap->table, table);
3037 EXPORT_SYMBOL_GPL(s390_replace_asce);