1 // SPDX-License-Identifier: GPL-2.0
3 * KVM guest address space mapping code
5 * Copyright IBM Corp. 2007, 2016, 2018
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>
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>
21 #include <asm/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 * @mm: pointer to the parent mm_struct
31 * @limit: maximum address of the gmap address space
33 * Returns a guest address space structure.
35 static struct gmap *gmap_alloc(unsigned long limit)
40 unsigned long etype, atype;
42 if (limit < _REGION3_SIZE) {
43 limit = _REGION3_SIZE - 1;
44 atype = _ASCE_TYPE_SEGMENT;
45 etype = _SEGMENT_ENTRY_EMPTY;
46 } else if (limit < _REGION2_SIZE) {
47 limit = _REGION2_SIZE - 1;
48 atype = _ASCE_TYPE_REGION3;
49 etype = _REGION3_ENTRY_EMPTY;
50 } else if (limit < _REGION1_SIZE) {
51 limit = _REGION1_SIZE - 1;
52 atype = _ASCE_TYPE_REGION2;
53 etype = _REGION2_ENTRY_EMPTY;
56 atype = _ASCE_TYPE_REGION1;
57 etype = _REGION1_ENTRY_EMPTY;
59 gmap = kzalloc(sizeof(struct gmap), GFP_KERNEL);
62 INIT_LIST_HEAD(&gmap->crst_list);
63 INIT_LIST_HEAD(&gmap->children);
64 INIT_LIST_HEAD(&gmap->pt_list);
65 INIT_RADIX_TREE(&gmap->guest_to_host, GFP_KERNEL);
66 INIT_RADIX_TREE(&gmap->host_to_guest, GFP_ATOMIC);
67 INIT_RADIX_TREE(&gmap->host_to_rmap, GFP_ATOMIC);
68 spin_lock_init(&gmap->guest_table_lock);
69 spin_lock_init(&gmap->shadow_lock);
70 atomic_set(&gmap->ref_count, 1);
71 page = alloc_pages(GFP_KERNEL, CRST_ALLOC_ORDER);
75 list_add(&page->lru, &gmap->crst_list);
76 table = (unsigned long *) page_to_phys(page);
77 crst_table_init(table, etype);
79 gmap->asce = atype | _ASCE_TABLE_LENGTH |
80 _ASCE_USER_BITS | __pa(table);
81 gmap->asce_end = limit;
91 * gmap_create - create a guest address space
92 * @mm: pointer to the parent mm_struct
93 * @limit: maximum size of the gmap address space
95 * Returns a guest address space structure.
97 struct gmap *gmap_create(struct mm_struct *mm, unsigned long limit)
100 unsigned long gmap_asce;
102 gmap = gmap_alloc(limit);
106 spin_lock(&mm->context.lock);
107 list_add_rcu(&gmap->list, &mm->context.gmap_list);
108 if (list_is_singular(&mm->context.gmap_list))
109 gmap_asce = gmap->asce;
112 WRITE_ONCE(mm->context.gmap_asce, gmap_asce);
113 spin_unlock(&mm->context.lock);
116 EXPORT_SYMBOL_GPL(gmap_create);
118 static void gmap_flush_tlb(struct gmap *gmap)
120 if (MACHINE_HAS_IDTE)
121 __tlb_flush_idte(gmap->asce);
123 __tlb_flush_global();
126 static void gmap_radix_tree_free(struct radix_tree_root *root)
128 struct radix_tree_iter iter;
129 unsigned long indices[16];
134 /* A radix tree is freed by deleting all of its entries */
138 radix_tree_for_each_slot(slot, root, &iter, index) {
139 indices[nr] = iter.index;
143 for (i = 0; i < nr; i++) {
145 radix_tree_delete(root, index);
150 static void gmap_rmap_radix_tree_free(struct radix_tree_root *root)
152 struct gmap_rmap *rmap, *rnext, *head;
153 struct radix_tree_iter iter;
154 unsigned long indices[16];
159 /* A radix tree is freed by deleting all of its entries */
163 radix_tree_for_each_slot(slot, root, &iter, index) {
164 indices[nr] = iter.index;
168 for (i = 0; i < nr; i++) {
170 head = radix_tree_delete(root, index);
171 gmap_for_each_rmap_safe(rmap, rnext, head)
178 * gmap_free - free a guest address space
179 * @gmap: pointer to the guest address space structure
181 * No locks required. There are no references to this gmap anymore.
183 static void gmap_free(struct gmap *gmap)
185 struct page *page, *next;
187 /* Flush tlb of all gmaps (if not already done for shadows) */
188 if (!(gmap_is_shadow(gmap) && gmap->removed))
189 gmap_flush_tlb(gmap);
190 /* Free all segment & region tables. */
191 list_for_each_entry_safe(page, next, &gmap->crst_list, lru)
192 __free_pages(page, CRST_ALLOC_ORDER);
193 gmap_radix_tree_free(&gmap->guest_to_host);
194 gmap_radix_tree_free(&gmap->host_to_guest);
196 /* Free additional data for a shadow gmap */
197 if (gmap_is_shadow(gmap)) {
198 /* Free all page tables. */
199 list_for_each_entry_safe(page, next, &gmap->pt_list, lru)
200 page_table_free_pgste(page);
201 gmap_rmap_radix_tree_free(&gmap->host_to_rmap);
202 /* Release reference to the parent */
203 gmap_put(gmap->parent);
210 * gmap_get - increase reference counter for guest address space
211 * @gmap: pointer to the guest address space structure
213 * Returns the gmap pointer
215 struct gmap *gmap_get(struct gmap *gmap)
217 atomic_inc(&gmap->ref_count);
220 EXPORT_SYMBOL_GPL(gmap_get);
223 * gmap_put - decrease reference counter for guest address space
224 * @gmap: pointer to the guest address space structure
226 * If the reference counter reaches zero the guest address space is freed.
228 void gmap_put(struct gmap *gmap)
230 if (atomic_dec_return(&gmap->ref_count) == 0)
233 EXPORT_SYMBOL_GPL(gmap_put);
236 * gmap_remove - remove a guest address space but do not free it yet
237 * @gmap: pointer to the guest address space structure
239 void gmap_remove(struct gmap *gmap)
241 struct gmap *sg, *next;
242 unsigned long gmap_asce;
244 /* Remove all shadow gmaps linked to this gmap */
245 if (!list_empty(&gmap->children)) {
246 spin_lock(&gmap->shadow_lock);
247 list_for_each_entry_safe(sg, next, &gmap->children, list) {
251 spin_unlock(&gmap->shadow_lock);
253 /* Remove gmap from the pre-mm list */
254 spin_lock(&gmap->mm->context.lock);
255 list_del_rcu(&gmap->list);
256 if (list_empty(&gmap->mm->context.gmap_list))
258 else if (list_is_singular(&gmap->mm->context.gmap_list))
259 gmap_asce = list_first_entry(&gmap->mm->context.gmap_list,
260 struct gmap, list)->asce;
263 WRITE_ONCE(gmap->mm->context.gmap_asce, gmap_asce);
264 spin_unlock(&gmap->mm->context.lock);
269 EXPORT_SYMBOL_GPL(gmap_remove);
272 * gmap_enable - switch primary space to the guest address space
273 * @gmap: pointer to the guest address space structure
275 void gmap_enable(struct gmap *gmap)
277 S390_lowcore.gmap = (unsigned long) gmap;
279 EXPORT_SYMBOL_GPL(gmap_enable);
282 * gmap_disable - switch back to the standard primary address space
283 * @gmap: pointer to the guest address space structure
285 void gmap_disable(struct gmap *gmap)
287 S390_lowcore.gmap = 0UL;
289 EXPORT_SYMBOL_GPL(gmap_disable);
292 * gmap_get_enabled - get a pointer to the currently enabled gmap
294 * Returns a pointer to the currently enabled gmap. 0 if none is enabled.
296 struct gmap *gmap_get_enabled(void)
298 return (struct gmap *) S390_lowcore.gmap;
300 EXPORT_SYMBOL_GPL(gmap_get_enabled);
303 * gmap_alloc_table is assumed to be called with mmap_sem held
305 static int gmap_alloc_table(struct gmap *gmap, unsigned long *table,
306 unsigned long init, unsigned long gaddr)
311 /* since we dont free the gmap table until gmap_free we can unlock */
312 page = alloc_pages(GFP_KERNEL, CRST_ALLOC_ORDER);
315 new = (unsigned long *) page_to_phys(page);
316 crst_table_init(new, init);
317 spin_lock(&gmap->guest_table_lock);
318 if (*table & _REGION_ENTRY_INVALID) {
319 list_add(&page->lru, &gmap->crst_list);
320 *table = (unsigned long) new | _REGION_ENTRY_LENGTH |
321 (*table & _REGION_ENTRY_TYPE_MASK);
325 spin_unlock(&gmap->guest_table_lock);
327 __free_pages(page, CRST_ALLOC_ORDER);
332 * __gmap_segment_gaddr - find virtual address from segment pointer
333 * @entry: pointer to a segment table entry in the guest address space
335 * Returns the virtual address in the guest address space for the segment
337 static unsigned long __gmap_segment_gaddr(unsigned long *entry)
340 unsigned long offset, mask;
342 offset = (unsigned long) entry / sizeof(unsigned long);
343 offset = (offset & (PTRS_PER_PMD - 1)) * PMD_SIZE;
344 mask = ~(PTRS_PER_PMD * sizeof(pmd_t) - 1);
345 page = virt_to_page((void *)((unsigned long) entry & mask));
346 return page->index + offset;
350 * __gmap_unlink_by_vmaddr - unlink a single segment via a host address
351 * @gmap: pointer to the guest address space structure
352 * @vmaddr: address in the host process address space
354 * Returns 1 if a TLB flush is required
356 static int __gmap_unlink_by_vmaddr(struct gmap *gmap, unsigned long vmaddr)
358 unsigned long *entry;
361 BUG_ON(gmap_is_shadow(gmap));
362 spin_lock(&gmap->guest_table_lock);
363 entry = radix_tree_delete(&gmap->host_to_guest, vmaddr >> PMD_SHIFT);
365 flush = (*entry != _SEGMENT_ENTRY_EMPTY);
366 *entry = _SEGMENT_ENTRY_EMPTY;
368 spin_unlock(&gmap->guest_table_lock);
373 * __gmap_unmap_by_gaddr - unmap a single segment via a guest address
374 * @gmap: pointer to the guest address space structure
375 * @gaddr: address in the guest address space
377 * Returns 1 if a TLB flush is required
379 static int __gmap_unmap_by_gaddr(struct gmap *gmap, unsigned long gaddr)
381 unsigned long vmaddr;
383 vmaddr = (unsigned long) radix_tree_delete(&gmap->guest_to_host,
385 return vmaddr ? __gmap_unlink_by_vmaddr(gmap, vmaddr) : 0;
389 * gmap_unmap_segment - unmap segment from the guest address space
390 * @gmap: pointer to the guest address space structure
391 * @to: address in the guest address space
392 * @len: length of the memory area to unmap
394 * Returns 0 if the unmap succeeded, -EINVAL if not.
396 int gmap_unmap_segment(struct gmap *gmap, unsigned long to, unsigned long len)
401 BUG_ON(gmap_is_shadow(gmap));
402 if ((to | len) & (PMD_SIZE - 1))
404 if (len == 0 || to + len < to)
408 down_write(&gmap->mm->mmap_sem);
409 for (off = 0; off < len; off += PMD_SIZE)
410 flush |= __gmap_unmap_by_gaddr(gmap, to + off);
411 up_write(&gmap->mm->mmap_sem);
413 gmap_flush_tlb(gmap);
416 EXPORT_SYMBOL_GPL(gmap_unmap_segment);
419 * gmap_map_segment - map a segment to the guest address space
420 * @gmap: pointer to the guest address space structure
421 * @from: source address in the parent address space
422 * @to: target address in the guest address space
423 * @len: length of the memory area to map
425 * Returns 0 if the mmap succeeded, -EINVAL or -ENOMEM if not.
427 int gmap_map_segment(struct gmap *gmap, unsigned long from,
428 unsigned long to, unsigned long len)
433 BUG_ON(gmap_is_shadow(gmap));
434 if ((from | to | len) & (PMD_SIZE - 1))
436 if (len == 0 || from + len < from || to + len < to ||
437 from + len - 1 > TASK_SIZE_MAX || to + len - 1 > gmap->asce_end)
441 down_write(&gmap->mm->mmap_sem);
442 for (off = 0; off < len; off += PMD_SIZE) {
443 /* Remove old translation */
444 flush |= __gmap_unmap_by_gaddr(gmap, to + off);
445 /* Store new translation */
446 if (radix_tree_insert(&gmap->guest_to_host,
447 (to + off) >> PMD_SHIFT,
448 (void *) from + off))
451 up_write(&gmap->mm->mmap_sem);
453 gmap_flush_tlb(gmap);
456 gmap_unmap_segment(gmap, to, len);
459 EXPORT_SYMBOL_GPL(gmap_map_segment);
462 * __gmap_translate - translate a guest address to a user space address
463 * @gmap: pointer to guest mapping meta data structure
464 * @gaddr: guest address
466 * Returns user space address which corresponds to the guest address or
467 * -EFAULT if no such mapping exists.
468 * This function does not establish potentially missing page table entries.
469 * The mmap_sem of the mm that belongs to the address space must be held
470 * when this function gets called.
472 * Note: Can also be called for shadow gmaps.
474 unsigned long __gmap_translate(struct gmap *gmap, unsigned long gaddr)
476 unsigned long vmaddr;
478 vmaddr = (unsigned long)
479 radix_tree_lookup(&gmap->guest_to_host, gaddr >> PMD_SHIFT);
480 /* Note: guest_to_host is empty for a shadow gmap */
481 return vmaddr ? (vmaddr | (gaddr & ~PMD_MASK)) : -EFAULT;
483 EXPORT_SYMBOL_GPL(__gmap_translate);
486 * gmap_translate - translate a guest address to a user space address
487 * @gmap: pointer to guest mapping meta data structure
488 * @gaddr: guest address
490 * Returns user space address which corresponds to the guest address or
491 * -EFAULT if no such mapping exists.
492 * This function does not establish potentially missing page table entries.
494 unsigned long gmap_translate(struct gmap *gmap, unsigned long gaddr)
498 down_read(&gmap->mm->mmap_sem);
499 rc = __gmap_translate(gmap, gaddr);
500 up_read(&gmap->mm->mmap_sem);
503 EXPORT_SYMBOL_GPL(gmap_translate);
506 * gmap_unlink - disconnect a page table from the gmap shadow tables
507 * @gmap: pointer to guest mapping meta data structure
508 * @table: pointer to the host page table
509 * @vmaddr: vm address associated with the host page table
511 void gmap_unlink(struct mm_struct *mm, unsigned long *table,
512 unsigned long vmaddr)
518 list_for_each_entry_rcu(gmap, &mm->context.gmap_list, list) {
519 flush = __gmap_unlink_by_vmaddr(gmap, vmaddr);
521 gmap_flush_tlb(gmap);
526 static void gmap_pmdp_xchg(struct gmap *gmap, pmd_t *old, pmd_t new,
527 unsigned long gaddr);
530 * gmap_link - set up shadow page tables to connect a host to a guest address
531 * @gmap: pointer to guest mapping meta data structure
532 * @gaddr: guest address
533 * @vmaddr: vm address
535 * Returns 0 on success, -ENOMEM for out of memory conditions, and -EFAULT
536 * if the vm address is already mapped to a different guest segment.
537 * The mmap_sem of the mm that belongs to the address space must be held
538 * when this function gets called.
540 int __gmap_link(struct gmap *gmap, unsigned long gaddr, unsigned long vmaddr)
542 struct mm_struct *mm;
543 unsigned long *table;
552 BUG_ON(gmap_is_shadow(gmap));
553 /* Create higher level tables in the gmap page table */
555 if ((gmap->asce & _ASCE_TYPE_MASK) >= _ASCE_TYPE_REGION1) {
556 table += (gaddr & _REGION1_INDEX) >> _REGION1_SHIFT;
557 if ((*table & _REGION_ENTRY_INVALID) &&
558 gmap_alloc_table(gmap, table, _REGION2_ENTRY_EMPTY,
559 gaddr & _REGION1_MASK))
561 table = (unsigned long *)(*table & _REGION_ENTRY_ORIGIN);
563 if ((gmap->asce & _ASCE_TYPE_MASK) >= _ASCE_TYPE_REGION2) {
564 table += (gaddr & _REGION2_INDEX) >> _REGION2_SHIFT;
565 if ((*table & _REGION_ENTRY_INVALID) &&
566 gmap_alloc_table(gmap, table, _REGION3_ENTRY_EMPTY,
567 gaddr & _REGION2_MASK))
569 table = (unsigned long *)(*table & _REGION_ENTRY_ORIGIN);
571 if ((gmap->asce & _ASCE_TYPE_MASK) >= _ASCE_TYPE_REGION3) {
572 table += (gaddr & _REGION3_INDEX) >> _REGION3_SHIFT;
573 if ((*table & _REGION_ENTRY_INVALID) &&
574 gmap_alloc_table(gmap, table, _SEGMENT_ENTRY_EMPTY,
575 gaddr & _REGION3_MASK))
577 table = (unsigned long *)(*table & _REGION_ENTRY_ORIGIN);
579 table += (gaddr & _SEGMENT_INDEX) >> _SEGMENT_SHIFT;
580 /* Walk the parent mm page table */
582 pgd = pgd_offset(mm, vmaddr);
583 VM_BUG_ON(pgd_none(*pgd));
584 p4d = p4d_offset(pgd, vmaddr);
585 VM_BUG_ON(p4d_none(*p4d));
586 pud = pud_offset(p4d, vmaddr);
587 VM_BUG_ON(pud_none(*pud));
588 /* large puds cannot yet be handled */
591 pmd = pmd_offset(pud, vmaddr);
592 VM_BUG_ON(pmd_none(*pmd));
593 /* Are we allowed to use huge pages? */
594 if (pmd_large(*pmd) && !gmap->mm->context.allow_gmap_hpage_1m)
596 /* Link gmap segment table entry location to page table. */
597 rc = radix_tree_preload(GFP_KERNEL);
600 ptl = pmd_lock(mm, pmd);
601 spin_lock(&gmap->guest_table_lock);
602 if (*table == _SEGMENT_ENTRY_EMPTY) {
603 rc = radix_tree_insert(&gmap->host_to_guest,
604 vmaddr >> PMD_SHIFT, table);
606 if (pmd_large(*pmd)) {
607 *table = (pmd_val(*pmd) &
608 _SEGMENT_ENTRY_HARDWARE_BITS_LARGE)
609 | _SEGMENT_ENTRY_GMAP_UC;
611 *table = pmd_val(*pmd) &
612 _SEGMENT_ENTRY_HARDWARE_BITS;
614 } else if (*table & _SEGMENT_ENTRY_PROTECT &&
615 !(pmd_val(*pmd) & _SEGMENT_ENTRY_PROTECT)) {
616 unprot = (u64)*table;
617 unprot &= ~_SEGMENT_ENTRY_PROTECT;
618 unprot |= _SEGMENT_ENTRY_GMAP_UC;
619 gmap_pmdp_xchg(gmap, (pmd_t *)table, __pmd(unprot), gaddr);
621 spin_unlock(&gmap->guest_table_lock);
623 radix_tree_preload_end();
628 * gmap_fault - resolve a fault on a guest address
629 * @gmap: pointer to guest mapping meta data structure
630 * @gaddr: guest address
631 * @fault_flags: flags to pass down to handle_mm_fault()
633 * Returns 0 on success, -ENOMEM for out of memory conditions, and -EFAULT
634 * if the vm address is already mapped to a different guest segment.
636 int gmap_fault(struct gmap *gmap, unsigned long gaddr,
637 unsigned int fault_flags)
639 unsigned long vmaddr;
643 down_read(&gmap->mm->mmap_sem);
647 vmaddr = __gmap_translate(gmap, gaddr);
648 if (IS_ERR_VALUE(vmaddr)) {
652 if (fixup_user_fault(current, gmap->mm, vmaddr, fault_flags,
658 * In the case that fixup_user_fault unlocked the mmap_sem during
659 * faultin redo __gmap_translate to not race with a map/unmap_segment.
664 rc = __gmap_link(gmap, gaddr, vmaddr);
666 up_read(&gmap->mm->mmap_sem);
669 EXPORT_SYMBOL_GPL(gmap_fault);
672 * this function is assumed to be called with mmap_sem held
674 void __gmap_zap(struct gmap *gmap, unsigned long gaddr)
676 unsigned long vmaddr;
680 /* Find the vm address for the guest address */
681 vmaddr = (unsigned long) radix_tree_lookup(&gmap->guest_to_host,
684 vmaddr |= gaddr & ~PMD_MASK;
685 /* Get pointer to the page table entry */
686 ptep = get_locked_pte(gmap->mm, vmaddr, &ptl);
688 ptep_zap_unused(gmap->mm, vmaddr, ptep, 0);
689 pte_unmap_unlock(ptep, ptl);
692 EXPORT_SYMBOL_GPL(__gmap_zap);
694 void gmap_discard(struct gmap *gmap, unsigned long from, unsigned long to)
696 unsigned long gaddr, vmaddr, size;
697 struct vm_area_struct *vma;
699 down_read(&gmap->mm->mmap_sem);
700 for (gaddr = from; gaddr < to;
701 gaddr = (gaddr + PMD_SIZE) & PMD_MASK) {
702 /* Find the vm address for the guest address */
703 vmaddr = (unsigned long)
704 radix_tree_lookup(&gmap->guest_to_host,
708 vmaddr |= gaddr & ~PMD_MASK;
709 /* Find vma in the parent mm */
710 vma = find_vma(gmap->mm, vmaddr);
714 * We do not discard pages that are backed by
715 * hugetlbfs, so we don't have to refault them.
717 if (is_vm_hugetlb_page(vma))
719 size = min(to - gaddr, PMD_SIZE - (gaddr & ~PMD_MASK));
720 zap_page_range(vma, vmaddr, size);
722 up_read(&gmap->mm->mmap_sem);
724 EXPORT_SYMBOL_GPL(gmap_discard);
726 static LIST_HEAD(gmap_notifier_list);
727 static DEFINE_SPINLOCK(gmap_notifier_lock);
730 * gmap_register_pte_notifier - register a pte invalidation callback
731 * @nb: pointer to the gmap notifier block
733 void gmap_register_pte_notifier(struct gmap_notifier *nb)
735 spin_lock(&gmap_notifier_lock);
736 list_add_rcu(&nb->list, &gmap_notifier_list);
737 spin_unlock(&gmap_notifier_lock);
739 EXPORT_SYMBOL_GPL(gmap_register_pte_notifier);
742 * gmap_unregister_pte_notifier - remove a pte invalidation callback
743 * @nb: pointer to the gmap notifier block
745 void gmap_unregister_pte_notifier(struct gmap_notifier *nb)
747 spin_lock(&gmap_notifier_lock);
748 list_del_rcu(&nb->list);
749 spin_unlock(&gmap_notifier_lock);
752 EXPORT_SYMBOL_GPL(gmap_unregister_pte_notifier);
755 * gmap_call_notifier - call all registered invalidation callbacks
756 * @gmap: pointer to guest mapping meta data structure
757 * @start: start virtual address in the guest address space
758 * @end: end virtual address in the guest address space
760 static void gmap_call_notifier(struct gmap *gmap, unsigned long start,
763 struct gmap_notifier *nb;
765 list_for_each_entry(nb, &gmap_notifier_list, list)
766 nb->notifier_call(gmap, start, end);
770 * gmap_table_walk - walk the gmap page tables
771 * @gmap: pointer to guest mapping meta data structure
772 * @gaddr: virtual address in the guest address space
773 * @level: page table level to stop at
775 * Returns a table entry pointer for the given guest address and @level
776 * @level=0 : returns a pointer to a page table table entry (or NULL)
777 * @level=1 : returns a pointer to a segment table entry (or NULL)
778 * @level=2 : returns a pointer to a region-3 table entry (or NULL)
779 * @level=3 : returns a pointer to a region-2 table entry (or NULL)
780 * @level=4 : returns a pointer to a region-1 table entry (or NULL)
782 * Returns NULL if the gmap page tables could not be walked to the
785 * Note: Can also be called for shadow gmaps.
787 static inline unsigned long *gmap_table_walk(struct gmap *gmap,
788 unsigned long gaddr, int level)
790 const int asce_type = gmap->asce & _ASCE_TYPE_MASK;
791 unsigned long *table;
793 if ((gmap->asce & _ASCE_TYPE_MASK) + 4 < (level * 4))
795 if (gmap_is_shadow(gmap) && gmap->removed)
798 if (asce_type != _ASCE_TYPE_REGION1 &&
799 gaddr & (-1UL << (31 + (asce_type >> 2) * 11)))
803 switch (gmap->asce & _ASCE_TYPE_MASK) {
804 case _ASCE_TYPE_REGION1:
805 table += (gaddr & _REGION1_INDEX) >> _REGION1_SHIFT;
808 if (*table & _REGION_ENTRY_INVALID)
810 table = (unsigned long *)(*table & _REGION_ENTRY_ORIGIN);
812 case _ASCE_TYPE_REGION2:
813 table += (gaddr & _REGION2_INDEX) >> _REGION2_SHIFT;
816 if (*table & _REGION_ENTRY_INVALID)
818 table = (unsigned long *)(*table & _REGION_ENTRY_ORIGIN);
820 case _ASCE_TYPE_REGION3:
821 table += (gaddr & _REGION3_INDEX) >> _REGION3_SHIFT;
824 if (*table & _REGION_ENTRY_INVALID)
826 table = (unsigned long *)(*table & _REGION_ENTRY_ORIGIN);
828 case _ASCE_TYPE_SEGMENT:
829 table += (gaddr & _SEGMENT_INDEX) >> _SEGMENT_SHIFT;
832 if (*table & _REGION_ENTRY_INVALID)
834 table = (unsigned long *)(*table & _SEGMENT_ENTRY_ORIGIN);
835 table += (gaddr & _PAGE_INDEX) >> _PAGE_SHIFT;
841 * gmap_pte_op_walk - walk the gmap page table, get the page table lock
842 * and return the pte pointer
843 * @gmap: pointer to guest mapping meta data structure
844 * @gaddr: virtual address in the guest address space
845 * @ptl: pointer to the spinlock pointer
847 * Returns a pointer to the locked pte for a guest address, or NULL
849 static pte_t *gmap_pte_op_walk(struct gmap *gmap, unsigned long gaddr,
852 unsigned long *table;
854 BUG_ON(gmap_is_shadow(gmap));
855 /* Walk the gmap page table, lock and get pte pointer */
856 table = gmap_table_walk(gmap, gaddr, 1); /* get segment pointer */
857 if (!table || *table & _SEGMENT_ENTRY_INVALID)
859 return pte_alloc_map_lock(gmap->mm, (pmd_t *) table, gaddr, ptl);
863 * gmap_pte_op_fixup - force a page in and connect the gmap page table
864 * @gmap: pointer to guest mapping meta data structure
865 * @gaddr: virtual address in the guest address space
866 * @vmaddr: address in the host process address space
867 * @prot: indicates access rights: PROT_NONE, PROT_READ or PROT_WRITE
869 * Returns 0 if the caller can retry __gmap_translate (might fail again),
870 * -ENOMEM if out of memory and -EFAULT if anything goes wrong while fixing
871 * up or connecting the gmap page table.
873 static int gmap_pte_op_fixup(struct gmap *gmap, unsigned long gaddr,
874 unsigned long vmaddr, int prot)
876 struct mm_struct *mm = gmap->mm;
877 unsigned int fault_flags;
878 bool unlocked = false;
880 BUG_ON(gmap_is_shadow(gmap));
881 fault_flags = (prot == PROT_WRITE) ? FAULT_FLAG_WRITE : 0;
882 if (fixup_user_fault(current, mm, vmaddr, fault_flags, &unlocked))
885 /* lost mmap_sem, caller has to retry __gmap_translate */
887 /* Connect the page tables */
888 return __gmap_link(gmap, gaddr, vmaddr);
892 * gmap_pte_op_end - release the page table lock
893 * @ptl: pointer to the spinlock pointer
895 static void gmap_pte_op_end(spinlock_t *ptl)
902 * gmap_pmd_op_walk - walk the gmap tables, get the guest table lock
903 * and return the pmd pointer
904 * @gmap: pointer to guest mapping meta data structure
905 * @gaddr: virtual address in the guest address space
907 * Returns a pointer to the pmd for a guest address, or NULL
909 static inline pmd_t *gmap_pmd_op_walk(struct gmap *gmap, unsigned long gaddr)
913 BUG_ON(gmap_is_shadow(gmap));
914 spin_lock(&gmap->guest_table_lock);
915 pmdp = (pmd_t *) gmap_table_walk(gmap, gaddr, 1);
917 if (!pmdp || pmd_none(*pmdp)) {
918 spin_unlock(&gmap->guest_table_lock);
922 /* 4k page table entries are locked via the pte (pte_alloc_map_lock). */
923 if (!pmd_large(*pmdp))
924 spin_unlock(&gmap->guest_table_lock);
929 * gmap_pmd_op_end - release the guest_table_lock if needed
930 * @gmap: pointer to the guest mapping meta data structure
931 * @pmdp: pointer to the pmd
933 static inline void gmap_pmd_op_end(struct gmap *gmap, pmd_t *pmdp)
935 if (pmd_large(*pmdp))
936 spin_unlock(&gmap->guest_table_lock);
940 * gmap_protect_pmd - remove access rights to memory and set pmd notification bits
941 * @pmdp: pointer to the pmd to be protected
942 * @prot: indicates access rights: PROT_NONE, PROT_READ or PROT_WRITE
943 * @bits: notification bits to set
946 * 0 if successfully protected
947 * -EAGAIN if a fixup is needed
948 * -EINVAL if unsupported notifier bits have been specified
950 * Expected to be called with sg->mm->mmap_sem in read and
951 * guest_table_lock held.
953 static int gmap_protect_pmd(struct gmap *gmap, unsigned long gaddr,
954 pmd_t *pmdp, int prot, unsigned long bits)
956 int pmd_i = pmd_val(*pmdp) & _SEGMENT_ENTRY_INVALID;
957 int pmd_p = pmd_val(*pmdp) & _SEGMENT_ENTRY_PROTECT;
961 if ((pmd_i && (prot != PROT_NONE)) || (pmd_p && (prot == PROT_WRITE)))
964 if (prot == PROT_NONE && !pmd_i) {
965 pmd_val(new) |= _SEGMENT_ENTRY_INVALID;
966 gmap_pmdp_xchg(gmap, pmdp, new, gaddr);
969 if (prot == PROT_READ && !pmd_p) {
970 pmd_val(new) &= ~_SEGMENT_ENTRY_INVALID;
971 pmd_val(new) |= _SEGMENT_ENTRY_PROTECT;
972 gmap_pmdp_xchg(gmap, pmdp, new, gaddr);
975 if (bits & GMAP_NOTIFY_MPROT)
976 pmd_val(*pmdp) |= _SEGMENT_ENTRY_GMAP_IN;
978 /* Shadow GMAP protection needs split PMDs */
979 if (bits & GMAP_NOTIFY_SHADOW)
986 * gmap_protect_pte - remove access rights to memory and set pgste bits
987 * @gmap: pointer to guest mapping meta data structure
988 * @gaddr: virtual address in the guest address space
989 * @pmdp: pointer to the pmd associated with the pte
990 * @prot: indicates access rights: PROT_NONE, PROT_READ or PROT_WRITE
991 * @bits: notification bits to set
993 * Returns 0 if successfully protected, -ENOMEM if out of memory and
994 * -EAGAIN if a fixup is needed.
996 * Expected to be called with sg->mm->mmap_sem in read
998 static int gmap_protect_pte(struct gmap *gmap, unsigned long gaddr,
999 pmd_t *pmdp, int prot, unsigned long bits)
1003 spinlock_t *ptl = NULL;
1004 unsigned long pbits = 0;
1006 if (pmd_val(*pmdp) & _SEGMENT_ENTRY_INVALID)
1009 ptep = pte_alloc_map_lock(gmap->mm, pmdp, gaddr, &ptl);
1013 pbits |= (bits & GMAP_NOTIFY_MPROT) ? PGSTE_IN_BIT : 0;
1014 pbits |= (bits & GMAP_NOTIFY_SHADOW) ? PGSTE_VSIE_BIT : 0;
1015 /* Protect and unlock. */
1016 rc = ptep_force_prot(gmap->mm, gaddr, ptep, prot, pbits);
1017 gmap_pte_op_end(ptl);
1022 * gmap_protect_range - remove access rights to memory and set pgste bits
1023 * @gmap: pointer to guest mapping meta data structure
1024 * @gaddr: virtual address in the guest address space
1025 * @len: size of area
1026 * @prot: indicates access rights: PROT_NONE, PROT_READ or PROT_WRITE
1027 * @bits: pgste notification bits to set
1029 * Returns 0 if successfully protected, -ENOMEM if out of memory and
1030 * -EFAULT if gaddr is invalid (or mapping for shadows is missing).
1032 * Called with sg->mm->mmap_sem in read.
1034 static int gmap_protect_range(struct gmap *gmap, unsigned long gaddr,
1035 unsigned long len, int prot, unsigned long bits)
1037 unsigned long vmaddr, dist;
1041 BUG_ON(gmap_is_shadow(gmap));
1044 pmdp = gmap_pmd_op_walk(gmap, gaddr);
1046 if (!pmd_large(*pmdp)) {
1047 rc = gmap_protect_pte(gmap, gaddr, pmdp, prot,
1054 rc = gmap_protect_pmd(gmap, gaddr, pmdp, prot,
1057 dist = HPAGE_SIZE - (gaddr & ~HPAGE_MASK);
1058 len = len < dist ? 0 : len - dist;
1059 gaddr = (gaddr & HPAGE_MASK) + HPAGE_SIZE;
1062 gmap_pmd_op_end(gmap, pmdp);
1068 /* -EAGAIN, fixup of userspace mm and gmap */
1069 vmaddr = __gmap_translate(gmap, gaddr);
1070 if (IS_ERR_VALUE(vmaddr))
1072 rc = gmap_pte_op_fixup(gmap, gaddr, vmaddr, prot);
1081 * gmap_mprotect_notify - change access rights for a range of ptes and
1082 * call the notifier if any pte changes again
1083 * @gmap: pointer to guest mapping meta data structure
1084 * @gaddr: virtual address in the guest address space
1085 * @len: size of area
1086 * @prot: indicates access rights: PROT_NONE, PROT_READ or PROT_WRITE
1088 * Returns 0 if for each page in the given range a gmap mapping exists,
1089 * the new access rights could be set and the notifier could be armed.
1090 * If the gmap mapping is missing for one or more pages -EFAULT is
1091 * returned. If no memory could be allocated -ENOMEM is returned.
1092 * This function establishes missing page table entries.
1094 int gmap_mprotect_notify(struct gmap *gmap, unsigned long gaddr,
1095 unsigned long len, int prot)
1099 if ((gaddr & ~PAGE_MASK) || (len & ~PAGE_MASK) || gmap_is_shadow(gmap))
1101 if (!MACHINE_HAS_ESOP && prot == PROT_READ)
1103 down_read(&gmap->mm->mmap_sem);
1104 rc = gmap_protect_range(gmap, gaddr, len, prot, GMAP_NOTIFY_MPROT);
1105 up_read(&gmap->mm->mmap_sem);
1108 EXPORT_SYMBOL_GPL(gmap_mprotect_notify);
1111 * gmap_read_table - get an unsigned long value from a guest page table using
1112 * absolute addressing, without marking the page referenced.
1113 * @gmap: pointer to guest mapping meta data structure
1114 * @gaddr: virtual address in the guest address space
1115 * @val: pointer to the unsigned long value to return
1117 * Returns 0 if the value was read, -ENOMEM if out of memory and -EFAULT
1118 * if reading using the virtual address failed. -EINVAL if called on a gmap
1121 * Called with gmap->mm->mmap_sem in read.
1123 int gmap_read_table(struct gmap *gmap, unsigned long gaddr, unsigned long *val)
1125 unsigned long address, vmaddr;
1130 if (gmap_is_shadow(gmap))
1135 ptep = gmap_pte_op_walk(gmap, gaddr, &ptl);
1138 if (pte_present(pte) && (pte_val(pte) & _PAGE_READ)) {
1139 address = pte_val(pte) & PAGE_MASK;
1140 address += gaddr & ~PAGE_MASK;
1141 *val = *(unsigned long *) address;
1142 pte_val(*ptep) |= _PAGE_YOUNG;
1143 /* Do *NOT* clear the _PAGE_INVALID bit! */
1146 gmap_pte_op_end(ptl);
1150 vmaddr = __gmap_translate(gmap, gaddr);
1151 if (IS_ERR_VALUE(vmaddr)) {
1155 rc = gmap_pte_op_fixup(gmap, gaddr, vmaddr, PROT_READ);
1161 EXPORT_SYMBOL_GPL(gmap_read_table);
1164 * gmap_insert_rmap - add a rmap to the host_to_rmap radix tree
1165 * @sg: pointer to the shadow guest address space structure
1166 * @vmaddr: vm address associated with the rmap
1167 * @rmap: pointer to the rmap structure
1169 * Called with the sg->guest_table_lock
1171 static inline void gmap_insert_rmap(struct gmap *sg, unsigned long vmaddr,
1172 struct gmap_rmap *rmap)
1176 BUG_ON(!gmap_is_shadow(sg));
1177 slot = radix_tree_lookup_slot(&sg->host_to_rmap, vmaddr >> PAGE_SHIFT);
1179 rmap->next = radix_tree_deref_slot_protected(slot,
1180 &sg->guest_table_lock);
1181 radix_tree_replace_slot(&sg->host_to_rmap, slot, rmap);
1184 radix_tree_insert(&sg->host_to_rmap, vmaddr >> PAGE_SHIFT,
1190 * gmap_protect_rmap - restrict access rights to memory (RO) and create an rmap
1191 * @sg: pointer to the shadow guest address space structure
1192 * @raddr: rmap address in the shadow gmap
1193 * @paddr: address in the parent guest address space
1194 * @len: length of the memory area to protect
1196 * Returns 0 if successfully protected and the rmap was created, -ENOMEM
1197 * if out of memory and -EFAULT if paddr is invalid.
1199 static int gmap_protect_rmap(struct gmap *sg, unsigned long raddr,
1200 unsigned long paddr, unsigned long len)
1202 struct gmap *parent;
1203 struct gmap_rmap *rmap;
1204 unsigned long vmaddr;
1209 BUG_ON(!gmap_is_shadow(sg));
1210 parent = sg->parent;
1212 vmaddr = __gmap_translate(parent, paddr);
1213 if (IS_ERR_VALUE(vmaddr))
1215 rmap = kzalloc(sizeof(*rmap), GFP_KERNEL);
1218 rmap->raddr = raddr;
1219 rc = radix_tree_preload(GFP_KERNEL);
1225 ptep = gmap_pte_op_walk(parent, paddr, &ptl);
1227 spin_lock(&sg->guest_table_lock);
1228 rc = ptep_force_prot(parent->mm, paddr, ptep, PROT_READ,
1231 gmap_insert_rmap(sg, vmaddr, rmap);
1232 spin_unlock(&sg->guest_table_lock);
1233 gmap_pte_op_end(ptl);
1235 radix_tree_preload_end();
1238 rc = gmap_pte_op_fixup(parent, paddr, vmaddr, PROT_READ);
1249 #define _SHADOW_RMAP_MASK 0x7
1250 #define _SHADOW_RMAP_REGION1 0x5
1251 #define _SHADOW_RMAP_REGION2 0x4
1252 #define _SHADOW_RMAP_REGION3 0x3
1253 #define _SHADOW_RMAP_SEGMENT 0x2
1254 #define _SHADOW_RMAP_PGTABLE 0x1
1257 * gmap_idte_one - invalidate a single region or segment table entry
1258 * @asce: region or segment table *origin* + table-type bits
1259 * @vaddr: virtual address to identify the table entry to flush
1261 * The invalid bit of a single region or segment table entry is set
1262 * and the associated TLB entries depending on the entry are flushed.
1263 * The table-type of the @asce identifies the portion of the @vaddr
1264 * that is used as the invalidation index.
1266 static inline void gmap_idte_one(unsigned long asce, unsigned long vaddr)
1269 " .insn rrf,0xb98e0000,%0,%1,0,0"
1270 : : "a" (asce), "a" (vaddr) : "cc", "memory");
1274 * gmap_unshadow_page - remove a page from a shadow page table
1275 * @sg: pointer to the shadow guest address space structure
1276 * @raddr: rmap address in the shadow guest address space
1278 * Called with the sg->guest_table_lock
1280 static void gmap_unshadow_page(struct gmap *sg, unsigned long raddr)
1282 unsigned long *table;
1284 BUG_ON(!gmap_is_shadow(sg));
1285 table = gmap_table_walk(sg, raddr, 0); /* get page table pointer */
1286 if (!table || *table & _PAGE_INVALID)
1288 gmap_call_notifier(sg, raddr, raddr + _PAGE_SIZE - 1);
1289 ptep_unshadow_pte(sg->mm, raddr, (pte_t *) table);
1293 * __gmap_unshadow_pgt - remove all entries from a shadow page table
1294 * @sg: pointer to the shadow guest address space structure
1295 * @raddr: rmap address in the shadow guest address space
1296 * @pgt: pointer to the start of a shadow page table
1298 * Called with the sg->guest_table_lock
1300 static void __gmap_unshadow_pgt(struct gmap *sg, unsigned long raddr,
1305 BUG_ON(!gmap_is_shadow(sg));
1306 for (i = 0; i < _PAGE_ENTRIES; i++, raddr += _PAGE_SIZE)
1307 pgt[i] = _PAGE_INVALID;
1311 * gmap_unshadow_pgt - remove a shadow page table from a segment entry
1312 * @sg: pointer to the shadow guest address space structure
1313 * @raddr: address in the shadow guest address space
1315 * Called with the sg->guest_table_lock
1317 static void gmap_unshadow_pgt(struct gmap *sg, unsigned long raddr)
1319 unsigned long sto, *ste, *pgt;
1322 BUG_ON(!gmap_is_shadow(sg));
1323 ste = gmap_table_walk(sg, raddr, 1); /* get segment pointer */
1324 if (!ste || !(*ste & _SEGMENT_ENTRY_ORIGIN))
1326 gmap_call_notifier(sg, raddr, raddr + _SEGMENT_SIZE - 1);
1327 sto = (unsigned long) (ste - ((raddr & _SEGMENT_INDEX) >> _SEGMENT_SHIFT));
1328 gmap_idte_one(sto | _ASCE_TYPE_SEGMENT, raddr);
1329 pgt = (unsigned long *)(*ste & _SEGMENT_ENTRY_ORIGIN);
1330 *ste = _SEGMENT_ENTRY_EMPTY;
1331 __gmap_unshadow_pgt(sg, raddr, pgt);
1332 /* Free page table */
1333 page = pfn_to_page(__pa(pgt) >> PAGE_SHIFT);
1334 list_del(&page->lru);
1335 page_table_free_pgste(page);
1339 * __gmap_unshadow_sgt - remove all entries from a shadow segment table
1340 * @sg: pointer to the shadow guest address space structure
1341 * @raddr: rmap address in the shadow guest address space
1342 * @sgt: pointer to the start of a shadow segment table
1344 * Called with the sg->guest_table_lock
1346 static void __gmap_unshadow_sgt(struct gmap *sg, unsigned long raddr,
1353 BUG_ON(!gmap_is_shadow(sg));
1354 for (i = 0; i < _CRST_ENTRIES; i++, raddr += _SEGMENT_SIZE) {
1355 if (!(sgt[i] & _SEGMENT_ENTRY_ORIGIN))
1357 pgt = (unsigned long *)(sgt[i] & _REGION_ENTRY_ORIGIN);
1358 sgt[i] = _SEGMENT_ENTRY_EMPTY;
1359 __gmap_unshadow_pgt(sg, raddr, pgt);
1360 /* Free page table */
1361 page = pfn_to_page(__pa(pgt) >> PAGE_SHIFT);
1362 list_del(&page->lru);
1363 page_table_free_pgste(page);
1368 * gmap_unshadow_sgt - remove a shadow segment table from a region-3 entry
1369 * @sg: pointer to the shadow guest address space structure
1370 * @raddr: rmap address in the shadow guest address space
1372 * Called with the shadow->guest_table_lock
1374 static void gmap_unshadow_sgt(struct gmap *sg, unsigned long raddr)
1376 unsigned long r3o, *r3e, *sgt;
1379 BUG_ON(!gmap_is_shadow(sg));
1380 r3e = gmap_table_walk(sg, raddr, 2); /* get region-3 pointer */
1381 if (!r3e || !(*r3e & _REGION_ENTRY_ORIGIN))
1383 gmap_call_notifier(sg, raddr, raddr + _REGION3_SIZE - 1);
1384 r3o = (unsigned long) (r3e - ((raddr & _REGION3_INDEX) >> _REGION3_SHIFT));
1385 gmap_idte_one(r3o | _ASCE_TYPE_REGION3, raddr);
1386 sgt = (unsigned long *)(*r3e & _REGION_ENTRY_ORIGIN);
1387 *r3e = _REGION3_ENTRY_EMPTY;
1388 __gmap_unshadow_sgt(sg, raddr, sgt);
1389 /* Free segment table */
1390 page = pfn_to_page(__pa(sgt) >> PAGE_SHIFT);
1391 list_del(&page->lru);
1392 __free_pages(page, CRST_ALLOC_ORDER);
1396 * __gmap_unshadow_r3t - remove all entries from a shadow region-3 table
1397 * @sg: pointer to the shadow guest address space structure
1398 * @raddr: address in the shadow guest address space
1399 * @r3t: pointer to the start of a shadow region-3 table
1401 * Called with the sg->guest_table_lock
1403 static void __gmap_unshadow_r3t(struct gmap *sg, unsigned long raddr,
1410 BUG_ON(!gmap_is_shadow(sg));
1411 for (i = 0; i < _CRST_ENTRIES; i++, raddr += _REGION3_SIZE) {
1412 if (!(r3t[i] & _REGION_ENTRY_ORIGIN))
1414 sgt = (unsigned long *)(r3t[i] & _REGION_ENTRY_ORIGIN);
1415 r3t[i] = _REGION3_ENTRY_EMPTY;
1416 __gmap_unshadow_sgt(sg, raddr, sgt);
1417 /* Free segment table */
1418 page = pfn_to_page(__pa(sgt) >> PAGE_SHIFT);
1419 list_del(&page->lru);
1420 __free_pages(page, CRST_ALLOC_ORDER);
1425 * gmap_unshadow_r3t - remove a shadow region-3 table from a region-2 entry
1426 * @sg: pointer to the shadow guest address space structure
1427 * @raddr: rmap address in the shadow guest address space
1429 * Called with the sg->guest_table_lock
1431 static void gmap_unshadow_r3t(struct gmap *sg, unsigned long raddr)
1433 unsigned long r2o, *r2e, *r3t;
1436 BUG_ON(!gmap_is_shadow(sg));
1437 r2e = gmap_table_walk(sg, raddr, 3); /* get region-2 pointer */
1438 if (!r2e || !(*r2e & _REGION_ENTRY_ORIGIN))
1440 gmap_call_notifier(sg, raddr, raddr + _REGION2_SIZE - 1);
1441 r2o = (unsigned long) (r2e - ((raddr & _REGION2_INDEX) >> _REGION2_SHIFT));
1442 gmap_idte_one(r2o | _ASCE_TYPE_REGION2, raddr);
1443 r3t = (unsigned long *)(*r2e & _REGION_ENTRY_ORIGIN);
1444 *r2e = _REGION2_ENTRY_EMPTY;
1445 __gmap_unshadow_r3t(sg, raddr, r3t);
1446 /* Free region 3 table */
1447 page = pfn_to_page(__pa(r3t) >> PAGE_SHIFT);
1448 list_del(&page->lru);
1449 __free_pages(page, CRST_ALLOC_ORDER);
1453 * __gmap_unshadow_r2t - remove all entries from a shadow region-2 table
1454 * @sg: pointer to the shadow guest address space structure
1455 * @raddr: rmap address in the shadow guest address space
1456 * @r2t: pointer to the start of a shadow region-2 table
1458 * Called with the sg->guest_table_lock
1460 static void __gmap_unshadow_r2t(struct gmap *sg, unsigned long raddr,
1467 BUG_ON(!gmap_is_shadow(sg));
1468 for (i = 0; i < _CRST_ENTRIES; i++, raddr += _REGION2_SIZE) {
1469 if (!(r2t[i] & _REGION_ENTRY_ORIGIN))
1471 r3t = (unsigned long *)(r2t[i] & _REGION_ENTRY_ORIGIN);
1472 r2t[i] = _REGION2_ENTRY_EMPTY;
1473 __gmap_unshadow_r3t(sg, raddr, r3t);
1474 /* Free region 3 table */
1475 page = pfn_to_page(__pa(r3t) >> PAGE_SHIFT);
1476 list_del(&page->lru);
1477 __free_pages(page, CRST_ALLOC_ORDER);
1482 * gmap_unshadow_r2t - remove a shadow region-2 table from a region-1 entry
1483 * @sg: pointer to the shadow guest address space structure
1484 * @raddr: rmap address in the shadow guest address space
1486 * Called with the sg->guest_table_lock
1488 static void gmap_unshadow_r2t(struct gmap *sg, unsigned long raddr)
1490 unsigned long r1o, *r1e, *r2t;
1493 BUG_ON(!gmap_is_shadow(sg));
1494 r1e = gmap_table_walk(sg, raddr, 4); /* get region-1 pointer */
1495 if (!r1e || !(*r1e & _REGION_ENTRY_ORIGIN))
1497 gmap_call_notifier(sg, raddr, raddr + _REGION1_SIZE - 1);
1498 r1o = (unsigned long) (r1e - ((raddr & _REGION1_INDEX) >> _REGION1_SHIFT));
1499 gmap_idte_one(r1o | _ASCE_TYPE_REGION1, raddr);
1500 r2t = (unsigned long *)(*r1e & _REGION_ENTRY_ORIGIN);
1501 *r1e = _REGION1_ENTRY_EMPTY;
1502 __gmap_unshadow_r2t(sg, raddr, r2t);
1503 /* Free region 2 table */
1504 page = pfn_to_page(__pa(r2t) >> PAGE_SHIFT);
1505 list_del(&page->lru);
1506 __free_pages(page, CRST_ALLOC_ORDER);
1510 * __gmap_unshadow_r1t - remove all entries from a shadow region-1 table
1511 * @sg: pointer to the shadow guest address space structure
1512 * @raddr: rmap address in the shadow guest address space
1513 * @r1t: pointer to the start of a shadow region-1 table
1515 * Called with the shadow->guest_table_lock
1517 static void __gmap_unshadow_r1t(struct gmap *sg, unsigned long raddr,
1520 unsigned long asce, *r2t;
1524 BUG_ON(!gmap_is_shadow(sg));
1525 asce = (unsigned long) r1t | _ASCE_TYPE_REGION1;
1526 for (i = 0; i < _CRST_ENTRIES; i++, raddr += _REGION1_SIZE) {
1527 if (!(r1t[i] & _REGION_ENTRY_ORIGIN))
1529 r2t = (unsigned long *)(r1t[i] & _REGION_ENTRY_ORIGIN);
1530 __gmap_unshadow_r2t(sg, raddr, r2t);
1531 /* Clear entry and flush translation r1t -> r2t */
1532 gmap_idte_one(asce, raddr);
1533 r1t[i] = _REGION1_ENTRY_EMPTY;
1534 /* Free region 2 table */
1535 page = pfn_to_page(__pa(r2t) >> PAGE_SHIFT);
1536 list_del(&page->lru);
1537 __free_pages(page, CRST_ALLOC_ORDER);
1542 * gmap_unshadow - remove a shadow page table completely
1543 * @sg: pointer to the shadow guest address space structure
1545 * Called with sg->guest_table_lock
1547 static void gmap_unshadow(struct gmap *sg)
1549 unsigned long *table;
1551 BUG_ON(!gmap_is_shadow(sg));
1555 gmap_call_notifier(sg, 0, -1UL);
1557 table = (unsigned long *)(sg->asce & _ASCE_ORIGIN);
1558 switch (sg->asce & _ASCE_TYPE_MASK) {
1559 case _ASCE_TYPE_REGION1:
1560 __gmap_unshadow_r1t(sg, 0, table);
1562 case _ASCE_TYPE_REGION2:
1563 __gmap_unshadow_r2t(sg, 0, table);
1565 case _ASCE_TYPE_REGION3:
1566 __gmap_unshadow_r3t(sg, 0, table);
1568 case _ASCE_TYPE_SEGMENT:
1569 __gmap_unshadow_sgt(sg, 0, table);
1575 * gmap_find_shadow - find a specific asce in the list of shadow tables
1576 * @parent: pointer to the parent gmap
1577 * @asce: ASCE for which the shadow table is created
1578 * @edat_level: edat level to be used for the shadow translation
1580 * Returns the pointer to a gmap if a shadow table with the given asce is
1581 * already available, ERR_PTR(-EAGAIN) if another one is just being created,
1584 static struct gmap *gmap_find_shadow(struct gmap *parent, unsigned long asce,
1589 list_for_each_entry(sg, &parent->children, list) {
1590 if (sg->orig_asce != asce || sg->edat_level != edat_level ||
1593 if (!sg->initialized)
1594 return ERR_PTR(-EAGAIN);
1595 atomic_inc(&sg->ref_count);
1602 * gmap_shadow_valid - check if a shadow guest address space matches the
1603 * given properties and is still valid
1604 * @sg: pointer to the shadow guest address space structure
1605 * @asce: ASCE for which the shadow table is requested
1606 * @edat_level: edat level to be used for the shadow translation
1608 * Returns 1 if the gmap shadow is still valid and matches the given
1609 * properties, the caller can continue using it. Returns 0 otherwise, the
1610 * caller has to request a new shadow gmap in this case.
1613 int gmap_shadow_valid(struct gmap *sg, unsigned long asce, int edat_level)
1617 return sg->orig_asce == asce && sg->edat_level == edat_level;
1619 EXPORT_SYMBOL_GPL(gmap_shadow_valid);
1622 * gmap_shadow - create/find a shadow guest address space
1623 * @parent: pointer to the parent gmap
1624 * @asce: ASCE for which the shadow table is created
1625 * @edat_level: edat level to be used for the shadow translation
1627 * The pages of the top level page table referred by the asce parameter
1628 * will be set to read-only and marked in the PGSTEs of the kvm process.
1629 * The shadow table will be removed automatically on any change to the
1630 * PTE mapping for the source table.
1632 * Returns a guest address space structure, ERR_PTR(-ENOMEM) if out of memory,
1633 * ERR_PTR(-EAGAIN) if the caller has to retry and ERR_PTR(-EFAULT) if the
1634 * parent gmap table could not be protected.
1636 struct gmap *gmap_shadow(struct gmap *parent, unsigned long asce,
1639 struct gmap *sg, *new;
1640 unsigned long limit;
1643 BUG_ON(parent->mm->context.allow_gmap_hpage_1m);
1644 BUG_ON(gmap_is_shadow(parent));
1645 spin_lock(&parent->shadow_lock);
1646 sg = gmap_find_shadow(parent, asce, edat_level);
1647 spin_unlock(&parent->shadow_lock);
1650 /* Create a new shadow gmap */
1651 limit = -1UL >> (33 - (((asce & _ASCE_TYPE_MASK) >> 2) * 11));
1652 if (asce & _ASCE_REAL_SPACE)
1654 new = gmap_alloc(limit);
1656 return ERR_PTR(-ENOMEM);
1657 new->mm = parent->mm;
1658 new->parent = gmap_get(parent);
1659 new->orig_asce = asce;
1660 new->edat_level = edat_level;
1661 new->initialized = false;
1662 spin_lock(&parent->shadow_lock);
1663 /* Recheck if another CPU created the same shadow */
1664 sg = gmap_find_shadow(parent, asce, edat_level);
1666 spin_unlock(&parent->shadow_lock);
1670 if (asce & _ASCE_REAL_SPACE) {
1671 /* only allow one real-space gmap shadow */
1672 list_for_each_entry(sg, &parent->children, list) {
1673 if (sg->orig_asce & _ASCE_REAL_SPACE) {
1674 spin_lock(&sg->guest_table_lock);
1676 spin_unlock(&sg->guest_table_lock);
1677 list_del(&sg->list);
1683 atomic_set(&new->ref_count, 2);
1684 list_add(&new->list, &parent->children);
1685 if (asce & _ASCE_REAL_SPACE) {
1686 /* nothing to protect, return right away */
1687 new->initialized = true;
1688 spin_unlock(&parent->shadow_lock);
1691 spin_unlock(&parent->shadow_lock);
1692 /* protect after insertion, so it will get properly invalidated */
1693 down_read(&parent->mm->mmap_sem);
1694 rc = gmap_protect_range(parent, asce & _ASCE_ORIGIN,
1695 ((asce & _ASCE_TABLE_LENGTH) + 1) * PAGE_SIZE,
1696 PROT_READ, GMAP_NOTIFY_SHADOW);
1697 up_read(&parent->mm->mmap_sem);
1698 spin_lock(&parent->shadow_lock);
1699 new->initialized = true;
1701 list_del(&new->list);
1705 spin_unlock(&parent->shadow_lock);
1708 EXPORT_SYMBOL_GPL(gmap_shadow);
1711 * gmap_shadow_r2t - create an empty shadow region 2 table
1712 * @sg: pointer to the shadow guest address space structure
1713 * @saddr: faulting address in the shadow gmap
1714 * @r2t: parent gmap address of the region 2 table to get shadowed
1715 * @fake: r2t references contiguous guest memory block, not a r2t
1717 * The r2t parameter specifies the address of the source table. The
1718 * four pages of the source table are made read-only in the parent gmap
1719 * address space. A write to the source table area @r2t will automatically
1720 * remove the shadow r2 table and all of its decendents.
1722 * Returns 0 if successfully shadowed or already shadowed, -EAGAIN if the
1723 * shadow table structure is incomplete, -ENOMEM if out of memory and
1724 * -EFAULT if an address in the parent gmap could not be resolved.
1726 * Called with sg->mm->mmap_sem in read.
1728 int gmap_shadow_r2t(struct gmap *sg, unsigned long saddr, unsigned long r2t,
1731 unsigned long raddr, origin, offset, len;
1732 unsigned long *s_r2t, *table;
1736 BUG_ON(!gmap_is_shadow(sg));
1737 /* Allocate a shadow region second table */
1738 page = alloc_pages(GFP_KERNEL, CRST_ALLOC_ORDER);
1741 page->index = r2t & _REGION_ENTRY_ORIGIN;
1743 page->index |= GMAP_SHADOW_FAKE_TABLE;
1744 s_r2t = (unsigned long *) page_to_phys(page);
1745 /* Install shadow region second table */
1746 spin_lock(&sg->guest_table_lock);
1747 table = gmap_table_walk(sg, saddr, 4); /* get region-1 pointer */
1749 rc = -EAGAIN; /* Race with unshadow */
1752 if (!(*table & _REGION_ENTRY_INVALID)) {
1753 rc = 0; /* Already established */
1755 } else if (*table & _REGION_ENTRY_ORIGIN) {
1756 rc = -EAGAIN; /* Race with shadow */
1759 crst_table_init(s_r2t, _REGION2_ENTRY_EMPTY);
1760 /* mark as invalid as long as the parent table is not protected */
1761 *table = (unsigned long) s_r2t | _REGION_ENTRY_LENGTH |
1762 _REGION_ENTRY_TYPE_R1 | _REGION_ENTRY_INVALID;
1763 if (sg->edat_level >= 1)
1764 *table |= (r2t & _REGION_ENTRY_PROTECT);
1765 list_add(&page->lru, &sg->crst_list);
1767 /* nothing to protect for fake tables */
1768 *table &= ~_REGION_ENTRY_INVALID;
1769 spin_unlock(&sg->guest_table_lock);
1772 spin_unlock(&sg->guest_table_lock);
1773 /* Make r2t read-only in parent gmap page table */
1774 raddr = (saddr & _REGION1_MASK) | _SHADOW_RMAP_REGION1;
1775 origin = r2t & _REGION_ENTRY_ORIGIN;
1776 offset = ((r2t & _REGION_ENTRY_OFFSET) >> 6) * PAGE_SIZE;
1777 len = ((r2t & _REGION_ENTRY_LENGTH) + 1) * PAGE_SIZE - offset;
1778 rc = gmap_protect_rmap(sg, raddr, origin + offset, len);
1779 spin_lock(&sg->guest_table_lock);
1781 table = gmap_table_walk(sg, saddr, 4);
1782 if (!table || (*table & _REGION_ENTRY_ORIGIN) !=
1783 (unsigned long) s_r2t)
1784 rc = -EAGAIN; /* Race with unshadow */
1786 *table &= ~_REGION_ENTRY_INVALID;
1788 gmap_unshadow_r2t(sg, raddr);
1790 spin_unlock(&sg->guest_table_lock);
1793 spin_unlock(&sg->guest_table_lock);
1794 __free_pages(page, CRST_ALLOC_ORDER);
1797 EXPORT_SYMBOL_GPL(gmap_shadow_r2t);
1800 * gmap_shadow_r3t - create a shadow region 3 table
1801 * @sg: pointer to the shadow guest address space structure
1802 * @saddr: faulting address in the shadow gmap
1803 * @r3t: parent gmap address of the region 3 table to get shadowed
1804 * @fake: r3t references contiguous guest memory block, not a r3t
1806 * Returns 0 if successfully shadowed or already shadowed, -EAGAIN if the
1807 * shadow table structure is incomplete, -ENOMEM if out of memory and
1808 * -EFAULT if an address in the parent gmap could not be resolved.
1810 * Called with sg->mm->mmap_sem in read.
1812 int gmap_shadow_r3t(struct gmap *sg, unsigned long saddr, unsigned long r3t,
1815 unsigned long raddr, origin, offset, len;
1816 unsigned long *s_r3t, *table;
1820 BUG_ON(!gmap_is_shadow(sg));
1821 /* Allocate a shadow region second table */
1822 page = alloc_pages(GFP_KERNEL, CRST_ALLOC_ORDER);
1825 page->index = r3t & _REGION_ENTRY_ORIGIN;
1827 page->index |= GMAP_SHADOW_FAKE_TABLE;
1828 s_r3t = (unsigned long *) page_to_phys(page);
1829 /* Install shadow region second table */
1830 spin_lock(&sg->guest_table_lock);
1831 table = gmap_table_walk(sg, saddr, 3); /* get region-2 pointer */
1833 rc = -EAGAIN; /* Race with unshadow */
1836 if (!(*table & _REGION_ENTRY_INVALID)) {
1837 rc = 0; /* Already established */
1839 } else if (*table & _REGION_ENTRY_ORIGIN) {
1840 rc = -EAGAIN; /* Race with shadow */
1843 crst_table_init(s_r3t, _REGION3_ENTRY_EMPTY);
1844 /* mark as invalid as long as the parent table is not protected */
1845 *table = (unsigned long) s_r3t | _REGION_ENTRY_LENGTH |
1846 _REGION_ENTRY_TYPE_R2 | _REGION_ENTRY_INVALID;
1847 if (sg->edat_level >= 1)
1848 *table |= (r3t & _REGION_ENTRY_PROTECT);
1849 list_add(&page->lru, &sg->crst_list);
1851 /* nothing to protect for fake tables */
1852 *table &= ~_REGION_ENTRY_INVALID;
1853 spin_unlock(&sg->guest_table_lock);
1856 spin_unlock(&sg->guest_table_lock);
1857 /* Make r3t read-only in parent gmap page table */
1858 raddr = (saddr & _REGION2_MASK) | _SHADOW_RMAP_REGION2;
1859 origin = r3t & _REGION_ENTRY_ORIGIN;
1860 offset = ((r3t & _REGION_ENTRY_OFFSET) >> 6) * PAGE_SIZE;
1861 len = ((r3t & _REGION_ENTRY_LENGTH) + 1) * PAGE_SIZE - offset;
1862 rc = gmap_protect_rmap(sg, raddr, origin + offset, len);
1863 spin_lock(&sg->guest_table_lock);
1865 table = gmap_table_walk(sg, saddr, 3);
1866 if (!table || (*table & _REGION_ENTRY_ORIGIN) !=
1867 (unsigned long) s_r3t)
1868 rc = -EAGAIN; /* Race with unshadow */
1870 *table &= ~_REGION_ENTRY_INVALID;
1872 gmap_unshadow_r3t(sg, raddr);
1874 spin_unlock(&sg->guest_table_lock);
1877 spin_unlock(&sg->guest_table_lock);
1878 __free_pages(page, CRST_ALLOC_ORDER);
1881 EXPORT_SYMBOL_GPL(gmap_shadow_r3t);
1884 * gmap_shadow_sgt - create a shadow segment table
1885 * @sg: pointer to the shadow guest address space structure
1886 * @saddr: faulting address in the shadow gmap
1887 * @sgt: parent gmap address of the segment table to get shadowed
1888 * @fake: sgt references contiguous guest memory block, not a sgt
1890 * Returns: 0 if successfully shadowed or already shadowed, -EAGAIN if the
1891 * shadow table structure is incomplete, -ENOMEM if out of memory and
1892 * -EFAULT if an address in the parent gmap could not be resolved.
1894 * Called with sg->mm->mmap_sem in read.
1896 int gmap_shadow_sgt(struct gmap *sg, unsigned long saddr, unsigned long sgt,
1899 unsigned long raddr, origin, offset, len;
1900 unsigned long *s_sgt, *table;
1904 BUG_ON(!gmap_is_shadow(sg) || (sgt & _REGION3_ENTRY_LARGE));
1905 /* Allocate a shadow segment table */
1906 page = alloc_pages(GFP_KERNEL, CRST_ALLOC_ORDER);
1909 page->index = sgt & _REGION_ENTRY_ORIGIN;
1911 page->index |= GMAP_SHADOW_FAKE_TABLE;
1912 s_sgt = (unsigned long *) page_to_phys(page);
1913 /* Install shadow region second table */
1914 spin_lock(&sg->guest_table_lock);
1915 table = gmap_table_walk(sg, saddr, 2); /* get region-3 pointer */
1917 rc = -EAGAIN; /* Race with unshadow */
1920 if (!(*table & _REGION_ENTRY_INVALID)) {
1921 rc = 0; /* Already established */
1923 } else if (*table & _REGION_ENTRY_ORIGIN) {
1924 rc = -EAGAIN; /* Race with shadow */
1927 crst_table_init(s_sgt, _SEGMENT_ENTRY_EMPTY);
1928 /* mark as invalid as long as the parent table is not protected */
1929 *table = (unsigned long) s_sgt | _REGION_ENTRY_LENGTH |
1930 _REGION_ENTRY_TYPE_R3 | _REGION_ENTRY_INVALID;
1931 if (sg->edat_level >= 1)
1932 *table |= sgt & _REGION_ENTRY_PROTECT;
1933 list_add(&page->lru, &sg->crst_list);
1935 /* nothing to protect for fake tables */
1936 *table &= ~_REGION_ENTRY_INVALID;
1937 spin_unlock(&sg->guest_table_lock);
1940 spin_unlock(&sg->guest_table_lock);
1941 /* Make sgt read-only in parent gmap page table */
1942 raddr = (saddr & _REGION3_MASK) | _SHADOW_RMAP_REGION3;
1943 origin = sgt & _REGION_ENTRY_ORIGIN;
1944 offset = ((sgt & _REGION_ENTRY_OFFSET) >> 6) * PAGE_SIZE;
1945 len = ((sgt & _REGION_ENTRY_LENGTH) + 1) * PAGE_SIZE - offset;
1946 rc = gmap_protect_rmap(sg, raddr, origin + offset, len);
1947 spin_lock(&sg->guest_table_lock);
1949 table = gmap_table_walk(sg, saddr, 2);
1950 if (!table || (*table & _REGION_ENTRY_ORIGIN) !=
1951 (unsigned long) s_sgt)
1952 rc = -EAGAIN; /* Race with unshadow */
1954 *table &= ~_REGION_ENTRY_INVALID;
1956 gmap_unshadow_sgt(sg, raddr);
1958 spin_unlock(&sg->guest_table_lock);
1961 spin_unlock(&sg->guest_table_lock);
1962 __free_pages(page, CRST_ALLOC_ORDER);
1965 EXPORT_SYMBOL_GPL(gmap_shadow_sgt);
1968 * gmap_shadow_lookup_pgtable - find a shadow page table
1969 * @sg: pointer to the shadow guest address space structure
1970 * @saddr: the address in the shadow aguest address space
1971 * @pgt: parent gmap address of the page table to get shadowed
1972 * @dat_protection: if the pgtable is marked as protected by dat
1973 * @fake: pgt references contiguous guest memory block, not a pgtable
1975 * Returns 0 if the shadow page table was found and -EAGAIN if the page
1976 * table was not found.
1978 * Called with sg->mm->mmap_sem in read.
1980 int gmap_shadow_pgt_lookup(struct gmap *sg, unsigned long saddr,
1981 unsigned long *pgt, int *dat_protection,
1984 unsigned long *table;
1988 BUG_ON(!gmap_is_shadow(sg));
1989 spin_lock(&sg->guest_table_lock);
1990 table = gmap_table_walk(sg, saddr, 1); /* get segment pointer */
1991 if (table && !(*table & _SEGMENT_ENTRY_INVALID)) {
1992 /* Shadow page tables are full pages (pte+pgste) */
1993 page = pfn_to_page(*table >> PAGE_SHIFT);
1994 *pgt = page->index & ~GMAP_SHADOW_FAKE_TABLE;
1995 *dat_protection = !!(*table & _SEGMENT_ENTRY_PROTECT);
1996 *fake = !!(page->index & GMAP_SHADOW_FAKE_TABLE);
2001 spin_unlock(&sg->guest_table_lock);
2005 EXPORT_SYMBOL_GPL(gmap_shadow_pgt_lookup);
2008 * gmap_shadow_pgt - instantiate a shadow page table
2009 * @sg: pointer to the shadow guest address space structure
2010 * @saddr: faulting address in the shadow gmap
2011 * @pgt: parent gmap address of the page table to get shadowed
2012 * @fake: pgt references contiguous guest memory block, not a pgtable
2014 * Returns 0 if successfully shadowed or already shadowed, -EAGAIN if the
2015 * shadow table structure is incomplete, -ENOMEM if out of memory,
2016 * -EFAULT if an address in the parent gmap could not be resolved and
2018 * Called with gmap->mm->mmap_sem in read
2020 int gmap_shadow_pgt(struct gmap *sg, unsigned long saddr, unsigned long pgt,
2023 unsigned long raddr, origin;
2024 unsigned long *s_pgt, *table;
2028 BUG_ON(!gmap_is_shadow(sg) || (pgt & _SEGMENT_ENTRY_LARGE));
2029 /* Allocate a shadow page table */
2030 page = page_table_alloc_pgste(sg->mm);
2033 page->index = pgt & _SEGMENT_ENTRY_ORIGIN;
2035 page->index |= GMAP_SHADOW_FAKE_TABLE;
2036 s_pgt = (unsigned long *) page_to_phys(page);
2037 /* Install shadow page table */
2038 spin_lock(&sg->guest_table_lock);
2039 table = gmap_table_walk(sg, saddr, 1); /* get segment pointer */
2041 rc = -EAGAIN; /* Race with unshadow */
2044 if (!(*table & _SEGMENT_ENTRY_INVALID)) {
2045 rc = 0; /* Already established */
2047 } else if (*table & _SEGMENT_ENTRY_ORIGIN) {
2048 rc = -EAGAIN; /* Race with shadow */
2051 /* mark as invalid as long as the parent table is not protected */
2052 *table = (unsigned long) s_pgt | _SEGMENT_ENTRY |
2053 (pgt & _SEGMENT_ENTRY_PROTECT) | _SEGMENT_ENTRY_INVALID;
2054 list_add(&page->lru, &sg->pt_list);
2056 /* nothing to protect for fake tables */
2057 *table &= ~_SEGMENT_ENTRY_INVALID;
2058 spin_unlock(&sg->guest_table_lock);
2061 spin_unlock(&sg->guest_table_lock);
2062 /* Make pgt read-only in parent gmap page table (not the pgste) */
2063 raddr = (saddr & _SEGMENT_MASK) | _SHADOW_RMAP_SEGMENT;
2064 origin = pgt & _SEGMENT_ENTRY_ORIGIN & PAGE_MASK;
2065 rc = gmap_protect_rmap(sg, raddr, origin, PAGE_SIZE);
2066 spin_lock(&sg->guest_table_lock);
2068 table = gmap_table_walk(sg, saddr, 1);
2069 if (!table || (*table & _SEGMENT_ENTRY_ORIGIN) !=
2070 (unsigned long) s_pgt)
2071 rc = -EAGAIN; /* Race with unshadow */
2073 *table &= ~_SEGMENT_ENTRY_INVALID;
2075 gmap_unshadow_pgt(sg, raddr);
2077 spin_unlock(&sg->guest_table_lock);
2080 spin_unlock(&sg->guest_table_lock);
2081 page_table_free_pgste(page);
2085 EXPORT_SYMBOL_GPL(gmap_shadow_pgt);
2088 * gmap_shadow_page - create a shadow page mapping
2089 * @sg: pointer to the shadow guest address space structure
2090 * @saddr: faulting address in the shadow gmap
2091 * @pte: pte in parent gmap address space to get shadowed
2093 * Returns 0 if successfully shadowed or already shadowed, -EAGAIN if the
2094 * shadow table structure is incomplete, -ENOMEM if out of memory and
2095 * -EFAULT if an address in the parent gmap could not be resolved.
2097 * Called with sg->mm->mmap_sem in read.
2099 int gmap_shadow_page(struct gmap *sg, unsigned long saddr, pte_t pte)
2101 struct gmap *parent;
2102 struct gmap_rmap *rmap;
2103 unsigned long vmaddr, paddr;
2105 pte_t *sptep, *tptep;
2109 BUG_ON(!gmap_is_shadow(sg));
2110 parent = sg->parent;
2111 prot = (pte_val(pte) & _PAGE_PROTECT) ? PROT_READ : PROT_WRITE;
2113 rmap = kzalloc(sizeof(*rmap), GFP_KERNEL);
2116 rmap->raddr = (saddr & PAGE_MASK) | _SHADOW_RMAP_PGTABLE;
2119 paddr = pte_val(pte) & PAGE_MASK;
2120 vmaddr = __gmap_translate(parent, paddr);
2121 if (IS_ERR_VALUE(vmaddr)) {
2125 rc = radix_tree_preload(GFP_KERNEL);
2129 sptep = gmap_pte_op_walk(parent, paddr, &ptl);
2131 spin_lock(&sg->guest_table_lock);
2132 /* Get page table pointer */
2133 tptep = (pte_t *) gmap_table_walk(sg, saddr, 0);
2135 spin_unlock(&sg->guest_table_lock);
2136 gmap_pte_op_end(ptl);
2137 radix_tree_preload_end();
2140 rc = ptep_shadow_pte(sg->mm, saddr, sptep, tptep, pte);
2142 /* Success and a new mapping */
2143 gmap_insert_rmap(sg, vmaddr, rmap);
2147 gmap_pte_op_end(ptl);
2148 spin_unlock(&sg->guest_table_lock);
2150 radix_tree_preload_end();
2153 rc = gmap_pte_op_fixup(parent, paddr, vmaddr, prot);
2160 EXPORT_SYMBOL_GPL(gmap_shadow_page);
2163 * gmap_shadow_notify - handle notifications for shadow gmap
2165 * Called with sg->parent->shadow_lock.
2167 static void gmap_shadow_notify(struct gmap *sg, unsigned long vmaddr,
2168 unsigned long gaddr)
2170 struct gmap_rmap *rmap, *rnext, *head;
2171 unsigned long start, end, bits, raddr;
2173 BUG_ON(!gmap_is_shadow(sg));
2175 spin_lock(&sg->guest_table_lock);
2177 spin_unlock(&sg->guest_table_lock);
2180 /* Check for top level table */
2181 start = sg->orig_asce & _ASCE_ORIGIN;
2182 end = start + ((sg->orig_asce & _ASCE_TABLE_LENGTH) + 1) * PAGE_SIZE;
2183 if (!(sg->orig_asce & _ASCE_REAL_SPACE) && gaddr >= start &&
2185 /* The complete shadow table has to go */
2187 spin_unlock(&sg->guest_table_lock);
2188 list_del(&sg->list);
2192 /* Remove the page table tree from on specific entry */
2193 head = radix_tree_delete(&sg->host_to_rmap, vmaddr >> PAGE_SHIFT);
2194 gmap_for_each_rmap_safe(rmap, rnext, head) {
2195 bits = rmap->raddr & _SHADOW_RMAP_MASK;
2196 raddr = rmap->raddr ^ bits;
2198 case _SHADOW_RMAP_REGION1:
2199 gmap_unshadow_r2t(sg, raddr);
2201 case _SHADOW_RMAP_REGION2:
2202 gmap_unshadow_r3t(sg, raddr);
2204 case _SHADOW_RMAP_REGION3:
2205 gmap_unshadow_sgt(sg, raddr);
2207 case _SHADOW_RMAP_SEGMENT:
2208 gmap_unshadow_pgt(sg, raddr);
2210 case _SHADOW_RMAP_PGTABLE:
2211 gmap_unshadow_page(sg, raddr);
2216 spin_unlock(&sg->guest_table_lock);
2220 * ptep_notify - call all invalidation callbacks for a specific pte.
2221 * @mm: pointer to the process mm_struct
2222 * @addr: virtual address in the process address space
2223 * @pte: pointer to the page table entry
2224 * @bits: bits from the pgste that caused the notify call
2226 * This function is assumed to be called with the page table lock held
2227 * for the pte to notify.
2229 void ptep_notify(struct mm_struct *mm, unsigned long vmaddr,
2230 pte_t *pte, unsigned long bits)
2232 unsigned long offset, gaddr = 0;
2233 unsigned long *table;
2234 struct gmap *gmap, *sg, *next;
2236 offset = ((unsigned long) pte) & (255 * sizeof(pte_t));
2237 offset = offset * (PAGE_SIZE / sizeof(pte_t));
2239 list_for_each_entry_rcu(gmap, &mm->context.gmap_list, list) {
2240 spin_lock(&gmap->guest_table_lock);
2241 table = radix_tree_lookup(&gmap->host_to_guest,
2242 vmaddr >> PMD_SHIFT);
2244 gaddr = __gmap_segment_gaddr(table) + offset;
2245 spin_unlock(&gmap->guest_table_lock);
2249 if (!list_empty(&gmap->children) && (bits & PGSTE_VSIE_BIT)) {
2250 spin_lock(&gmap->shadow_lock);
2251 list_for_each_entry_safe(sg, next,
2252 &gmap->children, list)
2253 gmap_shadow_notify(sg, vmaddr, gaddr);
2254 spin_unlock(&gmap->shadow_lock);
2256 if (bits & PGSTE_IN_BIT)
2257 gmap_call_notifier(gmap, gaddr, gaddr + PAGE_SIZE - 1);
2261 EXPORT_SYMBOL_GPL(ptep_notify);
2263 static void pmdp_notify_gmap(struct gmap *gmap, pmd_t *pmdp,
2264 unsigned long gaddr)
2266 pmd_val(*pmdp) &= ~_SEGMENT_ENTRY_GMAP_IN;
2267 gmap_call_notifier(gmap, gaddr, gaddr + HPAGE_SIZE - 1);
2271 * gmap_pmdp_xchg - exchange a gmap pmd with another
2272 * @gmap: pointer to the guest address space structure
2273 * @pmdp: pointer to the pmd entry
2274 * @new: replacement entry
2275 * @gaddr: the affected guest address
2277 * This function is assumed to be called with the guest_table_lock
2280 static void gmap_pmdp_xchg(struct gmap *gmap, pmd_t *pmdp, pmd_t new,
2281 unsigned long gaddr)
2283 gaddr &= HPAGE_MASK;
2284 pmdp_notify_gmap(gmap, pmdp, gaddr);
2285 pmd_val(new) &= ~_SEGMENT_ENTRY_GMAP_IN;
2286 if (MACHINE_HAS_TLB_GUEST)
2287 __pmdp_idte(gaddr, (pmd_t *)pmdp, IDTE_GUEST_ASCE, gmap->asce,
2289 else if (MACHINE_HAS_IDTE)
2290 __pmdp_idte(gaddr, (pmd_t *)pmdp, 0, 0, IDTE_GLOBAL);
2296 static void gmap_pmdp_clear(struct mm_struct *mm, unsigned long vmaddr,
2301 unsigned long gaddr;
2304 list_for_each_entry_rcu(gmap, &mm->context.gmap_list, list) {
2305 spin_lock(&gmap->guest_table_lock);
2306 pmdp = (pmd_t *)radix_tree_delete(&gmap->host_to_guest,
2307 vmaddr >> PMD_SHIFT);
2309 gaddr = __gmap_segment_gaddr((unsigned long *)pmdp);
2310 pmdp_notify_gmap(gmap, pmdp, gaddr);
2311 WARN_ON(pmd_val(*pmdp) & ~(_SEGMENT_ENTRY_HARDWARE_BITS_LARGE |
2312 _SEGMENT_ENTRY_GMAP_UC));
2315 pmd_val(*pmdp) = _SEGMENT_ENTRY_EMPTY;
2317 spin_unlock(&gmap->guest_table_lock);
2323 * gmap_pmdp_invalidate - invalidate all affected guest pmd entries without
2325 * @mm: pointer to the process mm_struct
2326 * @vmaddr: virtual address in the process address space
2328 void gmap_pmdp_invalidate(struct mm_struct *mm, unsigned long vmaddr)
2330 gmap_pmdp_clear(mm, vmaddr, 0);
2332 EXPORT_SYMBOL_GPL(gmap_pmdp_invalidate);
2335 * gmap_pmdp_csp - csp all affected guest pmd entries
2336 * @mm: pointer to the process mm_struct
2337 * @vmaddr: virtual address in the process address space
2339 void gmap_pmdp_csp(struct mm_struct *mm, unsigned long vmaddr)
2341 gmap_pmdp_clear(mm, vmaddr, 1);
2343 EXPORT_SYMBOL_GPL(gmap_pmdp_csp);
2346 * gmap_pmdp_idte_local - invalidate and clear a guest pmd entry
2347 * @mm: pointer to the process mm_struct
2348 * @vmaddr: virtual address in the process address space
2350 void gmap_pmdp_idte_local(struct mm_struct *mm, unsigned long vmaddr)
2352 unsigned long *entry, gaddr;
2357 list_for_each_entry_rcu(gmap, &mm->context.gmap_list, list) {
2358 spin_lock(&gmap->guest_table_lock);
2359 entry = radix_tree_delete(&gmap->host_to_guest,
2360 vmaddr >> PMD_SHIFT);
2362 pmdp = (pmd_t *)entry;
2363 gaddr = __gmap_segment_gaddr(entry);
2364 pmdp_notify_gmap(gmap, pmdp, gaddr);
2365 WARN_ON(*entry & ~(_SEGMENT_ENTRY_HARDWARE_BITS_LARGE |
2366 _SEGMENT_ENTRY_GMAP_UC));
2367 if (MACHINE_HAS_TLB_GUEST)
2368 __pmdp_idte(gaddr, pmdp, IDTE_GUEST_ASCE,
2369 gmap->asce, IDTE_LOCAL);
2370 else if (MACHINE_HAS_IDTE)
2371 __pmdp_idte(gaddr, pmdp, 0, 0, IDTE_LOCAL);
2372 *entry = _SEGMENT_ENTRY_EMPTY;
2374 spin_unlock(&gmap->guest_table_lock);
2378 EXPORT_SYMBOL_GPL(gmap_pmdp_idte_local);
2381 * gmap_pmdp_idte_global - invalidate and clear a guest pmd entry
2382 * @mm: pointer to the process mm_struct
2383 * @vmaddr: virtual address in the process address space
2385 void gmap_pmdp_idte_global(struct mm_struct *mm, unsigned long vmaddr)
2387 unsigned long *entry, gaddr;
2392 list_for_each_entry_rcu(gmap, &mm->context.gmap_list, list) {
2393 spin_lock(&gmap->guest_table_lock);
2394 entry = radix_tree_delete(&gmap->host_to_guest,
2395 vmaddr >> PMD_SHIFT);
2397 pmdp = (pmd_t *)entry;
2398 gaddr = __gmap_segment_gaddr(entry);
2399 pmdp_notify_gmap(gmap, pmdp, gaddr);
2400 WARN_ON(*entry & ~(_SEGMENT_ENTRY_HARDWARE_BITS_LARGE |
2401 _SEGMENT_ENTRY_GMAP_UC));
2402 if (MACHINE_HAS_TLB_GUEST)
2403 __pmdp_idte(gaddr, pmdp, IDTE_GUEST_ASCE,
2404 gmap->asce, IDTE_GLOBAL);
2405 else if (MACHINE_HAS_IDTE)
2406 __pmdp_idte(gaddr, pmdp, 0, 0, IDTE_GLOBAL);
2409 *entry = _SEGMENT_ENTRY_EMPTY;
2411 spin_unlock(&gmap->guest_table_lock);
2415 EXPORT_SYMBOL_GPL(gmap_pmdp_idte_global);
2418 * gmap_test_and_clear_dirty_pmd - test and reset segment dirty status
2419 * @gmap: pointer to guest address space
2420 * @pmdp: pointer to the pmd to be tested
2421 * @gaddr: virtual address in the guest address space
2423 * This function is assumed to be called with the guest_table_lock
2426 bool gmap_test_and_clear_dirty_pmd(struct gmap *gmap, pmd_t *pmdp,
2427 unsigned long gaddr)
2429 if (pmd_val(*pmdp) & _SEGMENT_ENTRY_INVALID)
2432 /* Already protected memory, which did not change is clean */
2433 if (pmd_val(*pmdp) & _SEGMENT_ENTRY_PROTECT &&
2434 !(pmd_val(*pmdp) & _SEGMENT_ENTRY_GMAP_UC))
2437 /* Clear UC indication and reset protection */
2438 pmd_val(*pmdp) &= ~_SEGMENT_ENTRY_GMAP_UC;
2439 gmap_protect_pmd(gmap, gaddr, pmdp, PROT_READ, 0);
2444 * gmap_sync_dirty_log_pmd - set bitmap based on dirty status of segment
2445 * @gmap: pointer to guest address space
2446 * @bitmap: dirty bitmap for this pmd
2447 * @gaddr: virtual address in the guest address space
2448 * @vmaddr: virtual address in the host address space
2450 * This function is assumed to be called with the guest_table_lock
2453 void gmap_sync_dirty_log_pmd(struct gmap *gmap, unsigned long bitmap[4],
2454 unsigned long gaddr, unsigned long vmaddr)
2461 pmdp = gmap_pmd_op_walk(gmap, gaddr);
2465 if (pmd_large(*pmdp)) {
2466 if (gmap_test_and_clear_dirty_pmd(gmap, pmdp, gaddr))
2467 bitmap_fill(bitmap, _PAGE_ENTRIES);
2469 for (i = 0; i < _PAGE_ENTRIES; i++, vmaddr += PAGE_SIZE) {
2470 ptep = pte_alloc_map_lock(gmap->mm, pmdp, vmaddr, &ptl);
2473 if (ptep_test_and_clear_uc(gmap->mm, vmaddr, ptep))
2478 gmap_pmd_op_end(gmap, pmdp);
2480 EXPORT_SYMBOL_GPL(gmap_sync_dirty_log_pmd);
2482 static inline void thp_split_mm(struct mm_struct *mm)
2484 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
2485 struct vm_area_struct *vma;
2488 for (vma = mm->mmap; vma != NULL; vma = vma->vm_next) {
2489 for (addr = vma->vm_start;
2492 follow_page(vma, addr, FOLL_SPLIT);
2493 vma->vm_flags &= ~VM_HUGEPAGE;
2494 vma->vm_flags |= VM_NOHUGEPAGE;
2496 mm->def_flags |= VM_NOHUGEPAGE;
2501 * Remove all empty zero pages from the mapping for lazy refaulting
2502 * - This must be called after mm->context.has_pgste is set, to avoid
2503 * future creation of zero pages
2504 * - This must be called after THP was enabled
2506 static int __zap_zero_pages(pmd_t *pmd, unsigned long start,
2507 unsigned long end, struct mm_walk *walk)
2511 for (addr = start; addr != end; addr += PAGE_SIZE) {
2515 ptep = pte_offset_map_lock(walk->mm, pmd, addr, &ptl);
2516 if (is_zero_pfn(pte_pfn(*ptep)))
2517 ptep_xchg_direct(walk->mm, addr, ptep, __pte(_PAGE_INVALID));
2518 pte_unmap_unlock(ptep, ptl);
2523 static inline void zap_zero_pages(struct mm_struct *mm)
2525 struct mm_walk walk = { .pmd_entry = __zap_zero_pages };
2528 walk_page_range(0, TASK_SIZE, &walk);
2532 * switch on pgstes for its userspace process (for kvm)
2534 int s390_enable_sie(void)
2536 struct mm_struct *mm = current->mm;
2538 /* Do we have pgstes? if yes, we are done */
2539 if (mm_has_pgste(mm))
2541 /* Fail if the page tables are 2K */
2542 if (!mm_alloc_pgste(mm))
2544 down_write(&mm->mmap_sem);
2545 mm->context.has_pgste = 1;
2546 /* split thp mappings and disable thp for future mappings */
2549 up_write(&mm->mmap_sem);
2552 EXPORT_SYMBOL_GPL(s390_enable_sie);
2555 * Enable storage key handling from now on and initialize the storage
2556 * keys with the default key.
2558 static int __s390_enable_skey_pte(pte_t *pte, unsigned long addr,
2559 unsigned long next, struct mm_walk *walk)
2561 /* Clear storage key */
2562 ptep_zap_key(walk->mm, addr, pte);
2566 static int __s390_enable_skey_hugetlb(pte_t *pte, unsigned long addr,
2567 unsigned long hmask, unsigned long next,
2568 struct mm_walk *walk)
2570 pmd_t *pmd = (pmd_t *)pte;
2571 unsigned long start, end;
2572 struct page *page = pmd_page(*pmd);
2575 * The write check makes sure we do not set a key on shared
2576 * memory. This is needed as the walker does not differentiate
2577 * between actual guest memory and the process executable or
2580 if (pmd_val(*pmd) & _SEGMENT_ENTRY_INVALID ||
2581 !(pmd_val(*pmd) & _SEGMENT_ENTRY_WRITE))
2584 start = pmd_val(*pmd) & HPAGE_MASK;
2585 end = start + HPAGE_SIZE - 1;
2586 __storage_key_init_range(start, end);
2587 set_bit(PG_arch_1, &page->flags);
2591 int s390_enable_skey(void)
2593 struct mm_walk walk = {
2594 .hugetlb_entry = __s390_enable_skey_hugetlb,
2595 .pte_entry = __s390_enable_skey_pte,
2597 struct mm_struct *mm = current->mm;
2598 struct vm_area_struct *vma;
2601 down_write(&mm->mmap_sem);
2602 if (mm_uses_skeys(mm))
2605 mm->context.uses_skeys = 1;
2606 for (vma = mm->mmap; vma; vma = vma->vm_next) {
2607 if (ksm_madvise(vma, vma->vm_start, vma->vm_end,
2608 MADV_UNMERGEABLE, &vma->vm_flags)) {
2609 mm->context.uses_skeys = 0;
2614 mm->def_flags &= ~VM_MERGEABLE;
2617 walk_page_range(0, TASK_SIZE, &walk);
2620 up_write(&mm->mmap_sem);
2623 EXPORT_SYMBOL_GPL(s390_enable_skey);
2626 * Reset CMMA state, make all pages stable again.
2628 static int __s390_reset_cmma(pte_t *pte, unsigned long addr,
2629 unsigned long next, struct mm_walk *walk)
2631 ptep_zap_unused(walk->mm, addr, pte, 1);
2635 void s390_reset_cmma(struct mm_struct *mm)
2637 struct mm_walk walk = { .pte_entry = __s390_reset_cmma };
2639 down_write(&mm->mmap_sem);
2641 walk_page_range(0, TASK_SIZE, &walk);
2642 up_write(&mm->mmap_sem);
2644 EXPORT_SYMBOL_GPL(s390_reset_cmma);