1 // SPDX-License-Identifier: GPL-2.0
3 * Copyright IBM Corp. 2007, 2011
4 * Author(s): Martin Schwidefsky <schwidefsky@de.ibm.com>
7 #include <linux/sched.h>
8 #include <linux/kernel.h>
9 #include <linux/errno.h>
10 #include <linux/gfp.h>
12 #include <linux/swap.h>
13 #include <linux/smp.h>
14 #include <linux/spinlock.h>
15 #include <linux/rcupdate.h>
16 #include <linux/slab.h>
17 #include <linux/swapops.h>
18 #include <linux/sysctl.h>
19 #include <linux/ksm.h>
20 #include <linux/mman.h>
22 #include <asm/pgtable.h>
23 #include <asm/pgalloc.h>
25 #include <asm/tlbflush.h>
26 #include <asm/mmu_context.h>
27 #include <asm/page-states.h>
29 static inline void ptep_ipte_local(struct mm_struct *mm, unsigned long addr,
30 pte_t *ptep, int nodat)
32 unsigned long opt, asce;
34 if (MACHINE_HAS_TLB_GUEST) {
36 asce = READ_ONCE(mm->context.gmap_asce);
37 if (asce == 0UL || nodat)
40 asce = asce ? : mm->context.asce;
41 opt |= IPTE_GUEST_ASCE;
43 __ptep_ipte(addr, ptep, opt, asce, IPTE_LOCAL);
45 __ptep_ipte(addr, ptep, 0, 0, IPTE_LOCAL);
49 static inline void ptep_ipte_global(struct mm_struct *mm, unsigned long addr,
50 pte_t *ptep, int nodat)
52 unsigned long opt, asce;
54 if (MACHINE_HAS_TLB_GUEST) {
56 asce = READ_ONCE(mm->context.gmap_asce);
57 if (asce == 0UL || nodat)
60 asce = asce ? : mm->context.asce;
61 opt |= IPTE_GUEST_ASCE;
63 __ptep_ipte(addr, ptep, opt, asce, IPTE_GLOBAL);
65 __ptep_ipte(addr, ptep, 0, 0, IPTE_GLOBAL);
69 static inline pte_t ptep_flush_direct(struct mm_struct *mm,
70 unsigned long addr, pte_t *ptep,
76 if (unlikely(pte_val(old) & _PAGE_INVALID))
78 atomic_inc(&mm->context.flush_count);
79 if (MACHINE_HAS_TLB_LC &&
80 cpumask_equal(mm_cpumask(mm), cpumask_of(smp_processor_id())))
81 ptep_ipte_local(mm, addr, ptep, nodat);
83 ptep_ipte_global(mm, addr, ptep, nodat);
84 atomic_dec(&mm->context.flush_count);
88 static inline pte_t ptep_flush_lazy(struct mm_struct *mm,
89 unsigned long addr, pte_t *ptep,
95 if (unlikely(pte_val(old) & _PAGE_INVALID))
97 atomic_inc(&mm->context.flush_count);
98 if (cpumask_equal(&mm->context.cpu_attach_mask,
99 cpumask_of(smp_processor_id()))) {
100 pte_val(*ptep) |= _PAGE_INVALID;
101 mm->context.flush_mm = 1;
103 ptep_ipte_global(mm, addr, ptep, nodat);
104 atomic_dec(&mm->context.flush_count);
108 static inline pgste_t pgste_get_lock(pte_t *ptep)
110 unsigned long new = 0;
117 " nihh %0,0xff7f\n" /* clear PCL bit in old */
118 " oihh %1,0x0080\n" /* set PCL bit in new */
121 : "=&d" (old), "=&d" (new), "=Q" (ptep[PTRS_PER_PTE])
122 : "Q" (ptep[PTRS_PER_PTE]) : "cc", "memory");
127 static inline void pgste_set_unlock(pte_t *ptep, pgste_t pgste)
131 " nihh %1,0xff7f\n" /* clear PCL bit */
133 : "=Q" (ptep[PTRS_PER_PTE])
134 : "d" (pgste_val(pgste)), "Q" (ptep[PTRS_PER_PTE])
139 static inline pgste_t pgste_get(pte_t *ptep)
141 unsigned long pgste = 0;
143 pgste = *(unsigned long *)(ptep + PTRS_PER_PTE);
145 return __pgste(pgste);
148 static inline void pgste_set(pte_t *ptep, pgste_t pgste)
151 *(pgste_t *)(ptep + PTRS_PER_PTE) = pgste;
155 static inline pgste_t pgste_update_all(pte_t pte, pgste_t pgste,
156 struct mm_struct *mm)
159 unsigned long address, bits, skey;
161 if (!mm_uses_skeys(mm) || pte_val(pte) & _PAGE_INVALID)
163 address = pte_val(pte) & PAGE_MASK;
164 skey = (unsigned long) page_get_storage_key(address);
165 bits = skey & (_PAGE_CHANGED | _PAGE_REFERENCED);
166 /* Transfer page changed & referenced bit to guest bits in pgste */
167 pgste_val(pgste) |= bits << 48; /* GR bit & GC bit */
168 /* Copy page access key and fetch protection bit to pgste */
169 pgste_val(pgste) &= ~(PGSTE_ACC_BITS | PGSTE_FP_BIT);
170 pgste_val(pgste) |= (skey & (_PAGE_ACC_BITS | _PAGE_FP_BIT)) << 56;
176 static inline void pgste_set_key(pte_t *ptep, pgste_t pgste, pte_t entry,
177 struct mm_struct *mm)
180 unsigned long address;
183 if (!mm_uses_skeys(mm) || pte_val(entry) & _PAGE_INVALID)
185 VM_BUG_ON(!(pte_val(*ptep) & _PAGE_INVALID));
186 address = pte_val(entry) & PAGE_MASK;
188 * Set page access key and fetch protection bit from pgste.
189 * The guest C/R information is still in the PGSTE, set real
192 nkey = (pgste_val(pgste) & (PGSTE_ACC_BITS | PGSTE_FP_BIT)) >> 56;
193 nkey |= (pgste_val(pgste) & (PGSTE_GR_BIT | PGSTE_GC_BIT)) >> 48;
194 page_set_storage_key(address, nkey, 0);
198 static inline pgste_t pgste_set_pte(pte_t *ptep, pgste_t pgste, pte_t entry)
201 if ((pte_val(entry) & _PAGE_PRESENT) &&
202 (pte_val(entry) & _PAGE_WRITE) &&
203 !(pte_val(entry) & _PAGE_INVALID)) {
204 if (!MACHINE_HAS_ESOP) {
206 * Without enhanced suppression-on-protection force
207 * the dirty bit on for all writable ptes.
209 pte_val(entry) |= _PAGE_DIRTY;
210 pte_val(entry) &= ~_PAGE_PROTECT;
212 if (!(pte_val(entry) & _PAGE_PROTECT))
213 /* This pte allows write access, set user-dirty */
214 pgste_val(pgste) |= PGSTE_UC_BIT;
221 static inline pgste_t pgste_pte_notify(struct mm_struct *mm,
223 pte_t *ptep, pgste_t pgste)
228 bits = pgste_val(pgste) & (PGSTE_IN_BIT | PGSTE_VSIE_BIT);
230 pgste_val(pgste) ^= bits;
231 ptep_notify(mm, addr, ptep, bits);
237 static inline pgste_t ptep_xchg_start(struct mm_struct *mm,
238 unsigned long addr, pte_t *ptep)
240 pgste_t pgste = __pgste(0);
242 if (mm_has_pgste(mm)) {
243 pgste = pgste_get_lock(ptep);
244 pgste = pgste_pte_notify(mm, addr, ptep, pgste);
249 static inline pte_t ptep_xchg_commit(struct mm_struct *mm,
250 unsigned long addr, pte_t *ptep,
251 pgste_t pgste, pte_t old, pte_t new)
253 if (mm_has_pgste(mm)) {
254 if (pte_val(old) & _PAGE_INVALID)
255 pgste_set_key(ptep, pgste, new, mm);
256 if (pte_val(new) & _PAGE_INVALID) {
257 pgste = pgste_update_all(old, pgste, mm);
258 if ((pgste_val(pgste) & _PGSTE_GPS_USAGE_MASK) ==
259 _PGSTE_GPS_USAGE_UNUSED)
260 pte_val(old) |= _PAGE_UNUSED;
262 pgste = pgste_set_pte(ptep, pgste, new);
263 pgste_set_unlock(ptep, pgste);
270 pte_t ptep_xchg_direct(struct mm_struct *mm, unsigned long addr,
271 pte_t *ptep, pte_t new)
278 pgste = ptep_xchg_start(mm, addr, ptep);
279 nodat = !!(pgste_val(pgste) & _PGSTE_GPS_NODAT);
280 old = ptep_flush_direct(mm, addr, ptep, nodat);
281 old = ptep_xchg_commit(mm, addr, ptep, pgste, old, new);
285 EXPORT_SYMBOL(ptep_xchg_direct);
287 pte_t ptep_xchg_lazy(struct mm_struct *mm, unsigned long addr,
288 pte_t *ptep, pte_t new)
295 pgste = ptep_xchg_start(mm, addr, ptep);
296 nodat = !!(pgste_val(pgste) & _PGSTE_GPS_NODAT);
297 old = ptep_flush_lazy(mm, addr, ptep, nodat);
298 old = ptep_xchg_commit(mm, addr, ptep, pgste, old, new);
302 EXPORT_SYMBOL(ptep_xchg_lazy);
304 pte_t ptep_modify_prot_start(struct mm_struct *mm, unsigned long addr,
312 pgste = ptep_xchg_start(mm, addr, ptep);
313 nodat = !!(pgste_val(pgste) & _PGSTE_GPS_NODAT);
314 old = ptep_flush_lazy(mm, addr, ptep, nodat);
315 if (mm_has_pgste(mm)) {
316 pgste = pgste_update_all(old, pgste, mm);
317 pgste_set(ptep, pgste);
322 void ptep_modify_prot_commit(struct mm_struct *mm, unsigned long addr,
323 pte_t *ptep, pte_t pte)
328 pte_val(pte) &= ~_PAGE_NOEXEC;
329 if (mm_has_pgste(mm)) {
330 pgste = pgste_get(ptep);
331 pgste_set_key(ptep, pgste, pte, mm);
332 pgste = pgste_set_pte(ptep, pgste, pte);
333 pgste_set_unlock(ptep, pgste);
340 static inline void pmdp_idte_local(struct mm_struct *mm,
341 unsigned long addr, pmd_t *pmdp)
343 if (MACHINE_HAS_TLB_GUEST)
344 __pmdp_idte(addr, pmdp, IDTE_NODAT | IDTE_GUEST_ASCE,
345 mm->context.asce, IDTE_LOCAL);
347 __pmdp_idte(addr, pmdp, 0, 0, IDTE_LOCAL);
348 if (mm_has_pgste(mm) && mm->context.allow_gmap_hpage_1m)
349 gmap_pmdp_idte_local(mm, addr);
352 static inline void pmdp_idte_global(struct mm_struct *mm,
353 unsigned long addr, pmd_t *pmdp)
355 if (MACHINE_HAS_TLB_GUEST) {
356 __pmdp_idte(addr, pmdp, IDTE_NODAT | IDTE_GUEST_ASCE,
357 mm->context.asce, IDTE_GLOBAL);
358 if (mm_has_pgste(mm) && mm->context.allow_gmap_hpage_1m)
359 gmap_pmdp_idte_global(mm, addr);
360 } else if (MACHINE_HAS_IDTE) {
361 __pmdp_idte(addr, pmdp, 0, 0, IDTE_GLOBAL);
362 if (mm_has_pgste(mm) && mm->context.allow_gmap_hpage_1m)
363 gmap_pmdp_idte_global(mm, addr);
366 if (mm_has_pgste(mm) && mm->context.allow_gmap_hpage_1m)
367 gmap_pmdp_csp(mm, addr);
371 static inline pmd_t pmdp_flush_direct(struct mm_struct *mm,
372 unsigned long addr, pmd_t *pmdp)
377 if (pmd_val(old) & _SEGMENT_ENTRY_INVALID)
379 atomic_inc(&mm->context.flush_count);
380 if (MACHINE_HAS_TLB_LC &&
381 cpumask_equal(mm_cpumask(mm), cpumask_of(smp_processor_id())))
382 pmdp_idte_local(mm, addr, pmdp);
384 pmdp_idte_global(mm, addr, pmdp);
385 atomic_dec(&mm->context.flush_count);
389 static inline pmd_t pmdp_flush_lazy(struct mm_struct *mm,
390 unsigned long addr, pmd_t *pmdp)
395 if (pmd_val(old) & _SEGMENT_ENTRY_INVALID)
397 atomic_inc(&mm->context.flush_count);
398 if (cpumask_equal(&mm->context.cpu_attach_mask,
399 cpumask_of(smp_processor_id()))) {
400 pmd_val(*pmdp) |= _SEGMENT_ENTRY_INVALID;
401 mm->context.flush_mm = 1;
402 if (mm_has_pgste(mm))
403 gmap_pmdp_invalidate(mm, addr);
405 pmdp_idte_global(mm, addr, pmdp);
407 atomic_dec(&mm->context.flush_count);
411 static pmd_t *pmd_alloc_map(struct mm_struct *mm, unsigned long addr)
418 pgd = pgd_offset(mm, addr);
419 p4d = p4d_alloc(mm, pgd, addr);
422 pud = pud_alloc(mm, p4d, addr);
425 pmd = pmd_alloc(mm, pud, addr);
429 pmd_t pmdp_xchg_direct(struct mm_struct *mm, unsigned long addr,
430 pmd_t *pmdp, pmd_t new)
435 old = pmdp_flush_direct(mm, addr, pmdp);
440 EXPORT_SYMBOL(pmdp_xchg_direct);
442 pmd_t pmdp_xchg_lazy(struct mm_struct *mm, unsigned long addr,
443 pmd_t *pmdp, pmd_t new)
448 old = pmdp_flush_lazy(mm, addr, pmdp);
453 EXPORT_SYMBOL(pmdp_xchg_lazy);
455 static inline void pudp_idte_local(struct mm_struct *mm,
456 unsigned long addr, pud_t *pudp)
458 if (MACHINE_HAS_TLB_GUEST)
459 __pudp_idte(addr, pudp, IDTE_NODAT | IDTE_GUEST_ASCE,
460 mm->context.asce, IDTE_LOCAL);
462 __pudp_idte(addr, pudp, 0, 0, IDTE_LOCAL);
465 static inline void pudp_idte_global(struct mm_struct *mm,
466 unsigned long addr, pud_t *pudp)
468 if (MACHINE_HAS_TLB_GUEST)
469 __pudp_idte(addr, pudp, IDTE_NODAT | IDTE_GUEST_ASCE,
470 mm->context.asce, IDTE_GLOBAL);
471 else if (MACHINE_HAS_IDTE)
472 __pudp_idte(addr, pudp, 0, 0, IDTE_GLOBAL);
475 * Invalid bit position is the same for pmd and pud, so we can
476 * re-use _pmd_csp() here
478 __pmdp_csp((pmd_t *) pudp);
481 static inline pud_t pudp_flush_direct(struct mm_struct *mm,
482 unsigned long addr, pud_t *pudp)
487 if (pud_val(old) & _REGION_ENTRY_INVALID)
489 atomic_inc(&mm->context.flush_count);
490 if (MACHINE_HAS_TLB_LC &&
491 cpumask_equal(mm_cpumask(mm), cpumask_of(smp_processor_id())))
492 pudp_idte_local(mm, addr, pudp);
494 pudp_idte_global(mm, addr, pudp);
495 atomic_dec(&mm->context.flush_count);
499 pud_t pudp_xchg_direct(struct mm_struct *mm, unsigned long addr,
500 pud_t *pudp, pud_t new)
505 old = pudp_flush_direct(mm, addr, pudp);
510 EXPORT_SYMBOL(pudp_xchg_direct);
512 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
513 void pgtable_trans_huge_deposit(struct mm_struct *mm, pmd_t *pmdp,
516 struct list_head *lh = (struct list_head *) pgtable;
518 assert_spin_locked(pmd_lockptr(mm, pmdp));
521 if (!pmd_huge_pte(mm, pmdp))
524 list_add(lh, (struct list_head *) pmd_huge_pte(mm, pmdp));
525 pmd_huge_pte(mm, pmdp) = pgtable;
528 pgtable_t pgtable_trans_huge_withdraw(struct mm_struct *mm, pmd_t *pmdp)
530 struct list_head *lh;
534 assert_spin_locked(pmd_lockptr(mm, pmdp));
537 pgtable = pmd_huge_pte(mm, pmdp);
538 lh = (struct list_head *) pgtable;
540 pmd_huge_pte(mm, pmdp) = NULL;
542 pmd_huge_pte(mm, pmdp) = (pgtable_t) lh->next;
545 ptep = (pte_t *) pgtable;
546 pte_val(*ptep) = _PAGE_INVALID;
548 pte_val(*ptep) = _PAGE_INVALID;
551 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
554 void ptep_set_pte_at(struct mm_struct *mm, unsigned long addr,
555 pte_t *ptep, pte_t entry)
559 /* the mm_has_pgste() check is done in set_pte_at() */
561 pgste = pgste_get_lock(ptep);
562 pgste_val(pgste) &= ~_PGSTE_GPS_ZERO;
563 pgste_set_key(ptep, pgste, entry, mm);
564 pgste = pgste_set_pte(ptep, pgste, entry);
565 pgste_set_unlock(ptep, pgste);
569 void ptep_set_notify(struct mm_struct *mm, unsigned long addr, pte_t *ptep)
574 pgste = pgste_get_lock(ptep);
575 pgste_val(pgste) |= PGSTE_IN_BIT;
576 pgste_set_unlock(ptep, pgste);
581 * ptep_force_prot - change access rights of a locked pte
582 * @mm: pointer to the process mm_struct
583 * @addr: virtual address in the guest address space
584 * @ptep: pointer to the page table entry
585 * @prot: indicates guest access rights: PROT_NONE, PROT_READ or PROT_WRITE
586 * @bit: pgste bit to set (e.g. for notification)
588 * Returns 0 if the access rights were changed and -EAGAIN if the current
589 * and requested access rights are incompatible.
591 int ptep_force_prot(struct mm_struct *mm, unsigned long addr,
592 pte_t *ptep, int prot, unsigned long bit)
596 int pte_i, pte_p, nodat;
598 pgste = pgste_get_lock(ptep);
600 /* Check pte entry after all locks have been acquired */
601 pte_i = pte_val(entry) & _PAGE_INVALID;
602 pte_p = pte_val(entry) & _PAGE_PROTECT;
603 if ((pte_i && (prot != PROT_NONE)) ||
604 (pte_p && (prot & PROT_WRITE))) {
605 pgste_set_unlock(ptep, pgste);
608 /* Change access rights and set pgste bit */
609 nodat = !!(pgste_val(pgste) & _PGSTE_GPS_NODAT);
610 if (prot == PROT_NONE && !pte_i) {
611 ptep_flush_direct(mm, addr, ptep, nodat);
612 pgste = pgste_update_all(entry, pgste, mm);
613 pte_val(entry) |= _PAGE_INVALID;
615 if (prot == PROT_READ && !pte_p) {
616 ptep_flush_direct(mm, addr, ptep, nodat);
617 pte_val(entry) &= ~_PAGE_INVALID;
618 pte_val(entry) |= _PAGE_PROTECT;
620 pgste_val(pgste) |= bit;
621 pgste = pgste_set_pte(ptep, pgste, entry);
622 pgste_set_unlock(ptep, pgste);
626 int ptep_shadow_pte(struct mm_struct *mm, unsigned long saddr,
627 pte_t *sptep, pte_t *tptep, pte_t pte)
629 pgste_t spgste, tpgste;
633 if (!(pte_val(*tptep) & _PAGE_INVALID))
634 return 0; /* already shadowed */
635 spgste = pgste_get_lock(sptep);
637 if (!(pte_val(spte) & _PAGE_INVALID) &&
638 !((pte_val(spte) & _PAGE_PROTECT) &&
639 !(pte_val(pte) & _PAGE_PROTECT))) {
640 pgste_val(spgste) |= PGSTE_VSIE_BIT;
641 tpgste = pgste_get_lock(tptep);
642 pte_val(tpte) = (pte_val(spte) & PAGE_MASK) |
643 (pte_val(pte) & _PAGE_PROTECT);
644 /* don't touch the storage key - it belongs to parent pgste */
645 tpgste = pgste_set_pte(tptep, tpgste, tpte);
646 pgste_set_unlock(tptep, tpgste);
649 pgste_set_unlock(sptep, spgste);
653 void ptep_unshadow_pte(struct mm_struct *mm, unsigned long saddr, pte_t *ptep)
658 pgste = pgste_get_lock(ptep);
659 /* notifier is called by the caller */
660 nodat = !!(pgste_val(pgste) & _PGSTE_GPS_NODAT);
661 ptep_flush_direct(mm, saddr, ptep, nodat);
662 /* don't touch the storage key - it belongs to parent pgste */
663 pgste = pgste_set_pte(ptep, pgste, __pte(_PAGE_INVALID));
664 pgste_set_unlock(ptep, pgste);
667 static void ptep_zap_swap_entry(struct mm_struct *mm, swp_entry_t entry)
669 if (!non_swap_entry(entry))
670 dec_mm_counter(mm, MM_SWAPENTS);
671 else if (is_migration_entry(entry)) {
672 struct page *page = migration_entry_to_page(entry);
674 dec_mm_counter(mm, mm_counter(page));
676 free_swap_and_cache(entry);
679 void ptep_zap_unused(struct mm_struct *mm, unsigned long addr,
680 pte_t *ptep, int reset)
682 unsigned long pgstev;
686 /* Zap unused and logically-zero pages */
688 pgste = pgste_get_lock(ptep);
689 pgstev = pgste_val(pgste);
691 if (!reset && pte_swap(pte) &&
692 ((pgstev & _PGSTE_GPS_USAGE_MASK) == _PGSTE_GPS_USAGE_UNUSED ||
693 (pgstev & _PGSTE_GPS_ZERO))) {
694 ptep_zap_swap_entry(mm, pte_to_swp_entry(pte));
695 pte_clear(mm, addr, ptep);
698 pgste_val(pgste) &= ~_PGSTE_GPS_USAGE_MASK;
699 pgste_set_unlock(ptep, pgste);
703 void ptep_zap_key(struct mm_struct *mm, unsigned long addr, pte_t *ptep)
708 /* Clear storage key ACC and F, but set R/C */
710 pgste = pgste_get_lock(ptep);
711 pgste_val(pgste) &= ~(PGSTE_ACC_BITS | PGSTE_FP_BIT);
712 pgste_val(pgste) |= PGSTE_GR_BIT | PGSTE_GC_BIT;
713 ptev = pte_val(*ptep);
714 if (!(ptev & _PAGE_INVALID) && (ptev & _PAGE_WRITE))
715 page_set_storage_key(ptev & PAGE_MASK, PAGE_DEFAULT_KEY, 1);
716 pgste_set_unlock(ptep, pgste);
721 * Test and reset if a guest page is dirty
723 bool ptep_test_and_clear_uc(struct mm_struct *mm, unsigned long addr,
731 pgste = pgste_get_lock(ptep);
732 dirty = !!(pgste_val(pgste) & PGSTE_UC_BIT);
733 pgste_val(pgste) &= ~PGSTE_UC_BIT;
735 if (dirty && (pte_val(pte) & _PAGE_PRESENT)) {
736 pgste = pgste_pte_notify(mm, addr, ptep, pgste);
737 nodat = !!(pgste_val(pgste) & _PGSTE_GPS_NODAT);
738 ptep_ipte_global(mm, addr, ptep, nodat);
739 if (MACHINE_HAS_ESOP || !(pte_val(pte) & _PAGE_WRITE))
740 pte_val(pte) |= _PAGE_PROTECT;
742 pte_val(pte) |= _PAGE_INVALID;
745 pgste_set_unlock(ptep, pgste);
748 EXPORT_SYMBOL_GPL(ptep_test_and_clear_uc);
750 int set_guest_storage_key(struct mm_struct *mm, unsigned long addr,
751 unsigned char key, bool nq)
753 unsigned long keyul, paddr;
759 pmdp = pmd_alloc_map(mm, addr);
763 ptl = pmd_lock(mm, pmdp);
764 if (!pmd_present(*pmdp)) {
769 if (pmd_large(*pmdp)) {
770 paddr = pmd_val(*pmdp) & HPAGE_MASK;
771 paddr |= addr & ~HPAGE_MASK;
773 * Huge pmds need quiescing operations, they are
776 page_set_storage_key(paddr, key, 1);
782 ptep = pte_alloc_map_lock(mm, pmdp, addr, &ptl);
786 new = old = pgste_get_lock(ptep);
787 pgste_val(new) &= ~(PGSTE_GR_BIT | PGSTE_GC_BIT |
788 PGSTE_ACC_BITS | PGSTE_FP_BIT);
789 keyul = (unsigned long) key;
790 pgste_val(new) |= (keyul & (_PAGE_CHANGED | _PAGE_REFERENCED)) << 48;
791 pgste_val(new) |= (keyul & (_PAGE_ACC_BITS | _PAGE_FP_BIT)) << 56;
792 if (!(pte_val(*ptep) & _PAGE_INVALID)) {
793 unsigned long bits, skey;
795 paddr = pte_val(*ptep) & PAGE_MASK;
796 skey = (unsigned long) page_get_storage_key(paddr);
797 bits = skey & (_PAGE_CHANGED | _PAGE_REFERENCED);
798 skey = key & (_PAGE_ACC_BITS | _PAGE_FP_BIT);
799 /* Set storage key ACC and FP */
800 page_set_storage_key(paddr, skey, !nq);
801 /* Merge host changed & referenced into pgste */
802 pgste_val(new) |= bits << 52;
804 /* changing the guest storage key is considered a change of the page */
805 if ((pgste_val(new) ^ pgste_val(old)) &
806 (PGSTE_ACC_BITS | PGSTE_FP_BIT | PGSTE_GR_BIT | PGSTE_GC_BIT))
807 pgste_val(new) |= PGSTE_UC_BIT;
809 pgste_set_unlock(ptep, new);
810 pte_unmap_unlock(ptep, ptl);
813 EXPORT_SYMBOL(set_guest_storage_key);
816 * Conditionally set a guest storage key (handling csske).
817 * oldkey will be updated when either mr or mc is set and a pointer is given.
819 * Returns 0 if a guests storage key update wasn't necessary, 1 if the guest
820 * storage key was updated and -EFAULT on access errors.
822 int cond_set_guest_storage_key(struct mm_struct *mm, unsigned long addr,
823 unsigned char key, unsigned char *oldkey,
824 bool nq, bool mr, bool mc)
826 unsigned char tmp, mask = _PAGE_ACC_BITS | _PAGE_FP_BIT;
829 /* we can drop the pgste lock between getting and setting the key */
831 rc = get_guest_storage_key(current->mm, addr, &tmp);
837 mask |= _PAGE_REFERENCED;
839 mask |= _PAGE_CHANGED;
840 if (!((tmp ^ key) & mask))
843 rc = set_guest_storage_key(current->mm, addr, key, nq);
844 return rc < 0 ? rc : 1;
846 EXPORT_SYMBOL(cond_set_guest_storage_key);
849 * Reset a guest reference bit (rrbe), returning the reference and changed bit.
851 * Returns < 0 in case of error, otherwise the cc to be reported to the guest.
853 int reset_guest_reference_bit(struct mm_struct *mm, unsigned long addr)
862 pmdp = pmd_alloc_map(mm, addr);
866 ptl = pmd_lock(mm, pmdp);
867 if (!pmd_present(*pmdp)) {
872 if (pmd_large(*pmdp)) {
873 paddr = pmd_val(*pmdp) & HPAGE_MASK;
874 paddr |= addr & ~HPAGE_MASK;
875 cc = page_reset_referenced(paddr);
881 ptep = pte_alloc_map_lock(mm, pmdp, addr, &ptl);
885 new = old = pgste_get_lock(ptep);
886 /* Reset guest reference bit only */
887 pgste_val(new) &= ~PGSTE_GR_BIT;
889 if (!(pte_val(*ptep) & _PAGE_INVALID)) {
890 paddr = pte_val(*ptep) & PAGE_MASK;
891 cc = page_reset_referenced(paddr);
892 /* Merge real referenced bit into host-set */
893 pgste_val(new) |= ((unsigned long) cc << 53) & PGSTE_HR_BIT;
895 /* Reflect guest's logical view, not physical */
896 cc |= (pgste_val(old) & (PGSTE_GR_BIT | PGSTE_GC_BIT)) >> 49;
897 /* Changing the guest storage key is considered a change of the page */
898 if ((pgste_val(new) ^ pgste_val(old)) & PGSTE_GR_BIT)
899 pgste_val(new) |= PGSTE_UC_BIT;
901 pgste_set_unlock(ptep, new);
902 pte_unmap_unlock(ptep, ptl);
905 EXPORT_SYMBOL(reset_guest_reference_bit);
907 int get_guest_storage_key(struct mm_struct *mm, unsigned long addr,
916 pmdp = pmd_alloc_map(mm, addr);
920 ptl = pmd_lock(mm, pmdp);
921 if (!pmd_present(*pmdp)) {
922 /* Not yet mapped memory has a zero key */
928 if (pmd_large(*pmdp)) {
929 paddr = pmd_val(*pmdp) & HPAGE_MASK;
930 paddr |= addr & ~HPAGE_MASK;
931 *key = page_get_storage_key(paddr);
937 ptep = pte_alloc_map_lock(mm, pmdp, addr, &ptl);
941 pgste = pgste_get_lock(ptep);
942 *key = (pgste_val(pgste) & (PGSTE_ACC_BITS | PGSTE_FP_BIT)) >> 56;
943 paddr = pte_val(*ptep) & PAGE_MASK;
944 if (!(pte_val(*ptep) & _PAGE_INVALID))
945 *key = page_get_storage_key(paddr);
946 /* Reflect guest's logical view, not physical */
947 *key |= (pgste_val(pgste) & (PGSTE_GR_BIT | PGSTE_GC_BIT)) >> 48;
948 pgste_set_unlock(ptep, pgste);
949 pte_unmap_unlock(ptep, ptl);
952 EXPORT_SYMBOL(get_guest_storage_key);
955 * pgste_perform_essa - perform ESSA actions on the PGSTE.
956 * @mm: the memory context. It must have PGSTEs, no check is performed here!
957 * @hva: the host virtual address of the page whose PGSTE is to be processed
958 * @orc: the specific action to perform, see the ESSA_SET_* macros.
959 * @oldpte: the PTE will be saved there if the pointer is not NULL.
960 * @oldpgste: the old PGSTE will be saved there if the pointer is not NULL.
962 * Return: 1 if the page is to be added to the CBRL, otherwise 0,
963 * or < 0 in case of error. -EINVAL is returned for invalid values
964 * of orc, -EFAULT for invalid addresses.
966 int pgste_perform_essa(struct mm_struct *mm, unsigned long hva, int orc,
967 unsigned long *oldpte, unsigned long *oldpgste)
969 unsigned long pgstev;
975 WARN_ON_ONCE(orc > ESSA_MAX);
976 if (unlikely(orc > ESSA_MAX))
978 ptep = get_locked_pte(mm, hva, &ptl);
981 pgste = pgste_get_lock(ptep);
982 pgstev = pgste_val(pgste);
984 *oldpte = pte_val(*ptep);
991 case ESSA_SET_STABLE:
992 pgstev &= ~(_PGSTE_GPS_USAGE_MASK | _PGSTE_GPS_NODAT);
993 pgstev |= _PGSTE_GPS_USAGE_STABLE;
995 case ESSA_SET_UNUSED:
996 pgstev &= ~_PGSTE_GPS_USAGE_MASK;
997 pgstev |= _PGSTE_GPS_USAGE_UNUSED;
998 if (pte_val(*ptep) & _PAGE_INVALID)
1001 case ESSA_SET_VOLATILE:
1002 pgstev &= ~_PGSTE_GPS_USAGE_MASK;
1003 pgstev |= _PGSTE_GPS_USAGE_VOLATILE;
1004 if (pte_val(*ptep) & _PAGE_INVALID)
1007 case ESSA_SET_POT_VOLATILE:
1008 pgstev &= ~_PGSTE_GPS_USAGE_MASK;
1009 if (!(pte_val(*ptep) & _PAGE_INVALID)) {
1010 pgstev |= _PGSTE_GPS_USAGE_POT_VOLATILE;
1013 if (pgstev & _PGSTE_GPS_ZERO) {
1014 pgstev |= _PGSTE_GPS_USAGE_VOLATILE;
1017 if (!(pgstev & PGSTE_GC_BIT)) {
1018 pgstev |= _PGSTE_GPS_USAGE_VOLATILE;
1023 case ESSA_SET_STABLE_RESIDENT:
1024 pgstev &= ~_PGSTE_GPS_USAGE_MASK;
1025 pgstev |= _PGSTE_GPS_USAGE_STABLE;
1027 * Since the resident state can go away any time after this
1028 * call, we will not make this page resident. We can revisit
1029 * this decision if a guest will ever start using this.
1032 case ESSA_SET_STABLE_IF_RESIDENT:
1033 if (!(pte_val(*ptep) & _PAGE_INVALID)) {
1034 pgstev &= ~_PGSTE_GPS_USAGE_MASK;
1035 pgstev |= _PGSTE_GPS_USAGE_STABLE;
1038 case ESSA_SET_STABLE_NODAT:
1039 pgstev &= ~_PGSTE_GPS_USAGE_MASK;
1040 pgstev |= _PGSTE_GPS_USAGE_STABLE | _PGSTE_GPS_NODAT;
1043 /* we should never get here! */
1046 /* If we are discarding a page, set it to logical zero */
1048 pgstev |= _PGSTE_GPS_ZERO;
1050 pgste_val(pgste) = pgstev;
1051 pgste_set_unlock(ptep, pgste);
1052 pte_unmap_unlock(ptep, ptl);
1055 EXPORT_SYMBOL(pgste_perform_essa);
1058 * set_pgste_bits - set specific PGSTE bits.
1059 * @mm: the memory context. It must have PGSTEs, no check is performed here!
1060 * @hva: the host virtual address of the page whose PGSTE is to be processed
1061 * @bits: a bitmask representing the bits that will be touched
1062 * @value: the values of the bits to be written. Only the bits in the mask
1065 * Return: 0 on success, < 0 in case of error.
1067 int set_pgste_bits(struct mm_struct *mm, unsigned long hva,
1068 unsigned long bits, unsigned long value)
1074 ptep = get_locked_pte(mm, hva, &ptl);
1075 if (unlikely(!ptep))
1077 new = pgste_get_lock(ptep);
1079 pgste_val(new) &= ~bits;
1080 pgste_val(new) |= value & bits;
1082 pgste_set_unlock(ptep, new);
1083 pte_unmap_unlock(ptep, ptl);
1086 EXPORT_SYMBOL(set_pgste_bits);
1089 * get_pgste - get the current PGSTE for the given address.
1090 * @mm: the memory context. It must have PGSTEs, no check is performed here!
1091 * @hva: the host virtual address of the page whose PGSTE is to be processed
1092 * @pgstep: will be written with the current PGSTE for the given address.
1094 * Return: 0 on success, < 0 in case of error.
1096 int get_pgste(struct mm_struct *mm, unsigned long hva, unsigned long *pgstep)
1101 ptep = get_locked_pte(mm, hva, &ptl);
1102 if (unlikely(!ptep))
1104 *pgstep = pgste_val(pgste_get(ptep));
1105 pte_unmap_unlock(ptep, ptl);
1108 EXPORT_SYMBOL(get_pgste);