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);
321 EXPORT_SYMBOL(ptep_modify_prot_start);
323 void ptep_modify_prot_commit(struct mm_struct *mm, unsigned long addr,
324 pte_t *ptep, pte_t pte)
329 pte_val(pte) &= ~_PAGE_NOEXEC;
330 if (mm_has_pgste(mm)) {
331 pgste = pgste_get(ptep);
332 pgste_set_key(ptep, pgste, pte, mm);
333 pgste = pgste_set_pte(ptep, pgste, pte);
334 pgste_set_unlock(ptep, pgste);
340 EXPORT_SYMBOL(ptep_modify_prot_commit);
342 static inline void pmdp_idte_local(struct mm_struct *mm,
343 unsigned long addr, pmd_t *pmdp)
345 if (MACHINE_HAS_TLB_GUEST)
346 __pmdp_idte(addr, pmdp, IDTE_NODAT | IDTE_GUEST_ASCE,
347 mm->context.asce, IDTE_LOCAL);
349 __pmdp_idte(addr, pmdp, 0, 0, IDTE_LOCAL);
350 if (mm_has_pgste(mm) && mm->context.allow_gmap_hpage_1m)
351 gmap_pmdp_idte_local(mm, addr);
354 static inline void pmdp_idte_global(struct mm_struct *mm,
355 unsigned long addr, pmd_t *pmdp)
357 if (MACHINE_HAS_TLB_GUEST) {
358 __pmdp_idte(addr, pmdp, IDTE_NODAT | IDTE_GUEST_ASCE,
359 mm->context.asce, IDTE_GLOBAL);
360 if (mm_has_pgste(mm) && mm->context.allow_gmap_hpage_1m)
361 gmap_pmdp_idte_global(mm, addr);
362 } else if (MACHINE_HAS_IDTE) {
363 __pmdp_idte(addr, pmdp, 0, 0, IDTE_GLOBAL);
364 if (mm_has_pgste(mm) && mm->context.allow_gmap_hpage_1m)
365 gmap_pmdp_idte_global(mm, addr);
368 if (mm_has_pgste(mm) && mm->context.allow_gmap_hpage_1m)
369 gmap_pmdp_csp(mm, addr);
373 static inline pmd_t pmdp_flush_direct(struct mm_struct *mm,
374 unsigned long addr, pmd_t *pmdp)
379 if (pmd_val(old) & _SEGMENT_ENTRY_INVALID)
381 atomic_inc(&mm->context.flush_count);
382 if (MACHINE_HAS_TLB_LC &&
383 cpumask_equal(mm_cpumask(mm), cpumask_of(smp_processor_id())))
384 pmdp_idte_local(mm, addr, pmdp);
386 pmdp_idte_global(mm, addr, pmdp);
387 atomic_dec(&mm->context.flush_count);
391 static inline pmd_t pmdp_flush_lazy(struct mm_struct *mm,
392 unsigned long addr, pmd_t *pmdp)
397 if (pmd_val(old) & _SEGMENT_ENTRY_INVALID)
399 atomic_inc(&mm->context.flush_count);
400 if (cpumask_equal(&mm->context.cpu_attach_mask,
401 cpumask_of(smp_processor_id()))) {
402 pmd_val(*pmdp) |= _SEGMENT_ENTRY_INVALID;
403 mm->context.flush_mm = 1;
404 if (mm_has_pgste(mm))
405 gmap_pmdp_invalidate(mm, addr);
407 pmdp_idte_global(mm, addr, pmdp);
409 atomic_dec(&mm->context.flush_count);
414 static pmd_t *pmd_alloc_map(struct mm_struct *mm, unsigned long addr)
421 pgd = pgd_offset(mm, addr);
422 p4d = p4d_alloc(mm, pgd, addr);
425 pud = pud_alloc(mm, p4d, addr);
428 pmd = pmd_alloc(mm, pud, addr);
433 pmd_t pmdp_xchg_direct(struct mm_struct *mm, unsigned long addr,
434 pmd_t *pmdp, pmd_t new)
439 old = pmdp_flush_direct(mm, addr, pmdp);
444 EXPORT_SYMBOL(pmdp_xchg_direct);
446 pmd_t pmdp_xchg_lazy(struct mm_struct *mm, unsigned long addr,
447 pmd_t *pmdp, pmd_t new)
452 old = pmdp_flush_lazy(mm, addr, pmdp);
457 EXPORT_SYMBOL(pmdp_xchg_lazy);
459 static inline void pudp_idte_local(struct mm_struct *mm,
460 unsigned long addr, pud_t *pudp)
462 if (MACHINE_HAS_TLB_GUEST)
463 __pudp_idte(addr, pudp, IDTE_NODAT | IDTE_GUEST_ASCE,
464 mm->context.asce, IDTE_LOCAL);
466 __pudp_idte(addr, pudp, 0, 0, IDTE_LOCAL);
469 static inline void pudp_idte_global(struct mm_struct *mm,
470 unsigned long addr, pud_t *pudp)
472 if (MACHINE_HAS_TLB_GUEST)
473 __pudp_idte(addr, pudp, IDTE_NODAT | IDTE_GUEST_ASCE,
474 mm->context.asce, IDTE_GLOBAL);
475 else if (MACHINE_HAS_IDTE)
476 __pudp_idte(addr, pudp, 0, 0, IDTE_GLOBAL);
479 * Invalid bit position is the same for pmd and pud, so we can
480 * re-use _pmd_csp() here
482 __pmdp_csp((pmd_t *) pudp);
485 static inline pud_t pudp_flush_direct(struct mm_struct *mm,
486 unsigned long addr, pud_t *pudp)
491 if (pud_val(old) & _REGION_ENTRY_INVALID)
493 atomic_inc(&mm->context.flush_count);
494 if (MACHINE_HAS_TLB_LC &&
495 cpumask_equal(mm_cpumask(mm), cpumask_of(smp_processor_id())))
496 pudp_idte_local(mm, addr, pudp);
498 pudp_idte_global(mm, addr, pudp);
499 atomic_dec(&mm->context.flush_count);
503 pud_t pudp_xchg_direct(struct mm_struct *mm, unsigned long addr,
504 pud_t *pudp, pud_t new)
509 old = pudp_flush_direct(mm, addr, pudp);
514 EXPORT_SYMBOL(pudp_xchg_direct);
516 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
517 void pgtable_trans_huge_deposit(struct mm_struct *mm, pmd_t *pmdp,
520 struct list_head *lh = (struct list_head *) pgtable;
522 assert_spin_locked(pmd_lockptr(mm, pmdp));
525 if (!pmd_huge_pte(mm, pmdp))
528 list_add(lh, (struct list_head *) pmd_huge_pte(mm, pmdp));
529 pmd_huge_pte(mm, pmdp) = pgtable;
532 pgtable_t pgtable_trans_huge_withdraw(struct mm_struct *mm, pmd_t *pmdp)
534 struct list_head *lh;
538 assert_spin_locked(pmd_lockptr(mm, pmdp));
541 pgtable = pmd_huge_pte(mm, pmdp);
542 lh = (struct list_head *) pgtable;
544 pmd_huge_pte(mm, pmdp) = NULL;
546 pmd_huge_pte(mm, pmdp) = (pgtable_t) lh->next;
549 ptep = (pte_t *) pgtable;
550 pte_val(*ptep) = _PAGE_INVALID;
552 pte_val(*ptep) = _PAGE_INVALID;
555 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
558 void ptep_set_pte_at(struct mm_struct *mm, unsigned long addr,
559 pte_t *ptep, pte_t entry)
563 /* the mm_has_pgste() check is done in set_pte_at() */
565 pgste = pgste_get_lock(ptep);
566 pgste_val(pgste) &= ~_PGSTE_GPS_ZERO;
567 pgste_set_key(ptep, pgste, entry, mm);
568 pgste = pgste_set_pte(ptep, pgste, entry);
569 pgste_set_unlock(ptep, pgste);
573 void ptep_set_notify(struct mm_struct *mm, unsigned long addr, pte_t *ptep)
578 pgste = pgste_get_lock(ptep);
579 pgste_val(pgste) |= PGSTE_IN_BIT;
580 pgste_set_unlock(ptep, pgste);
585 * ptep_force_prot - change access rights of a locked pte
586 * @mm: pointer to the process mm_struct
587 * @addr: virtual address in the guest address space
588 * @ptep: pointer to the page table entry
589 * @prot: indicates guest access rights: PROT_NONE, PROT_READ or PROT_WRITE
590 * @bit: pgste bit to set (e.g. for notification)
592 * Returns 0 if the access rights were changed and -EAGAIN if the current
593 * and requested access rights are incompatible.
595 int ptep_force_prot(struct mm_struct *mm, unsigned long addr,
596 pte_t *ptep, int prot, unsigned long bit)
600 int pte_i, pte_p, nodat;
602 pgste = pgste_get_lock(ptep);
604 /* Check pte entry after all locks have been acquired */
605 pte_i = pte_val(entry) & _PAGE_INVALID;
606 pte_p = pte_val(entry) & _PAGE_PROTECT;
607 if ((pte_i && (prot != PROT_NONE)) ||
608 (pte_p && (prot & PROT_WRITE))) {
609 pgste_set_unlock(ptep, pgste);
612 /* Change access rights and set pgste bit */
613 nodat = !!(pgste_val(pgste) & _PGSTE_GPS_NODAT);
614 if (prot == PROT_NONE && !pte_i) {
615 ptep_flush_direct(mm, addr, ptep, nodat);
616 pgste = pgste_update_all(entry, pgste, mm);
617 pte_val(entry) |= _PAGE_INVALID;
619 if (prot == PROT_READ && !pte_p) {
620 ptep_flush_direct(mm, addr, ptep, nodat);
621 pte_val(entry) &= ~_PAGE_INVALID;
622 pte_val(entry) |= _PAGE_PROTECT;
624 pgste_val(pgste) |= bit;
625 pgste = pgste_set_pte(ptep, pgste, entry);
626 pgste_set_unlock(ptep, pgste);
630 int ptep_shadow_pte(struct mm_struct *mm, unsigned long saddr,
631 pte_t *sptep, pte_t *tptep, pte_t pte)
633 pgste_t spgste, tpgste;
637 if (!(pte_val(*tptep) & _PAGE_INVALID))
638 return 0; /* already shadowed */
639 spgste = pgste_get_lock(sptep);
641 if (!(pte_val(spte) & _PAGE_INVALID) &&
642 !((pte_val(spte) & _PAGE_PROTECT) &&
643 !(pte_val(pte) & _PAGE_PROTECT))) {
644 pgste_val(spgste) |= PGSTE_VSIE_BIT;
645 tpgste = pgste_get_lock(tptep);
646 pte_val(tpte) = (pte_val(spte) & PAGE_MASK) |
647 (pte_val(pte) & _PAGE_PROTECT);
648 /* don't touch the storage key - it belongs to parent pgste */
649 tpgste = pgste_set_pte(tptep, tpgste, tpte);
650 pgste_set_unlock(tptep, tpgste);
653 pgste_set_unlock(sptep, spgste);
657 void ptep_unshadow_pte(struct mm_struct *mm, unsigned long saddr, pte_t *ptep)
662 pgste = pgste_get_lock(ptep);
663 /* notifier is called by the caller */
664 nodat = !!(pgste_val(pgste) & _PGSTE_GPS_NODAT);
665 ptep_flush_direct(mm, saddr, ptep, nodat);
666 /* don't touch the storage key - it belongs to parent pgste */
667 pgste = pgste_set_pte(ptep, pgste, __pte(_PAGE_INVALID));
668 pgste_set_unlock(ptep, pgste);
671 static void ptep_zap_swap_entry(struct mm_struct *mm, swp_entry_t entry)
673 if (!non_swap_entry(entry))
674 dec_mm_counter(mm, MM_SWAPENTS);
675 else if (is_migration_entry(entry)) {
676 struct page *page = migration_entry_to_page(entry);
678 dec_mm_counter(mm, mm_counter(page));
680 free_swap_and_cache(entry);
683 void ptep_zap_unused(struct mm_struct *mm, unsigned long addr,
684 pte_t *ptep, int reset)
686 unsigned long pgstev;
690 /* Zap unused and logically-zero pages */
692 pgste = pgste_get_lock(ptep);
693 pgstev = pgste_val(pgste);
695 if (!reset && pte_swap(pte) &&
696 ((pgstev & _PGSTE_GPS_USAGE_MASK) == _PGSTE_GPS_USAGE_UNUSED ||
697 (pgstev & _PGSTE_GPS_ZERO))) {
698 ptep_zap_swap_entry(mm, pte_to_swp_entry(pte));
699 pte_clear(mm, addr, ptep);
702 pgste_val(pgste) &= ~(_PGSTE_GPS_USAGE_MASK | _PGSTE_GPS_NODAT);
703 pgste_set_unlock(ptep, pgste);
707 void ptep_zap_key(struct mm_struct *mm, unsigned long addr, pte_t *ptep)
712 /* Clear storage key ACC and F, but set R/C */
714 pgste = pgste_get_lock(ptep);
715 pgste_val(pgste) &= ~(PGSTE_ACC_BITS | PGSTE_FP_BIT);
716 pgste_val(pgste) |= PGSTE_GR_BIT | PGSTE_GC_BIT;
717 ptev = pte_val(*ptep);
718 if (!(ptev & _PAGE_INVALID) && (ptev & _PAGE_WRITE))
719 page_set_storage_key(ptev & PAGE_MASK, PAGE_DEFAULT_KEY, 0);
720 pgste_set_unlock(ptep, pgste);
725 * Test and reset if a guest page is dirty
727 bool ptep_test_and_clear_uc(struct mm_struct *mm, unsigned long addr,
735 pgste = pgste_get_lock(ptep);
736 dirty = !!(pgste_val(pgste) & PGSTE_UC_BIT);
737 pgste_val(pgste) &= ~PGSTE_UC_BIT;
739 if (dirty && (pte_val(pte) & _PAGE_PRESENT)) {
740 pgste = pgste_pte_notify(mm, addr, ptep, pgste);
741 nodat = !!(pgste_val(pgste) & _PGSTE_GPS_NODAT);
742 ptep_ipte_global(mm, addr, ptep, nodat);
743 if (MACHINE_HAS_ESOP || !(pte_val(pte) & _PAGE_WRITE))
744 pte_val(pte) |= _PAGE_PROTECT;
746 pte_val(pte) |= _PAGE_INVALID;
749 pgste_set_unlock(ptep, pgste);
752 EXPORT_SYMBOL_GPL(ptep_test_and_clear_uc);
754 int set_guest_storage_key(struct mm_struct *mm, unsigned long addr,
755 unsigned char key, bool nq)
757 unsigned long keyul, paddr;
763 pmdp = pmd_alloc_map(mm, addr);
767 ptl = pmd_lock(mm, pmdp);
768 if (!pmd_present(*pmdp)) {
773 if (pmd_large(*pmdp)) {
774 paddr = pmd_val(*pmdp) & HPAGE_MASK;
775 paddr |= addr & ~HPAGE_MASK;
777 * Huge pmds need quiescing operations, they are
780 page_set_storage_key(paddr, key, 1);
786 ptep = pte_alloc_map_lock(mm, pmdp, addr, &ptl);
790 new = old = pgste_get_lock(ptep);
791 pgste_val(new) &= ~(PGSTE_GR_BIT | PGSTE_GC_BIT |
792 PGSTE_ACC_BITS | PGSTE_FP_BIT);
793 keyul = (unsigned long) key;
794 pgste_val(new) |= (keyul & (_PAGE_CHANGED | _PAGE_REFERENCED)) << 48;
795 pgste_val(new) |= (keyul & (_PAGE_ACC_BITS | _PAGE_FP_BIT)) << 56;
796 if (!(pte_val(*ptep) & _PAGE_INVALID)) {
797 unsigned long bits, skey;
799 paddr = pte_val(*ptep) & PAGE_MASK;
800 skey = (unsigned long) page_get_storage_key(paddr);
801 bits = skey & (_PAGE_CHANGED | _PAGE_REFERENCED);
802 skey = key & (_PAGE_ACC_BITS | _PAGE_FP_BIT);
803 /* Set storage key ACC and FP */
804 page_set_storage_key(paddr, skey, !nq);
805 /* Merge host changed & referenced into pgste */
806 pgste_val(new) |= bits << 52;
808 /* changing the guest storage key is considered a change of the page */
809 if ((pgste_val(new) ^ pgste_val(old)) &
810 (PGSTE_ACC_BITS | PGSTE_FP_BIT | PGSTE_GR_BIT | PGSTE_GC_BIT))
811 pgste_val(new) |= PGSTE_UC_BIT;
813 pgste_set_unlock(ptep, new);
814 pte_unmap_unlock(ptep, ptl);
817 EXPORT_SYMBOL(set_guest_storage_key);
820 * Conditionally set a guest storage key (handling csske).
821 * oldkey will be updated when either mr or mc is set and a pointer is given.
823 * Returns 0 if a guests storage key update wasn't necessary, 1 if the guest
824 * storage key was updated and -EFAULT on access errors.
826 int cond_set_guest_storage_key(struct mm_struct *mm, unsigned long addr,
827 unsigned char key, unsigned char *oldkey,
828 bool nq, bool mr, bool mc)
830 unsigned char tmp, mask = _PAGE_ACC_BITS | _PAGE_FP_BIT;
833 /* we can drop the pgste lock between getting and setting the key */
835 rc = get_guest_storage_key(current->mm, addr, &tmp);
841 mask |= _PAGE_REFERENCED;
843 mask |= _PAGE_CHANGED;
844 if (!((tmp ^ key) & mask))
847 rc = set_guest_storage_key(current->mm, addr, key, nq);
848 return rc < 0 ? rc : 1;
850 EXPORT_SYMBOL(cond_set_guest_storage_key);
853 * Reset a guest reference bit (rrbe), returning the reference and changed bit.
855 * Returns < 0 in case of error, otherwise the cc to be reported to the guest.
857 int reset_guest_reference_bit(struct mm_struct *mm, unsigned long addr)
866 pmdp = pmd_alloc_map(mm, addr);
870 ptl = pmd_lock(mm, pmdp);
871 if (!pmd_present(*pmdp)) {
876 if (pmd_large(*pmdp)) {
877 paddr = pmd_val(*pmdp) & HPAGE_MASK;
878 paddr |= addr & ~HPAGE_MASK;
879 cc = page_reset_referenced(paddr);
885 ptep = pte_alloc_map_lock(mm, pmdp, addr, &ptl);
889 new = old = pgste_get_lock(ptep);
890 /* Reset guest reference bit only */
891 pgste_val(new) &= ~PGSTE_GR_BIT;
893 if (!(pte_val(*ptep) & _PAGE_INVALID)) {
894 paddr = pte_val(*ptep) & PAGE_MASK;
895 cc = page_reset_referenced(paddr);
896 /* Merge real referenced bit into host-set */
897 pgste_val(new) |= ((unsigned long) cc << 53) & PGSTE_HR_BIT;
899 /* Reflect guest's logical view, not physical */
900 cc |= (pgste_val(old) & (PGSTE_GR_BIT | PGSTE_GC_BIT)) >> 49;
901 /* Changing the guest storage key is considered a change of the page */
902 if ((pgste_val(new) ^ pgste_val(old)) & PGSTE_GR_BIT)
903 pgste_val(new) |= PGSTE_UC_BIT;
905 pgste_set_unlock(ptep, new);
906 pte_unmap_unlock(ptep, ptl);
909 EXPORT_SYMBOL(reset_guest_reference_bit);
911 int get_guest_storage_key(struct mm_struct *mm, unsigned long addr,
920 pmdp = pmd_alloc_map(mm, addr);
924 ptl = pmd_lock(mm, pmdp);
925 if (!pmd_present(*pmdp)) {
926 /* Not yet mapped memory has a zero key */
932 if (pmd_large(*pmdp)) {
933 paddr = pmd_val(*pmdp) & HPAGE_MASK;
934 paddr |= addr & ~HPAGE_MASK;
935 *key = page_get_storage_key(paddr);
941 ptep = pte_alloc_map_lock(mm, pmdp, addr, &ptl);
945 pgste = pgste_get_lock(ptep);
946 *key = (pgste_val(pgste) & (PGSTE_ACC_BITS | PGSTE_FP_BIT)) >> 56;
947 paddr = pte_val(*ptep) & PAGE_MASK;
948 if (!(pte_val(*ptep) & _PAGE_INVALID))
949 *key = page_get_storage_key(paddr);
950 /* Reflect guest's logical view, not physical */
951 *key |= (pgste_val(pgste) & (PGSTE_GR_BIT | PGSTE_GC_BIT)) >> 48;
952 pgste_set_unlock(ptep, pgste);
953 pte_unmap_unlock(ptep, ptl);
956 EXPORT_SYMBOL(get_guest_storage_key);
959 * pgste_perform_essa - perform ESSA actions on the PGSTE.
960 * @mm: the memory context. It must have PGSTEs, no check is performed here!
961 * @hva: the host virtual address of the page whose PGSTE is to be processed
962 * @orc: the specific action to perform, see the ESSA_SET_* macros.
963 * @oldpte: the PTE will be saved there if the pointer is not NULL.
964 * @oldpgste: the old PGSTE will be saved there if the pointer is not NULL.
966 * Return: 1 if the page is to be added to the CBRL, otherwise 0,
967 * or < 0 in case of error. -EINVAL is returned for invalid values
968 * of orc, -EFAULT for invalid addresses.
970 int pgste_perform_essa(struct mm_struct *mm, unsigned long hva, int orc,
971 unsigned long *oldpte, unsigned long *oldpgste)
973 struct vm_area_struct *vma;
974 unsigned long pgstev;
980 WARN_ON_ONCE(orc > ESSA_MAX);
981 if (unlikely(orc > ESSA_MAX))
984 vma = find_vma(mm, hva);
985 if (!vma || hva < vma->vm_start || is_vm_hugetlb_page(vma))
987 ptep = get_locked_pte(mm, hva, &ptl);
990 pgste = pgste_get_lock(ptep);
991 pgstev = pgste_val(pgste);
993 *oldpte = pte_val(*ptep);
1000 case ESSA_SET_STABLE:
1001 pgstev &= ~(_PGSTE_GPS_USAGE_MASK | _PGSTE_GPS_NODAT);
1002 pgstev |= _PGSTE_GPS_USAGE_STABLE;
1004 case ESSA_SET_UNUSED:
1005 pgstev &= ~_PGSTE_GPS_USAGE_MASK;
1006 pgstev |= _PGSTE_GPS_USAGE_UNUSED;
1007 if (pte_val(*ptep) & _PAGE_INVALID)
1010 case ESSA_SET_VOLATILE:
1011 pgstev &= ~_PGSTE_GPS_USAGE_MASK;
1012 pgstev |= _PGSTE_GPS_USAGE_VOLATILE;
1013 if (pte_val(*ptep) & _PAGE_INVALID)
1016 case ESSA_SET_POT_VOLATILE:
1017 pgstev &= ~_PGSTE_GPS_USAGE_MASK;
1018 if (!(pte_val(*ptep) & _PAGE_INVALID)) {
1019 pgstev |= _PGSTE_GPS_USAGE_POT_VOLATILE;
1022 if (pgstev & _PGSTE_GPS_ZERO) {
1023 pgstev |= _PGSTE_GPS_USAGE_VOLATILE;
1026 if (!(pgstev & PGSTE_GC_BIT)) {
1027 pgstev |= _PGSTE_GPS_USAGE_VOLATILE;
1032 case ESSA_SET_STABLE_RESIDENT:
1033 pgstev &= ~_PGSTE_GPS_USAGE_MASK;
1034 pgstev |= _PGSTE_GPS_USAGE_STABLE;
1036 * Since the resident state can go away any time after this
1037 * call, we will not make this page resident. We can revisit
1038 * this decision if a guest will ever start using this.
1041 case ESSA_SET_STABLE_IF_RESIDENT:
1042 if (!(pte_val(*ptep) & _PAGE_INVALID)) {
1043 pgstev &= ~_PGSTE_GPS_USAGE_MASK;
1044 pgstev |= _PGSTE_GPS_USAGE_STABLE;
1047 case ESSA_SET_STABLE_NODAT:
1048 pgstev &= ~_PGSTE_GPS_USAGE_MASK;
1049 pgstev |= _PGSTE_GPS_USAGE_STABLE | _PGSTE_GPS_NODAT;
1052 /* we should never get here! */
1055 /* If we are discarding a page, set it to logical zero */
1057 pgstev |= _PGSTE_GPS_ZERO;
1059 pgste_val(pgste) = pgstev;
1060 pgste_set_unlock(ptep, pgste);
1061 pte_unmap_unlock(ptep, ptl);
1064 EXPORT_SYMBOL(pgste_perform_essa);
1067 * set_pgste_bits - set specific PGSTE bits.
1068 * @mm: the memory context. It must have PGSTEs, no check is performed here!
1069 * @hva: the host virtual address of the page whose PGSTE is to be processed
1070 * @bits: a bitmask representing the bits that will be touched
1071 * @value: the values of the bits to be written. Only the bits in the mask
1074 * Return: 0 on success, < 0 in case of error.
1076 int set_pgste_bits(struct mm_struct *mm, unsigned long hva,
1077 unsigned long bits, unsigned long value)
1079 struct vm_area_struct *vma;
1084 vma = find_vma(mm, hva);
1085 if (!vma || hva < vma->vm_start || is_vm_hugetlb_page(vma))
1087 ptep = get_locked_pte(mm, hva, &ptl);
1088 if (unlikely(!ptep))
1090 new = pgste_get_lock(ptep);
1092 pgste_val(new) &= ~bits;
1093 pgste_val(new) |= value & bits;
1095 pgste_set_unlock(ptep, new);
1096 pte_unmap_unlock(ptep, ptl);
1099 EXPORT_SYMBOL(set_pgste_bits);
1102 * get_pgste - get the current PGSTE for the given address.
1103 * @mm: the memory context. It must have PGSTEs, no check is performed here!
1104 * @hva: the host virtual address of the page whose PGSTE is to be processed
1105 * @pgstep: will be written with the current PGSTE for the given address.
1107 * Return: 0 on success, < 0 in case of error.
1109 int get_pgste(struct mm_struct *mm, unsigned long hva, unsigned long *pgstep)
1111 struct vm_area_struct *vma;
1115 vma = find_vma(mm, hva);
1116 if (!vma || hva < vma->vm_start || is_vm_hugetlb_page(vma))
1118 ptep = get_locked_pte(mm, hva, &ptl);
1119 if (unlikely(!ptep))
1121 *pgstep = pgste_val(pgste_get(ptep));
1122 pte_unmap_unlock(ptep, ptl);
1125 EXPORT_SYMBOL(get_pgste);