LXR linux/arch/s390/mm/pgtable.c

   1// SPDX-License-Identifier: GPL-2.0
   2/*
   3 *    Copyright IBM Corp. 2007, 2011
   4 *    Author(s): Martin Schwidefsky <schwidefsky@de.ibm.com>
   5 */
   6
   7#include <linux/sched.h>
   8#include <linux/kernel.h>
   9#include <linux/errno.h>
  10#include <linux/gfp.h>
  11#include <linux/mm.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>
  21
  22#include <asm/pgtable.h>
  23#include <asm/pgalloc.h>
  24#include <asm/tlb.h>
  25#include <asm/tlbflush.h>
  26#include <asm/mmu_context.h>
  27#include <asm/page-states.h>
  28
  29static inline void ptep_ipte_local(struct mm_struct *mm, unsigned long addr,
  30                                   pte_t *ptep, int nodat)
  31{
  32        unsigned long opt, asce;
  33
  34        if (MACHINE_HAS_TLB_GUEST) {
  35                opt = 0;
  36                asce = READ_ONCE(mm->context.gmap_asce);
  37                if (asce == 0UL || nodat)
  38                        opt |= IPTE_NODAT;
  39                if (asce != -1UL) {
  40                        asce = asce ? : mm->context.asce;
  41                        opt |= IPTE_GUEST_ASCE;
  42                }
  43                __ptep_ipte(addr, ptep, opt, asce, IPTE_LOCAL);
  44        } else {
  45                __ptep_ipte(addr, ptep, 0, 0, IPTE_LOCAL);
  46        }
  47}
  48
  49static inline void ptep_ipte_global(struct mm_struct *mm, unsigned long addr,
  50                                    pte_t *ptep, int nodat)
  51{
  52        unsigned long opt, asce;
  53
  54        if (MACHINE_HAS_TLB_GUEST) {
  55                opt = 0;
  56                asce = READ_ONCE(mm->context.gmap_asce);
  57                if (asce == 0UL || nodat)
  58                        opt |= IPTE_NODAT;
  59                if (asce != -1UL) {
  60                        asce = asce ? : mm->context.asce;
  61                        opt |= IPTE_GUEST_ASCE;
  62                }
  63                __ptep_ipte(addr, ptep, opt, asce, IPTE_GLOBAL);
  64        } else {
  65                __ptep_ipte(addr, ptep, 0, 0, IPTE_GLOBAL);
  66        }
  67}
  68
  69static inline pte_t ptep_flush_direct(struct mm_struct *mm,
  70                                      unsigned long addr, pte_t *ptep,
  71                                      int nodat)
  72{
  73        pte_t old;
  74
  75        old = *ptep;
  76        if (unlikely(pte_val(old) & _PAGE_INVALID))
  77                return old;
  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);
  82        else
  83                ptep_ipte_global(mm, addr, ptep, nodat);
  84        atomic_dec(&mm->context.flush_count);
  85        return old;
  86}
  87
  88static inline pte_t ptep_flush_lazy(struct mm_struct *mm,
  89                                    unsigned long addr, pte_t *ptep,
  90                                    int nodat)
  91{
  92        pte_t old;
  93
  94        old = *ptep;
  95        if (unlikely(pte_val(old) & _PAGE_INVALID))
  96                return old;
  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;
 102        } else
 103                ptep_ipte_global(mm, addr, ptep, nodat);
 104        atomic_dec(&mm->context.flush_count);
 105        return old;
 106}
 107
 108static inline pgste_t pgste_get_lock(pte_t *ptep)
 109{
 110        unsigned long new = 0;
 111#ifdef CONFIG_PGSTE
 112        unsigned long old;
 113
 114        asm(
 115                "       lg      %0,%2\n"
 116                "0:     lgr     %1,%0\n"
 117                "       nihh    %0,0xff7f\n"    /* clear PCL bit in old */
 118                "       oihh    %1,0x0080\n"    /* set PCL bit in new */
 119                "       csg     %0,%1,%2\n"
 120                "       jl      0b\n"
 121                : "=&d" (old), "=&d" (new), "=Q" (ptep[PTRS_PER_PTE])
 122                : "Q" (ptep[PTRS_PER_PTE]) : "cc", "memory");
 123#endif
 124        return __pgste(new);
 125}
 126
 127static inline void pgste_set_unlock(pte_t *ptep, pgste_t pgste)
 128{
 129#ifdef CONFIG_PGSTE
 130        asm(
 131                "       nihh    %1,0xff7f\n"    /* clear PCL bit */
 132                "       stg     %1,%0\n"
 133                : "=Q" (ptep[PTRS_PER_PTE])
 134                : "d" (pgste_val(pgste)), "Q" (ptep[PTRS_PER_PTE])
 135                : "cc", "memory");
 136#endif
 137}
 138
 139static inline pgste_t pgste_get(pte_t *ptep)
 140{
 141        unsigned long pgste = 0;
 142#ifdef CONFIG_PGSTE
 143        pgste = *(unsigned long *)(ptep + PTRS_PER_PTE);
 144#endif
 145        return __pgste(pgste);
 146}
 147
 148static inline void pgste_set(pte_t *ptep, pgste_t pgste)
 149{
 150#ifdef CONFIG_PGSTE
 151        *(pgste_t *)(ptep + PTRS_PER_PTE) = pgste;
 152#endif
 153}
 154
 155static inline pgste_t pgste_update_all(pte_t pte, pgste_t pgste,
 156                                       struct mm_struct *mm)
 157{
 158#ifdef CONFIG_PGSTE
 159        unsigned long address, bits, skey;
 160
 161        if (!mm_uses_skeys(mm) || pte_val(pte) & _PAGE_INVALID)
 162                return pgste;
 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;
 171#endif
 172        return pgste;
 173
 174}
 175
 176static inline void pgste_set_key(pte_t *ptep, pgste_t pgste, pte_t entry,
 177                                 struct mm_struct *mm)
 178{
 179#ifdef CONFIG_PGSTE
 180        unsigned long address;
 181        unsigned long nkey;
 182
 183        if (!mm_uses_skeys(mm) || pte_val(entry) & _PAGE_INVALID)
 184                return;
 185        VM_BUG_ON(!(pte_val(*ptep) & _PAGE_INVALID));
 186        address = pte_val(entry) & PAGE_MASK;
 187        /*
 188         * Set page access key and fetch protection bit from pgste.
 189         * The guest C/R information is still in the PGSTE, set real
 190         * key C/R to 0.
 191         */
 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);
 195#endif
 196}
 197
 198static inline pgste_t pgste_set_pte(pte_t *ptep, pgste_t pgste, pte_t entry)
 199{
 200#ifdef CONFIG_PGSTE
 201        if ((pte_val(entry) & _PAGE_PRESENT) &&
 202            (pte_val(entry) & _PAGE_WRITE) &&
 203            !(pte_val(entry) & _PAGE_INVALID)) {
 204                if (!MACHINE_HAS_ESOP) {
 205                        /*
 206                         * Without enhanced suppression-on-protection force
 207                         * the dirty bit on for all writable ptes.
 208                         */
 209                        pte_val(entry) |= _PAGE_DIRTY;
 210                        pte_val(entry) &= ~_PAGE_PROTECT;
 211                }
 212                if (!(pte_val(entry) & _PAGE_PROTECT))
 213                        /* This pte allows write access, set user-dirty */
 214                        pgste_val(pgste) |= PGSTE_UC_BIT;
 215        }
 216#endif
 217        *ptep = entry;
 218        return pgste;
 219}
 220
 221static inline pgste_t pgste_pte_notify(struct mm_struct *mm,
 222                                       unsigned long addr,
 223                                       pte_t *ptep, pgste_t pgste)
 224{
 225#ifdef CONFIG_PGSTE
 226        unsigned long bits;
 227
 228        bits = pgste_val(pgste) & (PGSTE_IN_BIT | PGSTE_VSIE_BIT);
 229        if (bits) {
 230                pgste_val(pgste) ^= bits;
 231                ptep_notify(mm, addr, ptep, bits);
 232        }
 233#endif
 234        return pgste;
 235}
 236
 237static inline pgste_t ptep_xchg_start(struct mm_struct *mm,
 238                                      unsigned long addr, pte_t *ptep)
 239{
 240        pgste_t pgste = __pgste(0);
 241
 242        if (mm_has_pgste(mm)) {
 243                pgste = pgste_get_lock(ptep);
 244                pgste = pgste_pte_notify(mm, addr, ptep, pgste);
 245        }
 246        return pgste;
 247}
 248
 249static 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)
 252{
 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;
 261                }
 262                pgste = pgste_set_pte(ptep, pgste, new);
 263                pgste_set_unlock(ptep, pgste);
 264        } else {
 265                *ptep = new;
 266        }
 267        return old;
 268}
 269
 270pte_t ptep_xchg_direct(struct mm_struct *mm, unsigned long addr,
 271                       pte_t *ptep, pte_t new)
 272{
 273        pgste_t pgste;
 274        pte_t old;
 275        int nodat;
 276
 277        preempt_disable();
 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);
 282        preempt_enable();
 283        return old;
 284}
 285EXPORT_SYMBOL(ptep_xchg_direct);
 286
 287pte_t ptep_xchg_lazy(struct mm_struct *mm, unsigned long addr,
 288                     pte_t *ptep, pte_t new)
 289{
 290        pgste_t pgste;
 291        pte_t old;
 292        int nodat;
 293
 294        preempt_disable();
 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);
 299        preempt_enable();
 300        return old;
 301}
 302EXPORT_SYMBOL(ptep_xchg_lazy);
 303
 304pte_t ptep_modify_prot_start(struct vm_area_struct *vma, unsigned long addr,
 305                             pte_t *ptep)
 306{
 307        pgste_t pgste;
 308        pte_t old;
 309        int nodat;
 310        struct mm_struct *mm = vma->vm_mm;
 311
 312        preempt_disable();
 313        pgste = ptep_xchg_start(mm, addr, ptep);
 314        nodat = !!(pgste_val(pgste) & _PGSTE_GPS_NODAT);
 315        old = ptep_flush_lazy(mm, addr, ptep, nodat);
 316        if (mm_has_pgste(mm)) {
 317                pgste = pgste_update_all(old, pgste, mm);
 318                pgste_set(ptep, pgste);
 319        }
 320        return old;
 321}
 322
 323void ptep_modify_prot_commit(struct vm_area_struct *vma, unsigned long addr,
 324                             pte_t *ptep, pte_t old_pte, pte_t pte)
 325{
 326        pgste_t pgste;
 327        struct mm_struct *mm = vma->vm_mm;
 328
 329        if (!MACHINE_HAS_NX)
 330                pte_val(pte) &= ~_PAGE_NOEXEC;
 331        if (mm_has_pgste(mm)) {
 332                pgste = pgste_get(ptep);
 333                pgste_set_key(ptep, pgste, pte, mm);
 334                pgste = pgste_set_pte(ptep, pgste, pte);
 335                pgste_set_unlock(ptep, pgste);
 336        } else {
 337                *ptep = pte;
 338        }
 339        preempt_enable();
 340}
 341
 342static inline void pmdp_idte_local(struct mm_struct *mm,
 343                                   unsigned long addr, pmd_t *pmdp)
 344{
 345        if (MACHINE_HAS_TLB_GUEST)
 346                __pmdp_idte(addr, pmdp, IDTE_NODAT | IDTE_GUEST_ASCE,
 347                            mm->context.asce, IDTE_LOCAL);
 348        else
 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);
 352}
 353
 354static inline void pmdp_idte_global(struct mm_struct *mm,
 355                                    unsigned long addr, pmd_t *pmdp)
 356{
 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);
 366        } else {
 367                __pmdp_csp(pmdp);
 368                if (mm_has_pgste(mm) && mm->context.allow_gmap_hpage_1m)
 369                        gmap_pmdp_csp(mm, addr);
 370        }
 371}
 372
 373static inline pmd_t pmdp_flush_direct(struct mm_struct *mm,
 374                                      unsigned long addr, pmd_t *pmdp)
 375{
 376        pmd_t old;
 377
 378        old = *pmdp;
 379        if (pmd_val(old) & _SEGMENT_ENTRY_INVALID)
 380                return old;
 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);
 385        else
 386                pmdp_idte_global(mm, addr, pmdp);
 387        atomic_dec(&mm->context.flush_count);
 388        return old;
 389}
 390
 391static inline pmd_t pmdp_flush_lazy(struct mm_struct *mm,
 392                                    unsigned long addr, pmd_t *pmdp)
 393{
 394        pmd_t old;
 395
 396        old = *pmdp;
 397        if (pmd_val(old) & _SEGMENT_ENTRY_INVALID)
 398                return old;
 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);
 406        } else {
 407                pmdp_idte_global(mm, addr, pmdp);
 408        }
 409        atomic_dec(&mm->context.flush_count);
 410        return old;
 411}
 412
 413#ifdef CONFIG_PGSTE
 414static pmd_t *pmd_alloc_map(struct mm_struct *mm, unsigned long addr)
 415{
 416        pgd_t *pgd;
 417        p4d_t *p4d;
 418        pud_t *pud;
 419        pmd_t *pmd;
 420
 421        pgd = pgd_offset(mm, addr);
 422        p4d = p4d_alloc(mm, pgd, addr);
 423        if (!p4d)
 424                return NULL;
 425        pud = pud_alloc(mm, p4d, addr);
 426        if (!pud)
 427                return NULL;
 428        pmd = pmd_alloc(mm, pud, addr);
 429        return pmd;
 430}
 431#endif
 432
 433pmd_t pmdp_xchg_direct(struct mm_struct *mm, unsigned long addr,
 434                       pmd_t *pmdp, pmd_t new)
 435{
 436        pmd_t old;
 437
 438        preempt_disable();
 439        old = pmdp_flush_direct(mm, addr, pmdp);
 440        *pmdp = new;
 441        preempt_enable();
 442        return old;
 443}
 444EXPORT_SYMBOL(pmdp_xchg_direct);
 445
 446pmd_t pmdp_xchg_lazy(struct mm_struct *mm, unsigned long addr,
 447                     pmd_t *pmdp, pmd_t new)
 448{
 449        pmd_t old;
 450
 451        preempt_disable();
 452        old = pmdp_flush_lazy(mm, addr, pmdp);
 453        *pmdp = new;
 454        preempt_enable();
 455        return old;
 456}
 457EXPORT_SYMBOL(pmdp_xchg_lazy);
 458
 459static inline void pudp_idte_local(struct mm_struct *mm,
 460                                   unsigned long addr, pud_t *pudp)
 461{
 462        if (MACHINE_HAS_TLB_GUEST)
 463                __pudp_idte(addr, pudp, IDTE_NODAT | IDTE_GUEST_ASCE,
 464                            mm->context.asce, IDTE_LOCAL);
 465        else
 466                __pudp_idte(addr, pudp, 0, 0, IDTE_LOCAL);
 467}
 468
 469static inline void pudp_idte_global(struct mm_struct *mm,
 470                                    unsigned long addr, pud_t *pudp)
 471{
 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);
 477        else
 478                /*
 479                 * Invalid bit position is the same for pmd and pud, so we can
 480                 * re-use _pmd_csp() here
 481                 */
 482                __pmdp_csp((pmd_t *) pudp);
 483}
 484
 485static inline pud_t pudp_flush_direct(struct mm_struct *mm,
 486                                      unsigned long addr, pud_t *pudp)
 487{
 488        pud_t old;
 489
 490        old = *pudp;
 491        if (pud_val(old) & _REGION_ENTRY_INVALID)
 492                return old;
 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);
 497        else
 498                pudp_idte_global(mm, addr, pudp);
 499        atomic_dec(&mm->context.flush_count);
 500        return old;
 501}
 502
 503pud_t pudp_xchg_direct(struct mm_struct *mm, unsigned long addr,
 504                       pud_t *pudp, pud_t new)
 505{
 506        pud_t old;
 507
 508        preempt_disable();
 509        old = pudp_flush_direct(mm, addr, pudp);
 510        *pudp = new;
 511        preempt_enable();
 512        return old;
 513}
 514EXPORT_SYMBOL(pudp_xchg_direct);
 515
 516#ifdef CONFIG_TRANSPARENT_HUGEPAGE
 517void pgtable_trans_huge_deposit(struct mm_struct *mm, pmd_t *pmdp,
 518                                pgtable_t pgtable)
 519{
 520        struct list_head *lh = (struct list_head *) pgtable;
 521
 522        assert_spin_locked(pmd_lockptr(mm, pmdp));
 523
 524        /* FIFO */
 525        if (!pmd_huge_pte(mm, pmdp))
 526                INIT_LIST_HEAD(lh);
 527        else
 528                list_add(lh, (struct list_head *) pmd_huge_pte(mm, pmdp));
 529        pmd_huge_pte(mm, pmdp) = pgtable;
 530}
 531
 532pgtable_t pgtable_trans_huge_withdraw(struct mm_struct *mm, pmd_t *pmdp)
 533{
 534        struct list_head *lh;
 535        pgtable_t pgtable;
 536        pte_t *ptep;
 537
 538        assert_spin_locked(pmd_lockptr(mm, pmdp));
 539
 540        /* FIFO */
 541        pgtable = pmd_huge_pte(mm, pmdp);
 542        lh = (struct list_head *) pgtable;
 543        if (list_empty(lh))
 544                pmd_huge_pte(mm, pmdp) = NULL;
 545        else {
 546                pmd_huge_pte(mm, pmdp) = (pgtable_t) lh->next;
 547                list_del(lh);
 548        }
 549        ptep = (pte_t *) pgtable;
 550        pte_val(*ptep) = _PAGE_INVALID;
 551        ptep++;
 552        pte_val(*ptep) = _PAGE_INVALID;
 553        return pgtable;
 554}
 555#endif /* CONFIG_TRANSPARENT_HUGEPAGE */
 556
 557#ifdef CONFIG_PGSTE
 558void ptep_set_pte_at(struct mm_struct *mm, unsigned long addr,
 559                     pte_t *ptep, pte_t entry)
 560{
 561        pgste_t pgste;
 562
 563        /* the mm_has_pgste() check is done in set_pte_at() */
 564        preempt_disable();
 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);
 570        preempt_enable();
 571}
 572
 573void ptep_set_notify(struct mm_struct *mm, unsigned long addr, pte_t *ptep)
 574{
 575        pgste_t pgste;
 576
 577        preempt_disable();
 578        pgste = pgste_get_lock(ptep);
 579        pgste_val(pgste) |= PGSTE_IN_BIT;
 580        pgste_set_unlock(ptep, pgste);
 581        preempt_enable();
 582}
 583
 584/**
 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)
 591 *
 592 * Returns 0 if the access rights were changed and -EAGAIN if the current
 593 * and requested access rights are incompatible.
 594 */
 595int ptep_force_prot(struct mm_struct *mm, unsigned long addr,
 596                    pte_t *ptep, int prot, unsigned long bit)
 597{
 598        pte_t entry;
 599        pgste_t pgste;
 600        int pte_i, pte_p, nodat;
 601
 602        pgste = pgste_get_lock(ptep);
 603        entry = *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);
 610                return -EAGAIN;
 611        }
 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;
 618        }
 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;
 623        }
 624        pgste_val(pgste) |= bit;
 625        pgste = pgste_set_pte(ptep, pgste, entry);
 626        pgste_set_unlock(ptep, pgste);
 627        return 0;
 628}
 629
 630int ptep_shadow_pte(struct mm_struct *mm, unsigned long saddr,
 631                    pte_t *sptep, pte_t *tptep, pte_t pte)
 632{
 633        pgste_t spgste, tpgste;
 634        pte_t spte, tpte;
 635        int rc = -EAGAIN;
 636
 637        if (!(pte_val(*tptep) & _PAGE_INVALID))
 638                return 0;       /* already shadowed */
 639        spgste = pgste_get_lock(sptep);
 640        spte = *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);
 651                rc = 1;
 652        }
 653        pgste_set_unlock(sptep, spgste);
 654        return rc;
 655}
 656
 657void ptep_unshadow_pte(struct mm_struct *mm, unsigned long saddr, pte_t *ptep)
 658{
 659        pgste_t pgste;
 660        int nodat;
 661
 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);
 669}
 670
 671static void ptep_zap_swap_entry(struct mm_struct *mm, swp_entry_t entry)
 672{
 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);
 677
 678                dec_mm_counter(mm, mm_counter(page));
 679        }
 680        free_swap_and_cache(entry);
 681}
 682
 683void ptep_zap_unused(struct mm_struct *mm, unsigned long addr,
 684                     pte_t *ptep, int reset)
 685{
 686        unsigned long pgstev;
 687        pgste_t pgste;
 688        pte_t pte;
 689
 690        /* Zap unused and logically-zero pages */
 691        preempt_disable();
 692        pgste = pgste_get_lock(ptep);
 693        pgstev = pgste_val(pgste);
 694        pte = *ptep;
 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);
 700        }
 701        if (reset)
 702                pgste_val(pgste) &= ~_PGSTE_GPS_USAGE_MASK;
 703        pgste_set_unlock(ptep, pgste);
 704        preempt_enable();
 705}
 706
 707void ptep_zap_key(struct mm_struct *mm, unsigned long addr, pte_t *ptep)
 708{
 709        unsigned long ptev;
 710        pgste_t pgste;
 711
 712        /* Clear storage key ACC and F, but set R/C */
 713        preempt_disable();
 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, 1);
 720        pgste_set_unlock(ptep, pgste);
 721        preempt_enable();
 722}
 723
 724/*
 725 * Test and reset if a guest page is dirty
 726 */
 727bool ptep_test_and_clear_uc(struct mm_struct *mm, unsigned long addr,
 728                       pte_t *ptep)
 729{
 730        pgste_t pgste;
 731        pte_t pte;
 732        bool dirty;
 733        int nodat;
 734
 735        pgste = pgste_get_lock(ptep);
 736        dirty = !!(pgste_val(pgste) & PGSTE_UC_BIT);
 737        pgste_val(pgste) &= ~PGSTE_UC_BIT;
 738        pte = *ptep;
 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;
 745                else
 746                        pte_val(pte) |= _PAGE_INVALID;
 747                *ptep = pte;
 748        }
 749        pgste_set_unlock(ptep, pgste);
 750        return dirty;
 751}
 752EXPORT_SYMBOL_GPL(ptep_test_and_clear_uc);
 753
 754int set_guest_storage_key(struct mm_struct *mm, unsigned long addr,
 755                          unsigned char key, bool nq)
 756{
 757        unsigned long keyul, paddr;
 758        spinlock_t *ptl;
 759        pgste_t old, new;
 760        pmd_t *pmdp;
 761        pte_t *ptep;
 762
 763        pmdp = pmd_alloc_map(mm, addr);
 764        if (unlikely(!pmdp))
 765                return -EFAULT;
 766
 767        ptl = pmd_lock(mm, pmdp);
 768        if (!pmd_present(*pmdp)) {
 769                spin_unlock(ptl);
 770                return -EFAULT;
 771        }
 772
 773        if (pmd_large(*pmdp)) {
 774                paddr = pmd_val(*pmdp) & HPAGE_MASK;
 775                paddr |= addr & ~HPAGE_MASK;
 776                /*
 777                 * Huge pmds need quiescing operations, they are
 778                 * always mapped.
 779                 */
 780                page_set_storage_key(paddr, key, 1);
 781                spin_unlock(ptl);
 782                return 0;
 783        }
 784        spin_unlock(ptl);
 785
 786        ptep = pte_alloc_map_lock(mm, pmdp, addr, &ptl);
 787        if (unlikely(!ptep))
 788                return -EFAULT;
 789
 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;
 798
 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;
 807        }
 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;
 812
 813        pgste_set_unlock(ptep, new);
 814        pte_unmap_unlock(ptep, ptl);
 815        return 0;
 816}
 817EXPORT_SYMBOL(set_guest_storage_key);
 818
 819/**
 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.
 822 *
 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.
 825 */
 826int 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)
 829{
 830        unsigned char tmp, mask = _PAGE_ACC_BITS | _PAGE_FP_BIT;
 831        int rc;
 832
 833        /* we can drop the pgste lock between getting and setting the key */
 834        if (mr | mc) {
 835                rc = get_guest_storage_key(current->mm, addr, &tmp);
 836                if (rc)
 837                        return rc;
 838                if (oldkey)
 839                        *oldkey = tmp;
 840                if (!mr)
 841                        mask |= _PAGE_REFERENCED;
 842                if (!mc)
 843                        mask |= _PAGE_CHANGED;
 844                if (!((tmp ^ key) & mask))
 845                        return 0;
 846        }
 847        rc = set_guest_storage_key(current->mm, addr, key, nq);
 848        return rc < 0 ? rc : 1;
 849}
 850EXPORT_SYMBOL(cond_set_guest_storage_key);
 851
 852/**
 853 * Reset a guest reference bit (rrbe), returning the reference and changed bit.
 854 *
 855 * Returns < 0 in case of error, otherwise the cc to be reported to the guest.
 856 */
 857int reset_guest_reference_bit(struct mm_struct *mm, unsigned long addr)
 858{
 859        spinlock_t *ptl;
 860        unsigned long paddr;
 861        pgste_t old, new;
 862        pmd_t *pmdp;
 863        pte_t *ptep;
 864        int cc = 0;
 865
 866        pmdp = pmd_alloc_map(mm, addr);
 867        if (unlikely(!pmdp))
 868                return -EFAULT;
 869
 870        ptl = pmd_lock(mm, pmdp);
 871        if (!pmd_present(*pmdp)) {
 872                spin_unlock(ptl);
 873                return -EFAULT;
 874        }
 875
 876        if (pmd_large(*pmdp)) {
 877                paddr = pmd_val(*pmdp) & HPAGE_MASK;
 878                paddr |= addr & ~HPAGE_MASK;
 879                cc = page_reset_referenced(paddr);
 880                spin_unlock(ptl);
 881                return cc;
 882        }
 883        spin_unlock(ptl);
 884
 885        ptep = pte_alloc_map_lock(mm, pmdp, addr, &ptl);
 886        if (unlikely(!ptep))
 887                return -EFAULT;
 888
 889        new = old = pgste_get_lock(ptep);
 890        /* Reset guest reference bit only */
 891        pgste_val(new) &= ~PGSTE_GR_BIT;
 892
 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;
 898        }
 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;
 904
 905        pgste_set_unlock(ptep, new);
 906        pte_unmap_unlock(ptep, ptl);
 907        return cc;
 908}
 909EXPORT_SYMBOL(reset_guest_reference_bit);
 910
 911int get_guest_storage_key(struct mm_struct *mm, unsigned long addr,
 912                          unsigned char *key)
 913{
 914        unsigned long paddr;
 915        spinlock_t *ptl;
 916        pgste_t pgste;
 917        pmd_t *pmdp;
 918        pte_t *ptep;
 919
 920        pmdp = pmd_alloc_map(mm, addr);
 921        if (unlikely(!pmdp))
 922                return -EFAULT;
 923
 924        ptl = pmd_lock(mm, pmdp);
 925        if (!pmd_present(*pmdp)) {
 926                /* Not yet mapped memory has a zero key */
 927                spin_unlock(ptl);
 928                *key = 0;
 929                return 0;
 930        }
 931
 932        if (pmd_large(*pmdp)) {
 933                paddr = pmd_val(*pmdp) & HPAGE_MASK;
 934                paddr |= addr & ~HPAGE_MASK;
 935                *key = page_get_storage_key(paddr);
 936                spin_unlock(ptl);
 937                return 0;
 938        }
 939        spin_unlock(ptl);
 940
 941        ptep = pte_alloc_map_lock(mm, pmdp, addr, &ptl);
 942        if (unlikely(!ptep))
 943                return -EFAULT;
 944
 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);
 954        return 0;
 955}
 956EXPORT_SYMBOL(get_guest_storage_key);
 957
 958/**
 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.
 965 *
 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.
 969 */
 970int pgste_perform_essa(struct mm_struct *mm, unsigned long hva, int orc,
 971                        unsigned long *oldpte, unsigned long *oldpgste)
 972{
 973        unsigned long pgstev;
 974        spinlock_t *ptl;
 975        pgste_t pgste;
 976        pte_t *ptep;
 977        int res = 0;
 978
 979        WARN_ON_ONCE(orc > ESSA_MAX);
 980        if (unlikely(orc > ESSA_MAX))
 981                return -EINVAL;
 982        ptep = get_locked_pte(mm, hva, &ptl);
 983        if (unlikely(!ptep))
 984                return -EFAULT;
 985        pgste = pgste_get_lock(ptep);
 986        pgstev = pgste_val(pgste);
 987        if (oldpte)
 988                *oldpte = pte_val(*ptep);
 989        if (oldpgste)
 990                *oldpgste = pgstev;
 991
 992        switch (orc) {
 993        case ESSA_GET_STATE:
 994                break;
 995        case ESSA_SET_STABLE:
 996                pgstev &= ~(_PGSTE_GPS_USAGE_MASK | _PGSTE_GPS_NODAT);
 997                pgstev |= _PGSTE_GPS_USAGE_STABLE;
 998                break;
 999        case ESSA_SET_UNUSED:
1000                pgstev &= ~_PGSTE_GPS_USAGE_MASK;

1001                pgstev |= _PGSTE_GPS_USAGE_UNUSED;
1002                if (pte_val(*ptep) & _PAGE_INVALID)
1003                        res = 1;
1004                break;
1005        case ESSA_SET_VOLATILE:
1006                pgstev &= ~_PGSTE_GPS_USAGE_MASK;
1007                pgstev |= _PGSTE_GPS_USAGE_VOLATILE;
1008                if (pte_val(*ptep) & _PAGE_INVALID)
1009                        res = 1;
1010                break;
1011        case ESSA_SET_POT_VOLATILE:
1012                pgstev &= ~_PGSTE_GPS_USAGE_MASK;
1013                if (!(pte_val(*ptep) & _PAGE_INVALID)) {
1014                        pgstev |= _PGSTE_GPS_USAGE_POT_VOLATILE;
1015                        break;
1016                }
1017                if (pgstev & _PGSTE_GPS_ZERO) {
1018                        pgstev |= _PGSTE_GPS_USAGE_VOLATILE;
1019                        break;
1020                }
1021                if (!(pgstev & PGSTE_GC_BIT)) {
1022                        pgstev |= _PGSTE_GPS_USAGE_VOLATILE;
1023                        res = 1;
1024                        break;
1025                }
1026                break;
1027        case ESSA_SET_STABLE_RESIDENT:
1028                pgstev &= ~_PGSTE_GPS_USAGE_MASK;
1029                pgstev |= _PGSTE_GPS_USAGE_STABLE;
1030                /*
1031                 * Since the resident state can go away any time after this
1032                 * call, we will not make this page resident. We can revisit
1033                 * this decision if a guest will ever start using this.
1034                 */
1035                break;
1036        case ESSA_SET_STABLE_IF_RESIDENT:
1037                if (!(pte_val(*ptep) & _PAGE_INVALID)) {
1038                        pgstev &= ~_PGSTE_GPS_USAGE_MASK;
1039                        pgstev |= _PGSTE_GPS_USAGE_STABLE;
1040                }
1041                break;
1042        case ESSA_SET_STABLE_NODAT:
1043                pgstev &= ~_PGSTE_GPS_USAGE_MASK;
1044                pgstev |= _PGSTE_GPS_USAGE_STABLE | _PGSTE_GPS_NODAT;
1045                break;
1046        default:
1047                /* we should never get here! */
1048                break;
1049        }
1050        /* If we are discarding a page, set it to logical zero */
1051        if (res)
1052                pgstev |= _PGSTE_GPS_ZERO;
1053
1054        pgste_val(pgste) = pgstev;
1055        pgste_set_unlock(ptep, pgste);
1056        pte_unmap_unlock(ptep, ptl);
1057        return res;
1058}
1059EXPORT_SYMBOL(pgste_perform_essa);
1060
1061/**
1062 * set_pgste_bits - set specific PGSTE bits.
1063 * @mm: the memory context. It must have PGSTEs, no check is performed here!
1064 * @hva: the host virtual address of the page whose PGSTE is to be processed
1065 * @bits: a bitmask representing the bits that will be touched
1066 * @value: the values of the bits to be written. Only the bits in the mask
1067 *         will be written.
1068 *
1069 * Return: 0 on success, < 0 in case of error.
1070 */
1071int set_pgste_bits(struct mm_struct *mm, unsigned long hva,
1072                        unsigned long bits, unsigned long value)
1073{
1074        spinlock_t *ptl;
1075        pgste_t new;
1076        pte_t *ptep;
1077
1078        ptep = get_locked_pte(mm, hva, &ptl);
1079        if (unlikely(!ptep))
1080                return -EFAULT;
1081        new = pgste_get_lock(ptep);
1082
1083        pgste_val(new) &= ~bits;
1084        pgste_val(new) |= value & bits;
1085
1086        pgste_set_unlock(ptep, new);
1087        pte_unmap_unlock(ptep, ptl);
1088        return 0;
1089}
1090EXPORT_SYMBOL(set_pgste_bits);
1091
1092/**
1093 * get_pgste - get the current PGSTE for the given address.
1094 * @mm: the memory context. It must have PGSTEs, no check is performed here!
1095 * @hva: the host virtual address of the page whose PGSTE is to be processed
1096 * @pgstep: will be written with the current PGSTE for the given address.
1097 *
1098 * Return: 0 on success, < 0 in case of error.
1099 */
1100int get_pgste(struct mm_struct *mm, unsigned long hva, unsigned long *pgstep)
1101{
1102        spinlock_t *ptl;
1103        pte_t *ptep;
1104
1105        ptep = get_locked_pte(mm, hva, &ptl);
1106        if (unlikely(!ptep))
1107                return -EFAULT;
1108        *pgstep = pgste_val(pgste_get(ptep));
1109        pte_unmap_unlock(ptep, ptl);
1110        return 0;
1111}
1112EXPORT_SYMBOL(get_pgste);
1113#endif
1114