arch/powerpc/mm/pgtable.c

   1 /*
   2  * This file contains common routines for dealing with free of page tables
   3  * Along with common page table handling code
   4  *
   5  *  Derived from arch/powerpc/mm/tlb_64.c:
   6  *    Copyright (C) 1995-1996 Gary Thomas (gdt@linuxppc.org)
   7  *
   8  *  Modifications by Paul Mackerras (PowerMac) (paulus@cs.anu.edu.au)
   9  *  and Cort Dougan (PReP) (cort@cs.nmt.edu)
  10  *    Copyright (C) 1996 Paul Mackerras
  11  *
  12  *  Derived from "arch/i386/mm/init.c"
  13  *    Copyright (C) 1991, 1992, 1993, 1994  Linus Torvalds
  14  *
  15  *  Dave Engebretsen <engebret@us.ibm.com>
  16  *      Rework for PPC64 port.
  17  *
  18  *  This program is free software; you can redistribute it and/or
  19  *  modify it under the terms of the GNU General Public License
  20  *  as published by the Free Software Foundation; either version
  21  *  2 of the License, or (at your option) any later version.
  22  */
  23
  24 #include <linux/kernel.h>
  25 #include <linux/gfp.h>
  26 #include <linux/mm.h>
  27 #include <linux/init.h>
  28 #include <linux/percpu.h>
  29 #include <linux/hardirq.h>
  30 #include <linux/hugetlb.h>
  31 #include <asm/pgalloc.h>
  32 #include <asm/tlbflush.h>
  33 #include <asm/tlb.h>
  34
  35 #include "mmu_decl.h"
  36
  37 static inline int is_exec_fault(void)
  38 {
  39         return current->thread.regs && TRAP(current->thread.regs) == 0x400;
  40 }
  41
  42 /* We only try to do i/d cache coherency on stuff that looks like
  43  * reasonably "normal" PTEs. We currently require a PTE to be present
  44  * and we avoid _PAGE_SPECIAL and _PAGE_NO_CACHE. We also only do that
  45  * on userspace PTEs
  46  */
  47 static inline int pte_looks_normal(pte_t pte)
  48 {
  49         return (pte_val(pte) &
  50             (_PAGE_PRESENT | _PAGE_SPECIAL | _PAGE_NO_CACHE | _PAGE_USER)) ==
  51             (_PAGE_PRESENT | _PAGE_USER);
  52 }
  53
  54 struct page * maybe_pte_to_page(pte_t pte)
  55 {
  56         unsigned long pfn = pte_pfn(pte);
  57         struct page *page;
  58
  59         if (unlikely(!pfn_valid(pfn)))
  60                 return NULL;
  61         page = pfn_to_page(pfn);
  62         if (PageReserved(page))
  63                 return NULL;
  64         return page;
  65 }
  66
  67 #if defined(CONFIG_PPC_STD_MMU) || _PAGE_EXEC == 0
  68
  69 /* Server-style MMU handles coherency when hashing if HW exec permission
  70  * is supposed per page (currently 64-bit only). If not, then, we always
  71  * flush the cache for valid PTEs in set_pte. Embedded CPU without HW exec
  72  * support falls into the same category.
  73  */
  74
  75 static pte_t set_pte_filter(pte_t pte, unsigned long addr)
  76 {
  77         pte = __pte(pte_val(pte) & ~_PAGE_HPTEFLAGS);
  78         if (pte_looks_normal(pte) && !(cpu_has_feature(CPU_FTR_COHERENT_ICACHE) ||
  79                                        cpu_has_feature(CPU_FTR_NOEXECUTE))) {
  80                 struct page *pg = maybe_pte_to_page(pte);
  81                 if (!pg)
  82                         return pte;
  83                 if (!test_bit(PG_arch_1, &pg->flags)) {
  84 #ifdef CONFIG_8xx
  85                         /* On 8xx, cache control instructions (particularly
  86                          * "dcbst" from flush_dcache_icache) fault as write
  87                          * operation if there is an unpopulated TLB entry
  88                          * for the address in question. To workaround that,
  89                          * we invalidate the TLB here, thus avoiding dcbst
  90                          * misbehaviour.
  91                          */
  92                         /* 8xx doesn't care about PID, size or ind args */
  93                         _tlbil_va(addr, 0, 0, 0);
  94 #endif /* CONFIG_8xx */
  95                         flush_dcache_icache_page(pg);
  96                         set_bit(PG_arch_1, &pg->flags);
  97                 }
  98         }
  99         return pte;
 100 }
 101
 102 static pte_t set_access_flags_filter(pte_t pte, struct vm_area_struct *vma,
 103                                      int dirty)
 104 {
 105         return pte;
 106 }
 107
 108 #else /* defined(CONFIG_PPC_STD_MMU) || _PAGE_EXEC == 0 */
 109
 110 /* Embedded type MMU with HW exec support. This is a bit more complicated
 111  * as we don't have two bits to spare for _PAGE_EXEC and _PAGE_HWEXEC so
 112  * instead we "filter out" the exec permission for non clean pages.
 113  */
 114 static pte_t set_pte_filter(pte_t pte, unsigned long addr)
 115 {
 116         struct page *pg;
 117
 118         /* No exec permission in the first place, move on */
 119         if (!(pte_val(pte) & _PAGE_EXEC) || !pte_looks_normal(pte))
 120                 return pte;
 121
 122         /* If you set _PAGE_EXEC on weird pages you're on your own */
 123         pg = maybe_pte_to_page(pte);
 124         if (unlikely(!pg))
 125                 return pte;
 126
 127         /* If the page clean, we move on */
 128         if (test_bit(PG_arch_1, &pg->flags))
 129                 return pte;
 130
 131         /* If it's an exec fault, we flush the cache and make it clean */
 132         if (is_exec_fault()) {
 133                 flush_dcache_icache_page(pg);
 134                 set_bit(PG_arch_1, &pg->flags);
 135                 return pte;
 136         }
 137
 138         /* Else, we filter out _PAGE_EXEC */
 139         return __pte(pte_val(pte) & ~_PAGE_EXEC);
 140 }
 141
 142 static pte_t set_access_flags_filter(pte_t pte, struct vm_area_struct *vma,
 143                                      int dirty)
 144 {
 145         struct page *pg;
 146
 147         /* So here, we only care about exec faults, as we use them
 148          * to recover lost _PAGE_EXEC and perform I$/D$ coherency
 149          * if necessary. Also if _PAGE_EXEC is already set, same deal,
 150          * we just bail out
 151          */
 152         if (dirty || (pte_val(pte) & _PAGE_EXEC) || !is_exec_fault())
 153                 return pte;
 154
 155 #ifdef CONFIG_DEBUG_VM
 156         /* So this is an exec fault, _PAGE_EXEC is not set. If it was
 157          * an error we would have bailed out earlier in do_page_fault()
 158          * but let's make sure of it
 159          */
 160         if (WARN_ON(!(vma->vm_flags & VM_EXEC)))
 161                 return pte;
 162 #endif /* CONFIG_DEBUG_VM */
 163
 164         /* If you set _PAGE_EXEC on weird pages you're on your own */
 165         pg = maybe_pte_to_page(pte);
 166         if (unlikely(!pg))
 167                 goto bail;
 168
 169         /* If the page is already clean, we move on */
 170         if (test_bit(PG_arch_1, &pg->flags))
 171                 goto bail;
 172
 173         /* Clean the page and set PG_arch_1 */
 174         flush_dcache_icache_page(pg);
 175         set_bit(PG_arch_1, &pg->flags);
 176
 177  bail:
 178         return __pte(pte_val(pte) | _PAGE_EXEC);
 179 }
 180
 181 #endif /* !(defined(CONFIG_PPC_STD_MMU) || _PAGE_EXEC == 0) */
 182
 183 /*
 184  * set_pte stores a linux PTE into the linux page table.
 185  */
 186 void set_pte_at(struct mm_struct *mm, unsigned long addr, pte_t *ptep,
 187                 pte_t pte)
 188 {
 189 #ifdef CONFIG_DEBUG_VM
 190         WARN_ON(pte_present(*ptep));
 191 #endif
 192         /* Note: mm->context.id might not yet have been assigned as
 193          * this context might not have been activated yet when this
 194          * is called.
 195          */
 196         pte = set_pte_filter(pte, addr);
 197
 198         /* Perform the setting of the PTE */
 199         __set_pte_at(mm, addr, ptep, pte, 0);
 200 }
 201
 202 /*
 203  * This is called when relaxing access to a PTE. It's also called in the page
 204  * fault path when we don't hit any of the major fault cases, ie, a minor
 205  * update of _PAGE_ACCESSED, _PAGE_DIRTY, etc... The generic code will have
 206  * handled those two for us, we additionally deal with missing execute
 207  * permission here on some processors
 208  */
 209 int ptep_set_access_flags(struct vm_area_struct *vma, unsigned long address,
 210                           pte_t *ptep, pte_t entry, int dirty)
 211 {
 212         int changed;
 213         entry = set_access_flags_filter(entry, vma, dirty);
 214         changed = !pte_same(*(ptep), entry);
 215         if (changed) {
 216                 if (!is_vm_hugetlb_page(vma))
 217                         assert_pte_locked(vma->vm_mm, address);
 218                 __ptep_set_access_flags(ptep, entry);
 219                 flush_tlb_page_nohash(vma, address);
 220         }
 221         return changed;
 222 }
 223
 224 #ifdef CONFIG_DEBUG_VM
 225 void assert_pte_locked(struct mm_struct *mm, unsigned long addr)
 226 {
 227         pgd_t *pgd;
 228         pud_t *pud;
 229         pmd_t *pmd;
 230
 231         if (mm == &init_mm)
 232                 return;
 233         pgd = mm->pgd + pgd_index(addr);
 234         BUG_ON(pgd_none(*pgd));
 235         pud = pud_offset(pgd, addr);
 236         BUG_ON(pud_none(*pud));
 237         pmd = pmd_offset(pud, addr);
 238         /*
 239          * khugepaged to collapse normal pages to hugepage, first set
 240          * pmd to none to force page fault/gup to take mmap_sem. After
 241          * pmd is set to none, we do a pte_clear which does this assertion
 242          * so if we find pmd none, return.
 243          */
 244         if (pmd_none(*pmd))
 245                 return;
 246         BUG_ON(!pmd_present(*pmd));
 247         assert_spin_locked(pte_lockptr(mm, pmd));
 248 }
 249 #endif /* CONFIG_DEBUG_VM */
 250