2 * This file contains common routines for dealing with free of page tables
3 * Along with common page table handling code
5 * Derived from arch/powerpc/mm/tlb_64.c:
6 * Copyright (C) 1995-1996 Gary Thomas (gdt@linuxppc.org)
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
12 * Derived from "arch/i386/mm/init.c"
13 * Copyright (C) 1991, 1992, 1993, 1994 Linus Torvalds
15 * Dave Engebretsen <engebret@us.ibm.com>
16 * Rework for PPC64 port.
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.
24 #include <linux/kernel.h>
26 #include <linux/init.h>
27 #include <linux/percpu.h>
28 #include <linux/hardirq.h>
29 #include <asm/pgalloc.h>
30 #include <asm/tlbflush.h>
33 static DEFINE_PER_CPU(struct pte_freelist_batch *, pte_freelist_cur);
34 static unsigned long pte_freelist_forced_free;
36 struct pte_freelist_batch
40 pgtable_free_t tables[0];
43 #define PTE_FREELIST_SIZE \
44 ((PAGE_SIZE - sizeof(struct pte_freelist_batch)) \
45 / sizeof(pgtable_free_t))
47 static void pte_free_smp_sync(void *arg)
49 /* Do nothing, just ensure we sync with all CPUs */
52 /* This is only called when we are critically out of memory
53 * (and fail to get a page in pte_free_tlb).
55 static void pgtable_free_now(pgtable_free_t pgf)
57 pte_freelist_forced_free++;
59 smp_call_function(pte_free_smp_sync, NULL, 1);
64 static void pte_free_rcu_callback(struct rcu_head *head)
66 struct pte_freelist_batch *batch =
67 container_of(head, struct pte_freelist_batch, rcu);
70 for (i = 0; i < batch->index; i++)
71 pgtable_free(batch->tables[i]);
73 free_page((unsigned long)batch);
76 static void pte_free_submit(struct pte_freelist_batch *batch)
78 INIT_RCU_HEAD(&batch->rcu);
79 call_rcu(&batch->rcu, pte_free_rcu_callback);
82 void pgtable_free_tlb(struct mmu_gather *tlb, pgtable_free_t pgf)
84 /* This is safe since tlb_gather_mmu has disabled preemption */
85 struct pte_freelist_batch **batchp = &__get_cpu_var(pte_freelist_cur);
87 if (atomic_read(&tlb->mm->mm_users) < 2 ||
88 cpumask_equal(mm_cpumask(tlb->mm), cpumask_of(smp_processor_id()))){
93 if (*batchp == NULL) {
94 *batchp = (struct pte_freelist_batch *)__get_free_page(GFP_ATOMIC);
95 if (*batchp == NULL) {
96 pgtable_free_now(pgf);
101 (*batchp)->tables[(*batchp)->index++] = pgf;
102 if ((*batchp)->index == PTE_FREELIST_SIZE) {
103 pte_free_submit(*batchp);
108 void pte_free_finish(void)
110 /* This is safe since tlb_gather_mmu has disabled preemption */
111 struct pte_freelist_batch **batchp = &__get_cpu_var(pte_freelist_cur);
115 pte_free_submit(*batchp);
120 * Handle i/d cache flushing, called from set_pte_at() or ptep_set_access_flags()
122 static pte_t do_dcache_icache_coherency(pte_t pte)
124 unsigned long pfn = pte_pfn(pte);
127 if (unlikely(!pfn_valid(pfn)))
129 page = pfn_to_page(pfn);
131 if (!PageReserved(page) && !test_bit(PG_arch_1, &page->flags)) {
132 pr_debug("do_dcache_icache_coherency... flushing\n");
133 flush_dcache_icache_page(page);
134 set_bit(PG_arch_1, &page->flags);
137 pr_debug("do_dcache_icache_coherency... already clean\n");
138 return __pte(pte_val(pte) | _PAGE_HWEXEC);
141 static inline int is_exec_fault(void)
143 return current->thread.regs && TRAP(current->thread.regs) == 0x400;
146 /* We only try to do i/d cache coherency on stuff that looks like
147 * reasonably "normal" PTEs. We currently require a PTE to be present
148 * and we avoid _PAGE_SPECIAL and _PAGE_NO_CACHE
150 static inline int pte_looks_normal(pte_t pte)
152 return (pte_val(pte) &
153 (_PAGE_PRESENT | _PAGE_SPECIAL | _PAGE_NO_CACHE)) ==
157 #if defined(CONFIG_PPC_STD_MMU)
158 /* Server-style MMU handles coherency when hashing if HW exec permission
159 * is supposed per page (currently 64-bit only). Else, we always flush
160 * valid PTEs in set_pte.
162 static inline int pte_need_exec_flush(pte_t pte, int set_pte)
164 return set_pte && pte_looks_normal(pte) &&
165 !(cpu_has_feature(CPU_FTR_COHERENT_ICACHE) ||
166 cpu_has_feature(CPU_FTR_NOEXECUTE));
168 #elif _PAGE_HWEXEC == 0
169 /* Embedded type MMU without HW exec support (8xx only so far), we flush
170 * the cache for any present PTE
172 static inline int pte_need_exec_flush(pte_t pte, int set_pte)
174 return set_pte && pte_looks_normal(pte);
177 /* Other embedded CPUs with HW exec support per-page, we flush on exec
178 * fault if HWEXEC is not set
180 static inline int pte_need_exec_flush(pte_t pte, int set_pte)
182 return pte_looks_normal(pte) && is_exec_fault() &&
183 !(pte_val(pte) & _PAGE_HWEXEC);
188 * set_pte stores a linux PTE into the linux page table.
190 void set_pte_at(struct mm_struct *mm, unsigned long addr, pte_t *ptep, pte_t pte)
192 #ifdef CONFIG_DEBUG_VM
193 WARN_ON(pte_present(*ptep));
195 /* Note: mm->context.id might not yet have been assigned as
196 * this context might not have been activated yet when this
199 pte = __pte(pte_val(pte) & ~_PAGE_HPTEFLAGS);
200 if (pte_need_exec_flush(pte, 1))
201 pte = do_dcache_icache_coherency(pte);
203 /* Perform the setting of the PTE */
204 __set_pte_at(mm, addr, ptep, pte, 0);
208 * This is called when relaxing access to a PTE. It's also called in the page
209 * fault path when we don't hit any of the major fault cases, ie, a minor
210 * update of _PAGE_ACCESSED, _PAGE_DIRTY, etc... The generic code will have
211 * handled those two for us, we additionally deal with missing execute
212 * permission here on some processors
214 int ptep_set_access_flags(struct vm_area_struct *vma, unsigned long address,
215 pte_t *ptep, pte_t entry, int dirty)
218 if (!dirty && pte_need_exec_flush(entry, 0))
219 entry = do_dcache_icache_coherency(entry);
220 changed = !pte_same(*(ptep), entry);
222 if (!(vma->vm_flags & VM_HUGETLB))
223 assert_pte_locked(vma->vm_mm, address);
224 __ptep_set_access_flags(ptep, entry);
225 flush_tlb_page_nohash(vma, address);
230 #ifdef CONFIG_DEBUG_VM
231 void assert_pte_locked(struct mm_struct *mm, unsigned long addr)
239 pgd = mm->pgd + pgd_index(addr);
240 BUG_ON(pgd_none(*pgd));
241 pud = pud_offset(pgd, addr);
242 BUG_ON(pud_none(*pud));
243 pmd = pmd_offset(pud, addr);
244 BUG_ON(!pmd_present(*pmd));
245 BUG_ON(!spin_is_locked(pte_lockptr(mm, pmd)));
247 #endif /* CONFIG_DEBUG_VM */