blob: 83f1551ec2c9182d9299b0504fcdc2138ea3c917 [file] [log] [blame]
Kumar Gala0186f472008-11-19 12:50:04 +00001/*
2 * This file contains common routines for dealing with free of page tables
Benjamin Herrenschmidt8d30c142009-02-10 16:02:37 +00003 * Along with common page table handling code
Kumar Gala0186f472008-11-19 12:50:04 +00004 *
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/mm.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>
31#include <asm/tlb.h>
32
Benjamin Herrenschmidta8f77582009-07-23 23:15:45 +000033DEFINE_PER_CPU(struct mmu_gather, mmu_gathers);
34
Benjamin Herrenschmidtc7cc58a2009-07-23 23:15:28 +000035#ifdef CONFIG_SMP
36
37/*
38 * Handle batching of page table freeing on SMP. Page tables are
39 * queued up and send to be freed later by RCU in order to avoid
40 * freeing a page table page that is being walked without locks
41 */
42
Kumar Gala0186f472008-11-19 12:50:04 +000043static DEFINE_PER_CPU(struct pte_freelist_batch *, pte_freelist_cur);
44static unsigned long pte_freelist_forced_free;
45
46struct pte_freelist_batch
47{
48 struct rcu_head rcu;
49 unsigned int index;
50 pgtable_free_t tables[0];
51};
52
53#define PTE_FREELIST_SIZE \
54 ((PAGE_SIZE - sizeof(struct pte_freelist_batch)) \
55 / sizeof(pgtable_free_t))
56
57static void pte_free_smp_sync(void *arg)
58{
59 /* Do nothing, just ensure we sync with all CPUs */
60}
61
62/* This is only called when we are critically out of memory
63 * (and fail to get a page in pte_free_tlb).
64 */
65static void pgtable_free_now(pgtable_free_t pgf)
66{
67 pte_freelist_forced_free++;
68
69 smp_call_function(pte_free_smp_sync, NULL, 1);
70
71 pgtable_free(pgf);
72}
73
74static void pte_free_rcu_callback(struct rcu_head *head)
75{
76 struct pte_freelist_batch *batch =
77 container_of(head, struct pte_freelist_batch, rcu);
78 unsigned int i;
79
80 for (i = 0; i < batch->index; i++)
81 pgtable_free(batch->tables[i]);
82
83 free_page((unsigned long)batch);
84}
85
86static void pte_free_submit(struct pte_freelist_batch *batch)
87{
88 INIT_RCU_HEAD(&batch->rcu);
89 call_rcu(&batch->rcu, pte_free_rcu_callback);
90}
91
92void pgtable_free_tlb(struct mmu_gather *tlb, pgtable_free_t pgf)
93{
94 /* This is safe since tlb_gather_mmu has disabled preemption */
Kumar Gala0186f472008-11-19 12:50:04 +000095 struct pte_freelist_batch **batchp = &__get_cpu_var(pte_freelist_cur);
96
97 if (atomic_read(&tlb->mm->mm_users) < 2 ||
Rusty Russell56aa4122009-03-15 18:16:43 +000098 cpumask_equal(mm_cpumask(tlb->mm), cpumask_of(smp_processor_id()))){
Kumar Gala0186f472008-11-19 12:50:04 +000099 pgtable_free(pgf);
100 return;
101 }
102
103 if (*batchp == NULL) {
104 *batchp = (struct pte_freelist_batch *)__get_free_page(GFP_ATOMIC);
105 if (*batchp == NULL) {
106 pgtable_free_now(pgf);
107 return;
108 }
109 (*batchp)->index = 0;
110 }
111 (*batchp)->tables[(*batchp)->index++] = pgf;
112 if ((*batchp)->index == PTE_FREELIST_SIZE) {
113 pte_free_submit(*batchp);
114 *batchp = NULL;
115 }
116}
117
118void pte_free_finish(void)
119{
120 /* This is safe since tlb_gather_mmu has disabled preemption */
121 struct pte_freelist_batch **batchp = &__get_cpu_var(pte_freelist_cur);
122
123 if (*batchp == NULL)
124 return;
125 pte_free_submit(*batchp);
126 *batchp = NULL;
127}
Benjamin Herrenschmidt8d30c142009-02-10 16:02:37 +0000128
Benjamin Herrenschmidtc7cc58a2009-07-23 23:15:28 +0000129#endif /* CONFIG_SMP */
130
Benjamin Herrenschmidt8d30c142009-02-10 16:02:37 +0000131static inline int is_exec_fault(void)
132{
133 return current->thread.regs && TRAP(current->thread.regs) == 0x400;
134}
135
136/* We only try to do i/d cache coherency on stuff that looks like
137 * reasonably "normal" PTEs. We currently require a PTE to be present
Benjamin Herrenschmidtea3cc332009-08-18 19:00:34 +0000138 * and we avoid _PAGE_SPECIAL and _PAGE_NO_CACHE. We also only do that
139 * on userspace PTEs
Benjamin Herrenschmidt8d30c142009-02-10 16:02:37 +0000140 */
141static inline int pte_looks_normal(pte_t pte)
142{
143 return (pte_val(pte) &
Benjamin Herrenschmidtea3cc332009-08-18 19:00:34 +0000144 (_PAGE_PRESENT | _PAGE_SPECIAL | _PAGE_NO_CACHE | _PAGE_USER)) ==
145 (_PAGE_PRESENT | _PAGE_USER);
Benjamin Herrenschmidt8d30c142009-02-10 16:02:37 +0000146}
147
Benjamin Herrenschmidtea3cc332009-08-18 19:00:34 +0000148struct page * maybe_pte_to_page(pte_t pte)
149{
150 unsigned long pfn = pte_pfn(pte);
151 struct page *page;
152
153 if (unlikely(!pfn_valid(pfn)))
154 return NULL;
155 page = pfn_to_page(pfn);
156 if (PageReserved(page))
157 return NULL;
158 return page;
159}
160
161#if defined(CONFIG_PPC_STD_MMU) || _PAGE_EXEC == 0
162
Benjamin Herrenschmidt8d30c142009-02-10 16:02:37 +0000163/* Server-style MMU handles coherency when hashing if HW exec permission
Benjamin Herrenschmidtea3cc332009-08-18 19:00:34 +0000164 * is supposed per page (currently 64-bit only). If not, then, we always
165 * flush the cache for valid PTEs in set_pte. Embedded CPU without HW exec
166 * support falls into the same category.
Benjamin Herrenschmidt8d30c142009-02-10 16:02:37 +0000167 */
Benjamin Herrenschmidtea3cc332009-08-18 19:00:34 +0000168
169static pte_t set_pte_filter(pte_t pte)
Benjamin Herrenschmidt8d30c142009-02-10 16:02:37 +0000170{
Benjamin Herrenschmidtea3cc332009-08-18 19:00:34 +0000171 pte = __pte(pte_val(pte) & ~_PAGE_HPTEFLAGS);
172 if (pte_looks_normal(pte) && !(cpu_has_feature(CPU_FTR_COHERENT_ICACHE) ||
173 cpu_has_feature(CPU_FTR_NOEXECUTE))) {
174 struct page *pg = maybe_pte_to_page(pte);
175 if (!pg)
176 return pte;
177 if (!test_bit(PG_arch_1, &pg->flags)) {
178 flush_dcache_icache_page(pg);
179 set_bit(PG_arch_1, &pg->flags);
180 }
181 }
182 return pte;
Benjamin Herrenschmidt8d30c142009-02-10 16:02:37 +0000183}
Benjamin Herrenschmidtea3cc332009-08-18 19:00:34 +0000184
185static pte_t set_access_flags_filter(pte_t pte, struct vm_area_struct *vma,
186 int dirty)
187{
188 return pte;
189}
190
191#else /* defined(CONFIG_PPC_STD_MMU) || _PAGE_EXEC == 0 */
192
193/* Embedded type MMU with HW exec support. This is a bit more complicated
194 * as we don't have two bits to spare for _PAGE_EXEC and _PAGE_HWEXEC so
195 * instead we "filter out" the exec permission for non clean pages.
Benjamin Herrenschmidt8d30c142009-02-10 16:02:37 +0000196 */
Benjamin Herrenschmidtea3cc332009-08-18 19:00:34 +0000197static pte_t set_pte_filter(pte_t pte)
Benjamin Herrenschmidt8d30c142009-02-10 16:02:37 +0000198{
Benjamin Herrenschmidtea3cc332009-08-18 19:00:34 +0000199 struct page *pg;
200
201 /* No exec permission in the first place, move on */
202 if (!(pte_val(pte) & _PAGE_EXEC) || !pte_looks_normal(pte))
203 return pte;
204
205 /* If you set _PAGE_EXEC on weird pages you're on your own */
206 pg = maybe_pte_to_page(pte);
207 if (unlikely(!pg))
208 return pte;
209
210 /* If the page clean, we move on */
211 if (test_bit(PG_arch_1, &pg->flags))
212 return pte;
213
214 /* If it's an exec fault, we flush the cache and make it clean */
215 if (is_exec_fault()) {
216 flush_dcache_icache_page(pg);
217 set_bit(PG_arch_1, &pg->flags);
218 return pte;
219 }
220
221 /* Else, we filter out _PAGE_EXEC */
222 return __pte(pte_val(pte) & ~_PAGE_EXEC);
Benjamin Herrenschmidt8d30c142009-02-10 16:02:37 +0000223}
Benjamin Herrenschmidtea3cc332009-08-18 19:00:34 +0000224
225static pte_t set_access_flags_filter(pte_t pte, struct vm_area_struct *vma,
226 int dirty)
Benjamin Herrenschmidt8d30c142009-02-10 16:02:37 +0000227{
Benjamin Herrenschmidtea3cc332009-08-18 19:00:34 +0000228 struct page *pg;
229
230 /* So here, we only care about exec faults, as we use them
231 * to recover lost _PAGE_EXEC and perform I$/D$ coherency
232 * if necessary. Also if _PAGE_EXEC is already set, same deal,
233 * we just bail out
234 */
235 if (dirty || (pte_val(pte) & _PAGE_EXEC) || !is_exec_fault())
236 return pte;
237
238#ifdef CONFIG_DEBUG_VM
239 /* So this is an exec fault, _PAGE_EXEC is not set. If it was
240 * an error we would have bailed out earlier in do_page_fault()
241 * but let's make sure of it
242 */
243 if (WARN_ON(!(vma->vm_flags & VM_EXEC)))
244 return pte;
245#endif /* CONFIG_DEBUG_VM */
246
247 /* If you set _PAGE_EXEC on weird pages you're on your own */
248 pg = maybe_pte_to_page(pte);
249 if (unlikely(!pg))
250 goto bail;
251
252 /* If the page is already clean, we move on */
253 if (test_bit(PG_arch_1, &pg->flags))
254 goto bail;
255
256 /* Clean the page and set PG_arch_1 */
257 flush_dcache_icache_page(pg);
258 set_bit(PG_arch_1, &pg->flags);
259
260 bail:
261 return __pte(pte_val(pte) | _PAGE_EXEC);
Benjamin Herrenschmidt8d30c142009-02-10 16:02:37 +0000262}
Benjamin Herrenschmidtea3cc332009-08-18 19:00:34 +0000263
264#endif /* !(defined(CONFIG_PPC_STD_MMU) || _PAGE_EXEC == 0) */
Benjamin Herrenschmidt8d30c142009-02-10 16:02:37 +0000265
266/*
267 * set_pte stores a linux PTE into the linux page table.
268 */
Benjamin Herrenschmidtea3cc332009-08-18 19:00:34 +0000269void set_pte_at(struct mm_struct *mm, unsigned long addr, pte_t *ptep,
270 pte_t pte)
Benjamin Herrenschmidt8d30c142009-02-10 16:02:37 +0000271{
272#ifdef CONFIG_DEBUG_VM
273 WARN_ON(pte_present(*ptep));
274#endif
275 /* Note: mm->context.id might not yet have been assigned as
276 * this context might not have been activated yet when this
277 * is called.
278 */
Benjamin Herrenschmidtea3cc332009-08-18 19:00:34 +0000279 pte = set_pte_filter(pte);
Benjamin Herrenschmidt8d30c142009-02-10 16:02:37 +0000280
281 /* Perform the setting of the PTE */
282 __set_pte_at(mm, addr, ptep, pte, 0);
283}
284
285/*
286 * This is called when relaxing access to a PTE. It's also called in the page
287 * fault path when we don't hit any of the major fault cases, ie, a minor
288 * update of _PAGE_ACCESSED, _PAGE_DIRTY, etc... The generic code will have
289 * handled those two for us, we additionally deal with missing execute
290 * permission here on some processors
291 */
292int ptep_set_access_flags(struct vm_area_struct *vma, unsigned long address,
293 pte_t *ptep, pte_t entry, int dirty)
294{
295 int changed;
Benjamin Herrenschmidtea3cc332009-08-18 19:00:34 +0000296 entry = set_access_flags_filter(entry, vma, dirty);
Benjamin Herrenschmidt8d30c142009-02-10 16:02:37 +0000297 changed = !pte_same(*(ptep), entry);
298 if (changed) {
Mel Gormanaf3e4ac2009-04-30 10:59:19 +0000299 if (!(vma->vm_flags & VM_HUGETLB))
300 assert_pte_locked(vma->vm_mm, address);
Benjamin Herrenschmidt8d30c142009-02-10 16:02:37 +0000301 __ptep_set_access_flags(ptep, entry);
302 flush_tlb_page_nohash(vma, address);
303 }
304 return changed;
305}
306
307#ifdef CONFIG_DEBUG_VM
308void assert_pte_locked(struct mm_struct *mm, unsigned long addr)
309{
310 pgd_t *pgd;
311 pud_t *pud;
312 pmd_t *pmd;
313
314 if (mm == &init_mm)
315 return;
316 pgd = mm->pgd + pgd_index(addr);
317 BUG_ON(pgd_none(*pgd));
318 pud = pud_offset(pgd, addr);
319 BUG_ON(pud_none(*pud));
320 pmd = pmd_offset(pud, addr);
321 BUG_ON(!pmd_present(*pmd));
Kumar Gala797a7472009-08-18 15:21:40 +0000322 assert_spin_locked(pte_lockptr(mm, pmd));
Benjamin Herrenschmidt8d30c142009-02-10 16:02:37 +0000323}
324#endif /* CONFIG_DEBUG_VM */
325