blob: 8e43bdd45456017cd431c2cf9b4678c5b7e9f65d [file] [log] [blame]
Jeremy Fitzhardinge4f76cd32008-03-17 16:36:55 -07001#include <linux/mm.h>
2#include <asm/pgalloc.h>
Jeremy Fitzhardingeee5aa8d2008-03-17 16:37:03 -07003#include <asm/pgtable.h>
Jeremy Fitzhardinge4f76cd32008-03-17 16:36:55 -07004#include <asm/tlb.h>
Ingo Molnara1d5a862008-06-20 15:34:46 +02005#include <asm/fixmap.h>
Jeremy Fitzhardinge4f76cd32008-03-17 16:36:55 -07006
Vegard Nossum9e730232009-02-22 11:28:25 +01007#define PGALLOC_GFP GFP_KERNEL | __GFP_NOTRACK | __GFP_REPEAT | __GFP_ZERO
8
Jeremy Fitzhardinge4f76cd32008-03-17 16:36:55 -07009pte_t *pte_alloc_one_kernel(struct mm_struct *mm, unsigned long address)
10{
Vegard Nossum9e730232009-02-22 11:28:25 +010011 return (pte_t *)__get_free_page(PGALLOC_GFP);
Jeremy Fitzhardinge4f76cd32008-03-17 16:36:55 -070012}
13
14pgtable_t pte_alloc_one(struct mm_struct *mm, unsigned long address)
15{
16 struct page *pte;
17
18#ifdef CONFIG_HIGHPTE
Vegard Nossum9e730232009-02-22 11:28:25 +010019 pte = alloc_pages(PGALLOC_GFP | __GFP_HIGHMEM, 0);
Jeremy Fitzhardinge4f76cd32008-03-17 16:36:55 -070020#else
Vegard Nossum9e730232009-02-22 11:28:25 +010021 pte = alloc_pages(PGALLOC_GFP, 0);
Jeremy Fitzhardinge4f76cd32008-03-17 16:36:55 -070022#endif
23 if (pte)
24 pgtable_page_ctor(pte);
25 return pte;
26}
27
Jeremy Fitzhardinge397f6872008-03-17 16:36:57 -070028void __pte_free_tlb(struct mmu_gather *tlb, struct page *pte)
29{
30 pgtable_page_dtor(pte);
Jeremy Fitzhardinge6944a9c2008-03-17 16:37:01 -070031 paravirt_release_pte(page_to_pfn(pte));
Jeremy Fitzhardinge397f6872008-03-17 16:36:57 -070032 tlb_remove_page(tlb, pte);
33}
34
Jeremy Fitzhardinge170fdff2008-03-17 16:36:58 -070035#if PAGETABLE_LEVELS > 2
36void __pmd_free_tlb(struct mmu_gather *tlb, pmd_t *pmd)
37{
Jeremy Fitzhardinge6944a9c2008-03-17 16:37:01 -070038 paravirt_release_pmd(__pa(pmd) >> PAGE_SHIFT);
Jeremy Fitzhardinge170fdff2008-03-17 16:36:58 -070039 tlb_remove_page(tlb, virt_to_page(pmd));
40}
Jeremy Fitzhardinge5a5f8f42008-03-17 16:36:59 -070041
42#if PAGETABLE_LEVELS > 3
43void __pud_free_tlb(struct mmu_gather *tlb, pud_t *pud)
44{
Jeremy Fitzhardinge2761fa02008-03-17 16:37:02 -070045 paravirt_release_pud(__pa(pud) >> PAGE_SHIFT);
Jeremy Fitzhardinge5a5f8f42008-03-17 16:36:59 -070046 tlb_remove_page(tlb, virt_to_page(pud));
47}
48#endif /* PAGETABLE_LEVELS > 3 */
Jeremy Fitzhardinge170fdff2008-03-17 16:36:58 -070049#endif /* PAGETABLE_LEVELS > 2 */
50
Jeremy Fitzhardinge4f76cd32008-03-17 16:36:55 -070051static inline void pgd_list_add(pgd_t *pgd)
52{
53 struct page *page = virt_to_page(pgd);
Jeremy Fitzhardinge4f76cd32008-03-17 16:36:55 -070054
Jeremy Fitzhardinge4f76cd32008-03-17 16:36:55 -070055 list_add(&page->lru, &pgd_list);
Jeremy Fitzhardinge4f76cd32008-03-17 16:36:55 -070056}
57
58static inline void pgd_list_del(pgd_t *pgd)
59{
60 struct page *page = virt_to_page(pgd);
Jeremy Fitzhardinge4f76cd32008-03-17 16:36:55 -070061
Jeremy Fitzhardinge4f76cd32008-03-17 16:36:55 -070062 list_del(&page->lru);
Jeremy Fitzhardinge4f76cd32008-03-17 16:36:55 -070063}
64
Jeremy Fitzhardinge4f76cd32008-03-17 16:36:55 -070065#define UNSHARED_PTRS_PER_PGD \
Jeremy Fitzhardinge68db0652008-03-17 16:37:13 -070066 (SHARED_KERNEL_PMD ? KERNEL_PGD_BOUNDARY : PTRS_PER_PGD)
Jeremy Fitzhardinge4f76cd32008-03-17 16:36:55 -070067
Jan Beulich17b74622008-08-29 12:51:32 +010068static void pgd_ctor(pgd_t *pgd)
Jeremy Fitzhardinge4f76cd32008-03-17 16:36:55 -070069{
Jeremy Fitzhardinge4f76cd32008-03-17 16:36:55 -070070 /* If the pgd points to a shared pagetable level (either the
71 ptes in non-PAE, or shared PMD in PAE), then just copy the
72 references from swapper_pg_dir. */
73 if (PAGETABLE_LEVELS == 2 ||
Jeremy Fitzhardinge85958b42008-03-17 16:37:14 -070074 (PAGETABLE_LEVELS == 3 && SHARED_KERNEL_PMD) ||
75 PAGETABLE_LEVELS == 4) {
Jeremy Fitzhardinge68db0652008-03-17 16:37:13 -070076 clone_pgd_range(pgd + KERNEL_PGD_BOUNDARY,
77 swapper_pg_dir + KERNEL_PGD_BOUNDARY,
Jeremy Fitzhardinge4f76cd32008-03-17 16:36:55 -070078 KERNEL_PGD_PTRS);
Jeremy Fitzhardinge6944a9c2008-03-17 16:37:01 -070079 paravirt_alloc_pmd_clone(__pa(pgd) >> PAGE_SHIFT,
80 __pa(swapper_pg_dir) >> PAGE_SHIFT,
Jeremy Fitzhardinge68db0652008-03-17 16:37:13 -070081 KERNEL_PGD_BOUNDARY,
Jeremy Fitzhardinge6944a9c2008-03-17 16:37:01 -070082 KERNEL_PGD_PTRS);
Jeremy Fitzhardinge4f76cd32008-03-17 16:36:55 -070083 }
84
85 /* list required to sync kernel mapping updates */
86 if (!SHARED_KERNEL_PMD)
87 pgd_list_add(pgd);
Jeremy Fitzhardinge4f76cd32008-03-17 16:36:55 -070088}
89
Jan Beulich17b74622008-08-29 12:51:32 +010090static void pgd_dtor(pgd_t *pgd)
Jeremy Fitzhardinge4f76cd32008-03-17 16:36:55 -070091{
92 unsigned long flags; /* can be called from interrupt context */
93
94 if (SHARED_KERNEL_PMD)
95 return;
96
97 spin_lock_irqsave(&pgd_lock, flags);
98 pgd_list_del(pgd);
99 spin_unlock_irqrestore(&pgd_lock, flags);
100}
101
Jeremy Fitzhardinge85958b42008-03-17 16:37:14 -0700102/*
103 * List of all pgd's needed for non-PAE so it can invalidate entries
104 * in both cached and uncached pgd's; not needed for PAE since the
105 * kernel pmd is shared. If PAE were not to share the pmd a similar
106 * tactic would be needed. This is essentially codepath-based locking
107 * against pageattr.c; it is the unique case in which a valid change
108 * of kernel pagetables can't be lazily synchronized by vmalloc faults.
109 * vmalloc faults work because attached pagetables are never freed.
110 * -- wli
111 */
112
Jeremy Fitzhardinge4f76cd32008-03-17 16:36:55 -0700113#ifdef CONFIG_X86_PAE
114/*
Jeremy Fitzhardinge4f76cd32008-03-17 16:36:55 -0700115 * In PAE mode, we need to do a cr3 reload (=tlb flush) when
116 * updating the top-level pagetable entries to guarantee the
117 * processor notices the update. Since this is expensive, and
118 * all 4 top-level entries are used almost immediately in a
119 * new process's life, we just pre-populate them here.
120 *
121 * Also, if we're in a paravirt environment where the kernel pmd is
122 * not shared between pagetables (!SHARED_KERNEL_PMDS), we allocate
123 * and initialize the kernel pmds here.
124 */
Jeremy Fitzhardinged8d59002008-06-25 00:19:13 -0400125#define PREALLOCATED_PMDS UNSHARED_PTRS_PER_PGD
Ingo Molnar1ec1fe72008-03-19 20:30:40 +0100126
127void pud_populate(struct mm_struct *mm, pud_t *pudp, pmd_t *pmd)
128{
Jeremy Fitzhardinge6944a9c2008-03-17 16:37:01 -0700129 paravirt_alloc_pmd(mm, __pa(pmd) >> PAGE_SHIFT);
Ingo Molnar1ec1fe72008-03-19 20:30:40 +0100130
131 /* Note: almost everything apart from _PAGE_PRESENT is
132 reserved at the pmd (PDPT) level. */
133 set_pud(pudp, __pud(__pa(pmd) | _PAGE_PRESENT));
134
135 /*
136 * According to Intel App note "TLBs, Paging-Structure Caches,
137 * and Their Invalidation", April 2007, document 317080-001,
138 * section 8.1: in PAE mode we explicitly have to flush the
139 * TLB via cr3 if the top-level pgd is changed...
140 */
141 if (mm == current->active_mm)
142 write_cr3(read_cr3());
143}
Jeremy Fitzhardinge4f76cd32008-03-17 16:36:55 -0700144#else /* !CONFIG_X86_PAE */
Jeremy Fitzhardinged8d59002008-06-25 00:19:13 -0400145
Jeremy Fitzhardinge4f76cd32008-03-17 16:36:55 -0700146/* No need to prepopulate any pagetable entries in non-PAE modes. */
Jeremy Fitzhardinged8d59002008-06-25 00:19:13 -0400147#define PREALLOCATED_PMDS 0
148
149#endif /* CONFIG_X86_PAE */
150
151static void free_pmds(pmd_t *pmds[])
Jeremy Fitzhardinge4f76cd32008-03-17 16:36:55 -0700152{
Jeremy Fitzhardinged8d59002008-06-25 00:19:13 -0400153 int i;
154
155 for(i = 0; i < PREALLOCATED_PMDS; i++)
156 if (pmds[i])
157 free_page((unsigned long)pmds[i]);
Jeremy Fitzhardinge4f76cd32008-03-17 16:36:55 -0700158}
159
Jeremy Fitzhardinged8d59002008-06-25 00:19:13 -0400160static int preallocate_pmds(pmd_t *pmds[])
Jeremy Fitzhardinge4f76cd32008-03-17 16:36:55 -0700161{
Jeremy Fitzhardinged8d59002008-06-25 00:19:13 -0400162 int i;
163 bool failed = false;
164
165 for(i = 0; i < PREALLOCATED_PMDS; i++) {
Vegard Nossum9e730232009-02-22 11:28:25 +0100166 pmd_t *pmd = (pmd_t *)__get_free_page(PGALLOC_GFP);
Jeremy Fitzhardinged8d59002008-06-25 00:19:13 -0400167 if (pmd == NULL)
168 failed = true;
169 pmds[i] = pmd;
170 }
171
172 if (failed) {
173 free_pmds(pmds);
174 return -ENOMEM;
175 }
176
177 return 0;
Jeremy Fitzhardinge4f76cd32008-03-17 16:36:55 -0700178}
Jeremy Fitzhardinged8d59002008-06-25 00:19:13 -0400179
180/*
181 * Mop up any pmd pages which may still be attached to the pgd.
182 * Normally they will be freed by munmap/exit_mmap, but any pmd we
183 * preallocate which never got a corresponding vma will need to be
184 * freed manually.
185 */
186static void pgd_mop_up_pmds(struct mm_struct *mm, pgd_t *pgdp)
187{
188 int i;
189
190 for(i = 0; i < PREALLOCATED_PMDS; i++) {
191 pgd_t pgd = pgdp[i];
192
193 if (pgd_val(pgd) != 0) {
194 pmd_t *pmd = (pmd_t *)pgd_page_vaddr(pgd);
195
196 pgdp[i] = native_make_pgd(0);
197
198 paravirt_release_pmd(pgd_val(pgd) >> PAGE_SHIFT);
199 pmd_free(mm, pmd);
200 }
201 }
202}
203
204static void pgd_prepopulate_pmd(struct mm_struct *mm, pgd_t *pgd, pmd_t *pmds[])
205{
206 pud_t *pud;
207 unsigned long addr;
208 int i;
209
Jeremy Fitzhardingecf3e5052008-08-08 13:46:07 -0700210 if (PREALLOCATED_PMDS == 0) /* Work around gcc-3.4.x bug */
211 return;
212
Jeremy Fitzhardinged8d59002008-06-25 00:19:13 -0400213 pud = pud_offset(pgd, 0);
214
215 for (addr = i = 0; i < PREALLOCATED_PMDS;
216 i++, pud++, addr += PUD_SIZE) {
217 pmd_t *pmd = pmds[i];
218
219 if (i >= KERNEL_PGD_BOUNDARY)
220 memcpy(pmd, (pmd_t *)pgd_page_vaddr(swapper_pg_dir[i]),
221 sizeof(pmd_t) * PTRS_PER_PMD);
222
223 pud_populate(mm, pud, pmd);
224 }
225}
Jeremy Fitzhardinge4f76cd32008-03-17 16:36:55 -0700226
227pgd_t *pgd_alloc(struct mm_struct *mm)
228{
Jeremy Fitzhardinged8d59002008-06-25 00:19:13 -0400229 pgd_t *pgd;
230 pmd_t *pmds[PREALLOCATED_PMDS];
231 unsigned long flags;
Jeremy Fitzhardinge4f76cd32008-03-17 16:36:55 -0700232
Vegard Nossum9e730232009-02-22 11:28:25 +0100233 pgd = (pgd_t *)__get_free_page(PGALLOC_GFP);
Jeremy Fitzhardinged8d59002008-06-25 00:19:13 -0400234
235 if (pgd == NULL)
236 goto out;
237
Jeremy Fitzhardinge4f76cd32008-03-17 16:36:55 -0700238 mm->pgd = pgd;
Jeremy Fitzhardinge4f76cd32008-03-17 16:36:55 -0700239
Jeremy Fitzhardinged8d59002008-06-25 00:19:13 -0400240 if (preallocate_pmds(pmds) != 0)
241 goto out_free_pgd;
242
243 if (paravirt_pgd_alloc(mm) != 0)
244 goto out_free_pmds;
245
246 /*
247 * Make sure that pre-populating the pmds is atomic with
248 * respect to anything walking the pgd_list, so that they
249 * never see a partially populated pgd.
250 */
251 spin_lock_irqsave(&pgd_lock, flags);
252
253 pgd_ctor(pgd);
254 pgd_prepopulate_pmd(mm, pgd, pmds);
255
256 spin_unlock_irqrestore(&pgd_lock, flags);
Jeremy Fitzhardinge4f76cd32008-03-17 16:36:55 -0700257
258 return pgd;
Jeremy Fitzhardinged8d59002008-06-25 00:19:13 -0400259
260out_free_pmds:
261 free_pmds(pmds);
262out_free_pgd:
263 free_page((unsigned long)pgd);
264out:
265 return NULL;
Jeremy Fitzhardinge4f76cd32008-03-17 16:36:55 -0700266}
267
268void pgd_free(struct mm_struct *mm, pgd_t *pgd)
269{
270 pgd_mop_up_pmds(mm, pgd);
271 pgd_dtor(pgd);
Jeremy Fitzhardingeeba00452008-06-25 00:19:12 -0400272 paravirt_pgd_free(mm, pgd);
Jeremy Fitzhardinge4f76cd32008-03-17 16:36:55 -0700273 free_page((unsigned long)pgd);
274}
Jeremy Fitzhardingeee5aa8d2008-03-17 16:37:03 -0700275
276int ptep_set_access_flags(struct vm_area_struct *vma,
277 unsigned long address, pte_t *ptep,
278 pte_t entry, int dirty)
279{
280 int changed = !pte_same(*ptep, entry);
281
282 if (changed && dirty) {
283 *ptep = entry;
284 pte_update_defer(vma->vm_mm, address, ptep);
285 flush_tlb_page(vma, address);
286 }
287
288 return changed;
289}
Jeremy Fitzhardingef9fbf1a2008-03-17 16:37:04 -0700290
291int ptep_test_and_clear_young(struct vm_area_struct *vma,
292 unsigned long addr, pte_t *ptep)
293{
294 int ret = 0;
295
296 if (pte_young(*ptep))
297 ret = test_and_clear_bit(_PAGE_BIT_ACCESSED,
Thomas Gleixner48e23952008-05-24 17:24:34 +0200298 (unsigned long *) &ptep->pte);
Jeremy Fitzhardingef9fbf1a2008-03-17 16:37:04 -0700299
300 if (ret)
301 pte_update(vma->vm_mm, addr, ptep);
302
303 return ret;
304}
Jeremy Fitzhardingec20311e2008-03-17 16:37:05 -0700305
306int ptep_clear_flush_young(struct vm_area_struct *vma,
307 unsigned long address, pte_t *ptep)
308{
309 int young;
310
311 young = ptep_test_and_clear_young(vma, address, ptep);
312 if (young)
313 flush_tlb_page(vma, address);
314
315 return young;
316}
Jeremy Fitzhardinge7c7e6e02008-06-17 11:41:54 -0700317
Gustavo F. Padovanfd862dd2009-02-15 21:48:54 -0300318/**
319 * reserve_top_address - reserves a hole in the top of kernel address space
320 * @reserve - size of hole to reserve
321 *
322 * Can be used to relocate the fixmap area and poke a hole in the top
323 * of kernel address space to make room for a hypervisor.
324 */
325void __init reserve_top_address(unsigned long reserve)
326{
327#ifdef CONFIG_X86_32
328 BUG_ON(fixmaps_set > 0);
329 printk(KERN_INFO "Reserving virtual address space above 0x%08x\n",
330 (int)-reserve);
331 __FIXADDR_TOP = -reserve - PAGE_SIZE;
332 __VMALLOC_RESERVE += reserve;
333#endif
334}
335
Jeremy Fitzhardinge7c7e6e02008-06-17 11:41:54 -0700336int fixmaps_set;
337
Jeremy Fitzhardingeaeaaa592008-06-17 11:42:01 -0700338void __native_set_fixmap(enum fixed_addresses idx, pte_t pte)
Jeremy Fitzhardinge7c7e6e02008-06-17 11:41:54 -0700339{
340 unsigned long address = __fix_to_virt(idx);
341
342 if (idx >= __end_of_fixed_addresses) {
343 BUG();
344 return;
345 }
Jeremy Fitzhardingeaeaaa592008-06-17 11:42:01 -0700346 set_pte_vaddr(address, pte);
Jeremy Fitzhardinge7c7e6e02008-06-17 11:41:54 -0700347 fixmaps_set++;
348}
Jeremy Fitzhardingeaeaaa592008-06-17 11:42:01 -0700349
Masami Hiramatsu3b3809a2009-04-09 10:55:33 -0700350void native_set_fixmap(enum fixed_addresses idx, phys_addr_t phys,
351 pgprot_t flags)
Jeremy Fitzhardingeaeaaa592008-06-17 11:42:01 -0700352{
353 __native_set_fixmap(idx, pfn_pte(phys >> PAGE_SHIFT, flags));
354}