3 #include <asm/pgalloc.h>
4 #include <asm/pgtable.h>
6 #include <asm/fixmap.h>
8 #define PGALLOC_GFP GFP_KERNEL | __GFP_NOTRACK | __GFP_REPEAT | __GFP_ZERO
11 #define PGALLOC_USER_GFP __GFP_HIGHMEM
13 #define PGALLOC_USER_GFP 0
16 gfp_t __userpte_alloc_gfp = PGALLOC_GFP | PGALLOC_USER_GFP;
18 pte_t *pte_alloc_one_kernel(struct mm_struct *mm, unsigned long address)
20 return (pte_t *)__get_free_page(PGALLOC_GFP);
23 pgtable_t pte_alloc_one(struct mm_struct *mm, unsigned long address)
27 pte = alloc_pages(__userpte_alloc_gfp, 0);
29 pgtable_page_ctor(pte);
33 static int __init setup_userpte(char *arg)
39 * "userpte=nohigh" disables allocation of user pagetables in
42 if (strcmp(arg, "nohigh") == 0)
43 __userpte_alloc_gfp &= ~__GFP_HIGHMEM;
48 early_param("userpte", setup_userpte);
50 void ___pte_free_tlb(struct mmu_gather *tlb, struct page *pte)
52 pgtable_page_dtor(pte);
53 paravirt_release_pte(page_to_pfn(pte));
54 tlb_remove_page(tlb, pte);
57 #if PAGETABLE_LEVELS > 2
58 void ___pmd_free_tlb(struct mmu_gather *tlb, pmd_t *pmd)
60 paravirt_release_pmd(__pa(pmd) >> PAGE_SHIFT);
61 tlb_remove_page(tlb, virt_to_page(pmd));
64 #if PAGETABLE_LEVELS > 3
65 void ___pud_free_tlb(struct mmu_gather *tlb, pud_t *pud)
67 paravirt_release_pud(__pa(pud) >> PAGE_SHIFT);
68 tlb_remove_page(tlb, virt_to_page(pud));
70 #endif /* PAGETABLE_LEVELS > 3 */
71 #endif /* PAGETABLE_LEVELS > 2 */
73 static inline void pgd_list_add(pgd_t *pgd)
75 struct page *page = virt_to_page(pgd);
77 list_add(&page->lru, &pgd_list);
80 static inline void pgd_list_del(pgd_t *pgd)
82 struct page *page = virt_to_page(pgd);
87 #define UNSHARED_PTRS_PER_PGD \
88 (SHARED_KERNEL_PMD ? KERNEL_PGD_BOUNDARY : PTRS_PER_PGD)
91 static void pgd_set_mm(pgd_t *pgd, struct mm_struct *mm)
93 BUILD_BUG_ON(sizeof(virt_to_page(pgd)->index) < sizeof(mm));
94 virt_to_page(pgd)->index = (pgoff_t)mm;
97 struct mm_struct *pgd_page_get_mm(struct page *page)
99 return (struct mm_struct *)page->index;
102 static void pgd_ctor(struct mm_struct *mm, pgd_t *pgd)
104 /* If the pgd points to a shared pagetable level (either the
105 ptes in non-PAE, or shared PMD in PAE), then just copy the
106 references from swapper_pg_dir. */
107 if (PAGETABLE_LEVELS == 2 ||
108 (PAGETABLE_LEVELS == 3 && SHARED_KERNEL_PMD) ||
109 PAGETABLE_LEVELS == 4) {
110 clone_pgd_range(pgd + KERNEL_PGD_BOUNDARY,
111 swapper_pg_dir + KERNEL_PGD_BOUNDARY,
115 /* list required to sync kernel mapping updates */
116 if (!SHARED_KERNEL_PMD) {
122 static void pgd_dtor(pgd_t *pgd)
124 if (SHARED_KERNEL_PMD)
127 spin_lock(&pgd_lock);
129 spin_unlock(&pgd_lock);
133 * List of all pgd's needed for non-PAE so it can invalidate entries
134 * in both cached and uncached pgd's; not needed for PAE since the
135 * kernel pmd is shared. If PAE were not to share the pmd a similar
136 * tactic would be needed. This is essentially codepath-based locking
137 * against pageattr.c; it is the unique case in which a valid change
138 * of kernel pagetables can't be lazily synchronized by vmalloc faults.
139 * vmalloc faults work because attached pagetables are never freed.
143 #ifdef CONFIG_X86_PAE
145 * In PAE mode, we need to do a cr3 reload (=tlb flush) when
146 * updating the top-level pagetable entries to guarantee the
147 * processor notices the update. Since this is expensive, and
148 * all 4 top-level entries are used almost immediately in a
149 * new process's life, we just pre-populate them here.
151 * Also, if we're in a paravirt environment where the kernel pmd is
152 * not shared between pagetables (!SHARED_KERNEL_PMDS), we allocate
153 * and initialize the kernel pmds here.
155 #define PREALLOCATED_PMDS UNSHARED_PTRS_PER_PGD
157 void pud_populate(struct mm_struct *mm, pud_t *pudp, pmd_t *pmd)
159 paravirt_alloc_pmd(mm, __pa(pmd) >> PAGE_SHIFT);
161 /* Note: almost everything apart from _PAGE_PRESENT is
162 reserved at the pmd (PDPT) level. */
163 set_pud(pudp, __pud(__pa(pmd) | _PAGE_PRESENT));
166 * According to Intel App note "TLBs, Paging-Structure Caches,
167 * and Their Invalidation", April 2007, document 317080-001,
168 * section 8.1: in PAE mode we explicitly have to flush the
169 * TLB via cr3 if the top-level pgd is changed...
173 #else /* !CONFIG_X86_PAE */
175 /* No need to prepopulate any pagetable entries in non-PAE modes. */
176 #define PREALLOCATED_PMDS 0
178 #endif /* CONFIG_X86_PAE */
180 static void free_pmds(pmd_t *pmds[])
184 for(i = 0; i < PREALLOCATED_PMDS; i++)
186 free_page((unsigned long)pmds[i]);
189 static int preallocate_pmds(pmd_t *pmds[])
194 for(i = 0; i < PREALLOCATED_PMDS; i++) {
195 pmd_t *pmd = (pmd_t *)__get_free_page(PGALLOC_GFP);
210 * Mop up any pmd pages which may still be attached to the pgd.
211 * Normally they will be freed by munmap/exit_mmap, but any pmd we
212 * preallocate which never got a corresponding vma will need to be
215 static void pgd_mop_up_pmds(struct mm_struct *mm, pgd_t *pgdp)
219 for(i = 0; i < PREALLOCATED_PMDS; i++) {
222 if (pgd_val(pgd) != 0) {
223 pmd_t *pmd = (pmd_t *)pgd_page_vaddr(pgd);
225 pgdp[i] = native_make_pgd(0);
227 paravirt_release_pmd(pgd_val(pgd) >> PAGE_SHIFT);
233 static void pgd_prepopulate_pmd(struct mm_struct *mm, pgd_t *pgd, pmd_t *pmds[])
239 if (PREALLOCATED_PMDS == 0) /* Work around gcc-3.4.x bug */
242 pud = pud_offset(pgd, 0);
244 for (addr = i = 0; i < PREALLOCATED_PMDS;
245 i++, pud++, addr += PUD_SIZE) {
246 pmd_t *pmd = pmds[i];
248 if (i >= KERNEL_PGD_BOUNDARY)
249 memcpy(pmd, (pmd_t *)pgd_page_vaddr(swapper_pg_dir[i]),
250 sizeof(pmd_t) * PTRS_PER_PMD);
252 pud_populate(mm, pud, pmd);
256 pgd_t *pgd_alloc(struct mm_struct *mm)
259 pmd_t *pmds[PREALLOCATED_PMDS];
261 pgd = (pgd_t *)__get_free_page(PGALLOC_GFP);
268 if (preallocate_pmds(pmds) != 0)
271 if (paravirt_pgd_alloc(mm) != 0)
275 * Make sure that pre-populating the pmds is atomic with
276 * respect to anything walking the pgd_list, so that they
277 * never see a partially populated pgd.
279 spin_lock(&pgd_lock);
282 pgd_prepopulate_pmd(mm, pgd, pmds);
284 spin_unlock(&pgd_lock);
291 free_page((unsigned long)pgd);
296 void pgd_free(struct mm_struct *mm, pgd_t *pgd)
298 pgd_mop_up_pmds(mm, pgd);
300 paravirt_pgd_free(mm, pgd);
301 free_page((unsigned long)pgd);
304 int ptep_set_access_flags(struct vm_area_struct *vma,
305 unsigned long address, pte_t *ptep,
306 pte_t entry, int dirty)
308 int changed = !pte_same(*ptep, entry);
310 if (changed && dirty) {
312 pte_update_defer(vma->vm_mm, address, ptep);
313 flush_tlb_page(vma, address);
319 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
320 int pmdp_set_access_flags(struct vm_area_struct *vma,
321 unsigned long address, pmd_t *pmdp,
322 pmd_t entry, int dirty)
324 int changed = !pmd_same(*pmdp, entry);
326 VM_BUG_ON(address & ~HPAGE_PMD_MASK);
328 if (changed && dirty) {
330 pmd_update_defer(vma->vm_mm, address, pmdp);
331 flush_tlb_range(vma, address, address + HPAGE_PMD_SIZE);
338 int ptep_test_and_clear_young(struct vm_area_struct *vma,
339 unsigned long addr, pte_t *ptep)
343 if (pte_young(*ptep))
344 ret = test_and_clear_bit(_PAGE_BIT_ACCESSED,
345 (unsigned long *) &ptep->pte);
348 pte_update(vma->vm_mm, addr, ptep);
353 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
354 int pmdp_test_and_clear_young(struct vm_area_struct *vma,
355 unsigned long addr, pmd_t *pmdp)
359 if (pmd_young(*pmdp))
360 ret = test_and_clear_bit(_PAGE_BIT_ACCESSED,
361 (unsigned long *)pmdp);
364 pmd_update(vma->vm_mm, addr, pmdp);
370 int ptep_clear_flush_young(struct vm_area_struct *vma,
371 unsigned long address, pte_t *ptep)
375 young = ptep_test_and_clear_young(vma, address, ptep);
377 flush_tlb_page(vma, address);
382 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
383 int pmdp_clear_flush_young(struct vm_area_struct *vma,
384 unsigned long address, pmd_t *pmdp)
388 VM_BUG_ON(address & ~HPAGE_PMD_MASK);
390 young = pmdp_test_and_clear_young(vma, address, pmdp);
392 flush_tlb_range(vma, address, address + HPAGE_PMD_SIZE);
397 void pmdp_splitting_flush(struct vm_area_struct *vma,
398 unsigned long address, pmd_t *pmdp)
401 VM_BUG_ON(address & ~HPAGE_PMD_MASK);
402 set = !test_and_set_bit(_PAGE_BIT_SPLITTING,
403 (unsigned long *)pmdp);
405 pmd_update(vma->vm_mm, address, pmdp);
406 /* need tlb flush only to serialize against gup-fast */
407 flush_tlb_range(vma, address, address + HPAGE_PMD_SIZE);
413 * reserve_top_address - reserves a hole in the top of kernel address space
414 * @reserve - size of hole to reserve
416 * Can be used to relocate the fixmap area and poke a hole in the top
417 * of kernel address space to make room for a hypervisor.
419 void __init reserve_top_address(unsigned long reserve)
422 BUG_ON(fixmaps_set > 0);
423 printk(KERN_INFO "Reserving virtual address space above 0x%08x\n",
425 __FIXADDR_TOP = -reserve - PAGE_SIZE;
431 void __native_set_fixmap(enum fixed_addresses idx, pte_t pte)
433 unsigned long address = __fix_to_virt(idx);
435 if (idx >= __end_of_fixed_addresses) {
439 set_pte_vaddr(address, pte);
443 void native_set_fixmap(enum fixed_addresses idx, phys_addr_t phys,
446 __native_set_fixmap(idx, pfn_pte(phys >> PAGE_SHIFT, flags));