blob: 8927222b3ab2595202afcbb5f88810a8f8716a02 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * Copyright 2002 Andi Kleen, SuSE Labs.
3 * Thanks to Ben LaHaise for precious feedback.
4 */
5
Linus Torvalds1da177e2005-04-16 15:20:36 -07006#include <linux/mm.h>
7#include <linux/sched.h>
8#include <linux/highmem.h>
9#include <linux/module.h>
10#include <linux/slab.h>
11#include <asm/uaccess.h>
12#include <asm/processor.h>
13#include <asm/tlbflush.h>
Zachary Amsdenc9b02a22005-09-03 15:56:40 -070014#include <asm/pgalloc.h>
Dave Jonesf8af0952006-01-06 00:12:10 -080015#include <asm/sections.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070016
17static DEFINE_SPINLOCK(cpa_lock);
18static struct list_head df_list = LIST_HEAD_INIT(df_list);
19
20
21pte_t *lookup_address(unsigned long address)
22{
23 pgd_t *pgd = pgd_offset_k(address);
24 pud_t *pud;
25 pmd_t *pmd;
26 if (pgd_none(*pgd))
27 return NULL;
28 pud = pud_offset(pgd, address);
29 if (pud_none(*pud))
30 return NULL;
31 pmd = pmd_offset(pud, address);
32 if (pmd_none(*pmd))
33 return NULL;
34 if (pmd_large(*pmd))
35 return (pte_t *)pmd;
36 return pte_offset_kernel(pmd, address);
37}
38
Dave Jonesf8af0952006-01-06 00:12:10 -080039static struct page *split_large_page(unsigned long address, pgprot_t prot,
40 pgprot_t ref_prot)
Linus Torvalds1da177e2005-04-16 15:20:36 -070041{
42 int i;
43 unsigned long addr;
44 struct page *base;
45 pte_t *pbase;
46
47 spin_unlock_irq(&cpa_lock);
48 base = alloc_pages(GFP_KERNEL, 0);
49 spin_lock_irq(&cpa_lock);
50 if (!base)
51 return NULL;
52
Nick Piggin84d1c052006-03-22 00:08:31 -080053 /*
54 * page_private is used to track the number of entries in
55 * the page table page that have non standard attributes.
56 */
57 SetPagePrivate(base);
58 page_private(base) = 0;
59
Linus Torvalds1da177e2005-04-16 15:20:36 -070060 address = __pa(address);
61 addr = address & LARGE_PAGE_MASK;
62 pbase = (pte_t *)page_address(base);
Jeremy Fitzhardingefdb4c332007-07-17 18:37:03 -070063 paravirt_alloc_pt(&init_mm, page_to_pfn(base));
Linus Torvalds1da177e2005-04-16 15:20:36 -070064 for (i = 0; i < PTRS_PER_PTE; i++, addr += PAGE_SIZE) {
Zachary Amsdenc9b02a22005-09-03 15:56:40 -070065 set_pte(&pbase[i], pfn_pte(addr >> PAGE_SHIFT,
Dave Jonesf8af0952006-01-06 00:12:10 -080066 addr == address ? prot : ref_prot));
Linus Torvalds1da177e2005-04-16 15:20:36 -070067 }
68 return base;
69}
70
Andi Kleen018d2ad2007-06-20 12:23:36 +020071static void cache_flush_page(struct page *p)
Linus Torvalds1da177e2005-04-16 15:20:36 -070072{
Andi Kleen018d2ad2007-06-20 12:23:36 +020073 unsigned long adr = (unsigned long)page_address(p);
74 int i;
75 for (i = 0; i < PAGE_SIZE; i += boot_cpu_data.x86_clflush_size)
76 asm volatile("clflush (%0)" :: "r" (adr + i));
77}
Andi Kleen3760dd62006-12-07 02:14:05 +010078
Andi Kleen018d2ad2007-06-20 12:23:36 +020079static void flush_kernel_map(void *arg)
80{
81 struct list_head *lh = (struct list_head *)arg;
82 struct page *p;
83
84 /* High level code is not ready for clflush yet */
Andi Kleen65d2f0b2007-07-21 17:09:51 +020085 if (cpu_has_clflush) {
Andi Kleen018d2ad2007-06-20 12:23:36 +020086 list_for_each_entry (p, lh, lru)
87 cache_flush_page(p);
Andi Kleen3760dd62006-12-07 02:14:05 +010088 } else if (boot_cpu_data.x86_model >= 4)
Zachary Amsden4bb0d3e2005-09-03 15:56:36 -070089 wbinvd();
Andi Kleen3760dd62006-12-07 02:14:05 +010090
Linus Torvalds1da177e2005-04-16 15:20:36 -070091 /* Flush all to work around Errata in early athlons regarding
92 * large page flushing.
93 */
94 __flush_tlb_all();
95}
96
97static void set_pmd_pte(pte_t *kpte, unsigned long address, pte_t pte)
98{
99 struct page *page;
100 unsigned long flags;
101
102 set_pte_atomic(kpte, pte); /* change init_mm */
Jeremy Fitzhardinge5311ab62007-05-02 19:27:13 +0200103 if (SHARED_KERNEL_PMD)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700104 return;
105
106 spin_lock_irqsave(&pgd_lock, flags);
107 for (page = pgd_list; page; page = (struct page *)page->index) {
108 pgd_t *pgd;
109 pud_t *pud;
110 pmd_t *pmd;
111 pgd = (pgd_t *)page_address(page) + pgd_index(address);
112 pud = pud_offset(pgd, address);
113 pmd = pmd_offset(pud, address);
114 set_pte_atomic((pte_t *)pmd, pte);
115 }
116 spin_unlock_irqrestore(&pgd_lock, flags);
117}
118
119/*
120 * No more special protections in this 2/4MB area - revert to a
121 * large page again.
122 */
123static inline void revert_page(struct page *kpte_page, unsigned long address)
124{
Dave Jonesf8af0952006-01-06 00:12:10 -0800125 pgprot_t ref_prot;
126 pte_t *linear;
127
128 ref_prot =
129 ((address & LARGE_PAGE_MASK) < (unsigned long)&_etext)
130 ? PAGE_KERNEL_LARGE_EXEC : PAGE_KERNEL_LARGE;
131
132 linear = (pte_t *)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700133 pmd_offset(pud_offset(pgd_offset_k(address), address), address);
134 set_pmd_pte(linear, address,
135 pfn_pte((__pa(address) & LARGE_PAGE_MASK) >> PAGE_SHIFT,
Dave Jonesf8af0952006-01-06 00:12:10 -0800136 ref_prot));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700137}
138
Andi Kleen65d2f0b2007-07-21 17:09:51 +0200139static inline void save_page(struct page *kpte_page)
140{
141 if (!test_and_set_bit(PG_arch_1, &kpte_page->flags))
142 list_add(&kpte_page->lru, &df_list);
143}
144
Linus Torvalds1da177e2005-04-16 15:20:36 -0700145static int
146__change_page_attr(struct page *page, pgprot_t prot)
147{
148 pte_t *kpte;
149 unsigned long address;
150 struct page *kpte_page;
151
152 BUG_ON(PageHighMem(page));
153 address = (unsigned long)page_address(page);
154
155 kpte = lookup_address(address);
156 if (!kpte)
157 return -EINVAL;
158 kpte_page = virt_to_page(kpte);
Andi Kleen65d2f0b2007-07-21 17:09:51 +0200159 BUG_ON(PageLRU(kpte_page));
160 BUG_ON(PageCompound(kpte_page));
161
Linus Torvalds1da177e2005-04-16 15:20:36 -0700162 if (pgprot_val(prot) != pgprot_val(PAGE_KERNEL)) {
Jan Beulichd01ad8d2007-05-02 19:27:10 +0200163 if (!pte_huge(*kpte)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700164 set_pte_atomic(kpte, mk_pte(page, prot));
165 } else {
Dave Jonesf8af0952006-01-06 00:12:10 -0800166 pgprot_t ref_prot;
167 struct page *split;
168
169 ref_prot =
170 ((address & LARGE_PAGE_MASK) < (unsigned long)&_etext)
171 ? PAGE_KERNEL_EXEC : PAGE_KERNEL;
172 split = split_large_page(address, prot, ref_prot);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700173 if (!split)
174 return -ENOMEM;
Dave Jonesf8af0952006-01-06 00:12:10 -0800175 set_pmd_pte(kpte,address,mk_pte(split, ref_prot));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700176 kpte_page = split;
Nick Piggin84d1c052006-03-22 00:08:31 -0800177 }
178 page_private(kpte_page)++;
Jan Beulichd01ad8d2007-05-02 19:27:10 +0200179 } else if (!pte_huge(*kpte)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700180 set_pte_atomic(kpte, mk_pte(page, PAGE_KERNEL));
Nick Piggin84d1c052006-03-22 00:08:31 -0800181 BUG_ON(page_private(kpte_page) == 0);
182 page_private(kpte_page)--;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700183 } else
184 BUG();
185
186 /*
187 * If the pte was reserved, it means it was created at boot
188 * time (not via split_large_page) and in turn we must not
189 * replace it with a largepage.
190 */
Andi Kleen65d2f0b2007-07-21 17:09:51 +0200191
192 save_page(kpte_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700193 if (!PageReserved(kpte_page)) {
Nick Piggin84d1c052006-03-22 00:08:31 -0800194 if (cpu_has_pse && (page_private(kpte_page) == 0)) {
Zachary Amsdenc119ecc2007-02-13 13:26:21 +0100195 paravirt_release_pt(page_to_pfn(kpte_page));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700196 revert_page(kpte_page, address);
197 }
198 }
199 return 0;
200}
201
Andi Kleen018d2ad2007-06-20 12:23:36 +0200202static inline void flush_map(struct list_head *l)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700203{
Andi Kleen018d2ad2007-06-20 12:23:36 +0200204 on_each_cpu(flush_kernel_map, l, 1, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700205}
206
207/*
208 * Change the page attributes of an page in the linear mapping.
209 *
210 * This should be used when a page is mapped with a different caching policy
211 * than write-back somewhere - some CPUs do not like it when mappings with
212 * different caching policies exist. This changes the page attributes of the
213 * in kernel linear mapping too.
214 *
215 * The caller needs to ensure that there are no conflicting mappings elsewhere.
216 * This function only deals with the kernel linear map.
217 *
218 * Caller must call global_flush_tlb() after this.
219 */
220int change_page_attr(struct page *page, int numpages, pgprot_t prot)
221{
222 int err = 0;
223 int i;
224 unsigned long flags;
225
226 spin_lock_irqsave(&cpa_lock, flags);
227 for (i = 0; i < numpages; i++, page++) {
228 err = __change_page_attr(page, prot);
229 if (err)
230 break;
231 }
232 spin_unlock_irqrestore(&cpa_lock, flags);
233 return err;
234}
235
236void global_flush_tlb(void)
Oleg Nesterov626ab0e2006-06-23 02:05:55 -0700237{
238 struct list_head l;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700239 struct page *pg, *next;
240
241 BUG_ON(irqs_disabled());
242
243 spin_lock_irq(&cpa_lock);
Oleg Nesterov626ab0e2006-06-23 02:05:55 -0700244 list_replace_init(&df_list, &l);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700245 spin_unlock_irq(&cpa_lock);
Andi Kleen018d2ad2007-06-20 12:23:36 +0200246 flush_map(&l);
Andi Kleen3760dd62006-12-07 02:14:05 +0100247 list_for_each_entry_safe(pg, next, &l, lru) {
Andi Kleen65d2f0b2007-07-21 17:09:51 +0200248 list_del(&pg->lru);
249 clear_bit(PG_arch_1, &pg->flags);
250 if (PageReserved(pg) || !cpu_has_pse || page_private(pg) != 0)
251 continue;
252 ClearPagePrivate(pg);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700253 __free_page(pg);
Andi Kleen3760dd62006-12-07 02:14:05 +0100254 }
Oleg Nesterov626ab0e2006-06-23 02:05:55 -0700255}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700256
257#ifdef CONFIG_DEBUG_PAGEALLOC
258void kernel_map_pages(struct page *page, int numpages, int enable)
259{
260 if (PageHighMem(page))
261 return;
Ingo Molnarde5097c2006-01-09 15:59:21 -0800262 if (!enable)
Ingo Molnarf9b84042006-06-27 02:54:49 -0700263 debug_check_no_locks_freed(page_address(page),
264 numpages * PAGE_SIZE);
Ingo Molnarde5097c2006-01-09 15:59:21 -0800265
Linus Torvalds1da177e2005-04-16 15:20:36 -0700266 /* the return value is ignored - the calls cannot fail,
267 * large pages are disabled at boot time.
268 */
269 change_page_attr(page, numpages, enable ? PAGE_KERNEL : __pgprot(0));
270 /* we should perform an IPI and flush all tlbs,
271 * but that can deadlock->flush only current cpu.
272 */
273 __flush_tlb_all();
274}
275#endif
276
277EXPORT_SYMBOL(change_page_attr);
278EXPORT_SYMBOL(global_flush_tlb);