]>
Commit | Line | Data |
---|---|---|
9f4c815c IM |
1 | /* |
2 | * Copyright 2002 Andi Kleen, SuSE Labs. | |
1da177e4 | 3 | * Thanks to Ben LaHaise for precious feedback. |
9f4c815c | 4 | */ |
1da177e4 | 5 | #include <linux/highmem.h> |
8192206d | 6 | #include <linux/bootmem.h> |
9f4c815c | 7 | #include <linux/sched.h> |
9f4c815c | 8 | #include <linux/mm.h> |
76ebd054 | 9 | #include <linux/interrupt.h> |
ee7ae7a1 TG |
10 | #include <linux/seq_file.h> |
11 | #include <linux/debugfs.h> | |
e59a1bb2 | 12 | #include <linux/pfn.h> |
8c4bfc6e | 13 | #include <linux/percpu.h> |
5a0e3ad6 | 14 | #include <linux/gfp.h> |
5bd5a452 | 15 | #include <linux/pci.h> |
d6472302 | 16 | #include <linux/vmalloc.h> |
9f4c815c | 17 | |
950f9d95 | 18 | #include <asm/e820.h> |
1da177e4 LT |
19 | #include <asm/processor.h> |
20 | #include <asm/tlbflush.h> | |
f8af095d | 21 | #include <asm/sections.h> |
93dbda7c | 22 | #include <asm/setup.h> |
9f4c815c IM |
23 | #include <asm/uaccess.h> |
24 | #include <asm/pgalloc.h> | |
c31c7d48 | 25 | #include <asm/proto.h> |
1219333d | 26 | #include <asm/pat.h> |
1da177e4 | 27 | |
9df84993 IM |
28 | /* |
29 | * The current flushing context - we pass it instead of 5 arguments: | |
30 | */ | |
72e458df | 31 | struct cpa_data { |
d75586ad | 32 | unsigned long *vaddr; |
0fd64c23 | 33 | pgd_t *pgd; |
72e458df TG |
34 | pgprot_t mask_set; |
35 | pgprot_t mask_clr; | |
74256377 | 36 | unsigned long numpages; |
d75586ad | 37 | int flags; |
c31c7d48 | 38 | unsigned long pfn; |
c9caa02c | 39 | unsigned force_split : 1; |
d75586ad | 40 | int curpage; |
9ae28475 | 41 | struct page **pages; |
72e458df TG |
42 | }; |
43 | ||
ad5ca55f SS |
44 | /* |
45 | * Serialize cpa() (for !DEBUG_PAGEALLOC which uses large identity mappings) | |
46 | * using cpa_lock. So that we don't allow any other cpu, with stale large tlb | |
47 | * entries change the page attribute in parallel to some other cpu | |
48 | * splitting a large page entry along with changing the attribute. | |
49 | */ | |
50 | static DEFINE_SPINLOCK(cpa_lock); | |
51 | ||
d75586ad SL |
52 | #define CPA_FLUSHTLB 1 |
53 | #define CPA_ARRAY 2 | |
9ae28475 | 54 | #define CPA_PAGES_ARRAY 4 |
d75586ad | 55 | |
65280e61 | 56 | #ifdef CONFIG_PROC_FS |
ce0c0e50 AK |
57 | static unsigned long direct_pages_count[PG_LEVEL_NUM]; |
58 | ||
65280e61 | 59 | void update_page_count(int level, unsigned long pages) |
ce0c0e50 | 60 | { |
ce0c0e50 | 61 | /* Protect against CPA */ |
a79e53d8 | 62 | spin_lock(&pgd_lock); |
ce0c0e50 | 63 | direct_pages_count[level] += pages; |
a79e53d8 | 64 | spin_unlock(&pgd_lock); |
65280e61 TG |
65 | } |
66 | ||
67 | static void split_page_count(int level) | |
68 | { | |
c9e0d391 DJ |
69 | if (direct_pages_count[level] == 0) |
70 | return; | |
71 | ||
65280e61 TG |
72 | direct_pages_count[level]--; |
73 | direct_pages_count[level - 1] += PTRS_PER_PTE; | |
74 | } | |
75 | ||
e1759c21 | 76 | void arch_report_meminfo(struct seq_file *m) |
65280e61 | 77 | { |
b9c3bfc2 | 78 | seq_printf(m, "DirectMap4k: %8lu kB\n", |
a06de630 HD |
79 | direct_pages_count[PG_LEVEL_4K] << 2); |
80 | #if defined(CONFIG_X86_64) || defined(CONFIG_X86_PAE) | |
b9c3bfc2 | 81 | seq_printf(m, "DirectMap2M: %8lu kB\n", |
a06de630 HD |
82 | direct_pages_count[PG_LEVEL_2M] << 11); |
83 | #else | |
b9c3bfc2 | 84 | seq_printf(m, "DirectMap4M: %8lu kB\n", |
a06de630 HD |
85 | direct_pages_count[PG_LEVEL_2M] << 12); |
86 | #endif | |
a06de630 | 87 | if (direct_gbpages) |
b9c3bfc2 | 88 | seq_printf(m, "DirectMap1G: %8lu kB\n", |
a06de630 | 89 | direct_pages_count[PG_LEVEL_1G] << 20); |
ce0c0e50 | 90 | } |
65280e61 TG |
91 | #else |
92 | static inline void split_page_count(int level) { } | |
93 | #endif | |
ce0c0e50 | 94 | |
c31c7d48 TG |
95 | #ifdef CONFIG_X86_64 |
96 | ||
97 | static inline unsigned long highmap_start_pfn(void) | |
98 | { | |
fc8d7826 | 99 | return __pa_symbol(_text) >> PAGE_SHIFT; |
c31c7d48 TG |
100 | } |
101 | ||
102 | static inline unsigned long highmap_end_pfn(void) | |
103 | { | |
fc8d7826 | 104 | return __pa_symbol(roundup(_brk_end, PMD_SIZE)) >> PAGE_SHIFT; |
c31c7d48 TG |
105 | } |
106 | ||
107 | #endif | |
108 | ||
92cb54a3 IM |
109 | #ifdef CONFIG_DEBUG_PAGEALLOC |
110 | # define debug_pagealloc 1 | |
111 | #else | |
112 | # define debug_pagealloc 0 | |
113 | #endif | |
114 | ||
ed724be6 AV |
115 | static inline int |
116 | within(unsigned long addr, unsigned long start, unsigned long end) | |
687c4825 | 117 | { |
ed724be6 AV |
118 | return addr >= start && addr < end; |
119 | } | |
120 | ||
d7c8f21a TG |
121 | /* |
122 | * Flushing functions | |
123 | */ | |
cd8ddf1a | 124 | |
cd8ddf1a TG |
125 | /** |
126 | * clflush_cache_range - flush a cache range with clflush | |
9efc31b8 | 127 | * @vaddr: virtual start address |
cd8ddf1a TG |
128 | * @size: number of bytes to flush |
129 | * | |
8b80fd8b RZ |
130 | * clflushopt is an unordered instruction which needs fencing with mfence or |
131 | * sfence to avoid ordering issues. | |
cd8ddf1a | 132 | */ |
4c61afcd | 133 | void clflush_cache_range(void *vaddr, unsigned int size) |
d7c8f21a | 134 | { |
1f1a89ac CW |
135 | const unsigned long clflush_size = boot_cpu_data.x86_clflush_size; |
136 | void *p = (void *)((unsigned long)vaddr & ~(clflush_size - 1)); | |
6c434d61 | 137 | void *vend = vaddr + size; |
1f1a89ac CW |
138 | |
139 | if (p >= vend) | |
140 | return; | |
d7c8f21a | 141 | |
cd8ddf1a | 142 | mb(); |
4c61afcd | 143 | |
1f1a89ac | 144 | for (; p < vend; p += clflush_size) |
6c434d61 | 145 | clflushopt(p); |
4c61afcd | 146 | |
cd8ddf1a | 147 | mb(); |
d7c8f21a | 148 | } |
e517a5e9 | 149 | EXPORT_SYMBOL_GPL(clflush_cache_range); |
d7c8f21a | 150 | |
af1e6844 | 151 | static void __cpa_flush_all(void *arg) |
d7c8f21a | 152 | { |
6bb8383b AK |
153 | unsigned long cache = (unsigned long)arg; |
154 | ||
d7c8f21a TG |
155 | /* |
156 | * Flush all to work around Errata in early athlons regarding | |
157 | * large page flushing. | |
158 | */ | |
159 | __flush_tlb_all(); | |
160 | ||
0b827537 | 161 | if (cache && boot_cpu_data.x86 >= 4) |
d7c8f21a TG |
162 | wbinvd(); |
163 | } | |
164 | ||
6bb8383b | 165 | static void cpa_flush_all(unsigned long cache) |
d7c8f21a TG |
166 | { |
167 | BUG_ON(irqs_disabled()); | |
168 | ||
15c8b6c1 | 169 | on_each_cpu(__cpa_flush_all, (void *) cache, 1); |
d7c8f21a TG |
170 | } |
171 | ||
57a6a46a TG |
172 | static void __cpa_flush_range(void *arg) |
173 | { | |
57a6a46a TG |
174 | /* |
175 | * We could optimize that further and do individual per page | |
176 | * tlb invalidates for a low number of pages. Caveat: we must | |
177 | * flush the high aliases on 64bit as well. | |
178 | */ | |
179 | __flush_tlb_all(); | |
57a6a46a TG |
180 | } |
181 | ||
6bb8383b | 182 | static void cpa_flush_range(unsigned long start, int numpages, int cache) |
57a6a46a | 183 | { |
4c61afcd IM |
184 | unsigned int i, level; |
185 | unsigned long addr; | |
186 | ||
57a6a46a | 187 | BUG_ON(irqs_disabled()); |
4c61afcd | 188 | WARN_ON(PAGE_ALIGN(start) != start); |
57a6a46a | 189 | |
15c8b6c1 | 190 | on_each_cpu(__cpa_flush_range, NULL, 1); |
57a6a46a | 191 | |
6bb8383b AK |
192 | if (!cache) |
193 | return; | |
194 | ||
3b233e52 TG |
195 | /* |
196 | * We only need to flush on one CPU, | |
197 | * clflush is a MESI-coherent instruction that | |
198 | * will cause all other CPUs to flush the same | |
199 | * cachelines: | |
200 | */ | |
4c61afcd IM |
201 | for (i = 0, addr = start; i < numpages; i++, addr += PAGE_SIZE) { |
202 | pte_t *pte = lookup_address(addr, &level); | |
203 | ||
204 | /* | |
205 | * Only flush present addresses: | |
206 | */ | |
7bfb72e8 | 207 | if (pte && (pte_val(*pte) & _PAGE_PRESENT)) |
4c61afcd IM |
208 | clflush_cache_range((void *) addr, PAGE_SIZE); |
209 | } | |
57a6a46a TG |
210 | } |
211 | ||
9ae28475 | 212 | static void cpa_flush_array(unsigned long *start, int numpages, int cache, |
213 | int in_flags, struct page **pages) | |
d75586ad SL |
214 | { |
215 | unsigned int i, level; | |
2171787b | 216 | unsigned long do_wbinvd = cache && numpages >= 1024; /* 4M threshold */ |
d75586ad SL |
217 | |
218 | BUG_ON(irqs_disabled()); | |
219 | ||
2171787b | 220 | on_each_cpu(__cpa_flush_all, (void *) do_wbinvd, 1); |
d75586ad | 221 | |
2171787b | 222 | if (!cache || do_wbinvd) |
d75586ad SL |
223 | return; |
224 | ||
d75586ad SL |
225 | /* |
226 | * We only need to flush on one CPU, | |
227 | * clflush is a MESI-coherent instruction that | |
228 | * will cause all other CPUs to flush the same | |
229 | * cachelines: | |
230 | */ | |
9ae28475 | 231 | for (i = 0; i < numpages; i++) { |
232 | unsigned long addr; | |
233 | pte_t *pte; | |
234 | ||
235 | if (in_flags & CPA_PAGES_ARRAY) | |
236 | addr = (unsigned long)page_address(pages[i]); | |
237 | else | |
238 | addr = start[i]; | |
239 | ||
240 | pte = lookup_address(addr, &level); | |
d75586ad SL |
241 | |
242 | /* | |
243 | * Only flush present addresses: | |
244 | */ | |
245 | if (pte && (pte_val(*pte) & _PAGE_PRESENT)) | |
9ae28475 | 246 | clflush_cache_range((void *)addr, PAGE_SIZE); |
d75586ad SL |
247 | } |
248 | } | |
249 | ||
ed724be6 AV |
250 | /* |
251 | * Certain areas of memory on x86 require very specific protection flags, | |
252 | * for example the BIOS area or kernel text. Callers don't always get this | |
253 | * right (again, ioremap() on BIOS memory is not uncommon) so this function | |
254 | * checks and fixes these known static required protection bits. | |
255 | */ | |
c31c7d48 TG |
256 | static inline pgprot_t static_protections(pgprot_t prot, unsigned long address, |
257 | unsigned long pfn) | |
ed724be6 AV |
258 | { |
259 | pgprot_t forbidden = __pgprot(0); | |
260 | ||
687c4825 | 261 | /* |
ed724be6 AV |
262 | * The BIOS area between 640k and 1Mb needs to be executable for |
263 | * PCI BIOS based config access (CONFIG_PCI_GOBIOS) support. | |
687c4825 | 264 | */ |
5bd5a452 MC |
265 | #ifdef CONFIG_PCI_BIOS |
266 | if (pcibios_enabled && within(pfn, BIOS_BEGIN >> PAGE_SHIFT, BIOS_END >> PAGE_SHIFT)) | |
ed724be6 | 267 | pgprot_val(forbidden) |= _PAGE_NX; |
5bd5a452 | 268 | #endif |
ed724be6 AV |
269 | |
270 | /* | |
271 | * The kernel text needs to be executable for obvious reasons | |
c31c7d48 TG |
272 | * Does not cover __inittext since that is gone later on. On |
273 | * 64bit we do not enforce !NX on the low mapping | |
ed724be6 AV |
274 | */ |
275 | if (within(address, (unsigned long)_text, (unsigned long)_etext)) | |
276 | pgprot_val(forbidden) |= _PAGE_NX; | |
cc0f21bb | 277 | |
cc0f21bb | 278 | /* |
c31c7d48 TG |
279 | * The .rodata section needs to be read-only. Using the pfn |
280 | * catches all aliases. | |
cc0f21bb | 281 | */ |
fc8d7826 AD |
282 | if (within(pfn, __pa_symbol(__start_rodata) >> PAGE_SHIFT, |
283 | __pa_symbol(__end_rodata) >> PAGE_SHIFT)) | |
cc0f21bb | 284 | pgprot_val(forbidden) |= _PAGE_RW; |
ed724be6 | 285 | |
55ca3cc1 | 286 | #if defined(CONFIG_X86_64) && defined(CONFIG_DEBUG_RODATA) |
74e08179 | 287 | /* |
502f6604 SS |
288 | * Once the kernel maps the text as RO (kernel_set_to_readonly is set), |
289 | * kernel text mappings for the large page aligned text, rodata sections | |
290 | * will be always read-only. For the kernel identity mappings covering | |
291 | * the holes caused by this alignment can be anything that user asks. | |
74e08179 SS |
292 | * |
293 | * This will preserve the large page mappings for kernel text/data | |
294 | * at no extra cost. | |
295 | */ | |
502f6604 SS |
296 | if (kernel_set_to_readonly && |
297 | within(address, (unsigned long)_text, | |
281ff33b SS |
298 | (unsigned long)__end_rodata_hpage_align)) { |
299 | unsigned int level; | |
300 | ||
301 | /* | |
302 | * Don't enforce the !RW mapping for the kernel text mapping, | |
303 | * if the current mapping is already using small page mapping. | |
304 | * No need to work hard to preserve large page mappings in this | |
305 | * case. | |
306 | * | |
307 | * This also fixes the Linux Xen paravirt guest boot failure | |
308 | * (because of unexpected read-only mappings for kernel identity | |
309 | * mappings). In this paravirt guest case, the kernel text | |
310 | * mapping and the kernel identity mapping share the same | |
311 | * page-table pages. Thus we can't really use different | |
312 | * protections for the kernel text and identity mappings. Also, | |
313 | * these shared mappings are made of small page mappings. | |
314 | * Thus this don't enforce !RW mapping for small page kernel | |
315 | * text mapping logic will help Linux Xen parvirt guest boot | |
0d2eb44f | 316 | * as well. |
281ff33b SS |
317 | */ |
318 | if (lookup_address(address, &level) && (level != PG_LEVEL_4K)) | |
319 | pgprot_val(forbidden) |= _PAGE_RW; | |
320 | } | |
74e08179 SS |
321 | #endif |
322 | ||
ed724be6 | 323 | prot = __pgprot(pgprot_val(prot) & ~pgprot_val(forbidden)); |
687c4825 IM |
324 | |
325 | return prot; | |
326 | } | |
327 | ||
426e34cc MF |
328 | /* |
329 | * Lookup the page table entry for a virtual address in a specific pgd. | |
330 | * Return a pointer to the entry and the level of the mapping. | |
331 | */ | |
332 | pte_t *lookup_address_in_pgd(pgd_t *pgd, unsigned long address, | |
333 | unsigned int *level) | |
9f4c815c | 334 | { |
1da177e4 LT |
335 | pud_t *pud; |
336 | pmd_t *pmd; | |
9f4c815c | 337 | |
30551bb3 TG |
338 | *level = PG_LEVEL_NONE; |
339 | ||
1da177e4 LT |
340 | if (pgd_none(*pgd)) |
341 | return NULL; | |
9df84993 | 342 | |
1da177e4 LT |
343 | pud = pud_offset(pgd, address); |
344 | if (pud_none(*pud)) | |
345 | return NULL; | |
c2f71ee2 AK |
346 | |
347 | *level = PG_LEVEL_1G; | |
348 | if (pud_large(*pud) || !pud_present(*pud)) | |
349 | return (pte_t *)pud; | |
350 | ||
1da177e4 LT |
351 | pmd = pmd_offset(pud, address); |
352 | if (pmd_none(*pmd)) | |
353 | return NULL; | |
30551bb3 TG |
354 | |
355 | *level = PG_LEVEL_2M; | |
9a14aefc | 356 | if (pmd_large(*pmd) || !pmd_present(*pmd)) |
1da177e4 | 357 | return (pte_t *)pmd; |
1da177e4 | 358 | |
30551bb3 | 359 | *level = PG_LEVEL_4K; |
9df84993 | 360 | |
9f4c815c IM |
361 | return pte_offset_kernel(pmd, address); |
362 | } | |
0fd64c23 BP |
363 | |
364 | /* | |
365 | * Lookup the page table entry for a virtual address. Return a pointer | |
366 | * to the entry and the level of the mapping. | |
367 | * | |
368 | * Note: We return pud and pmd either when the entry is marked large | |
369 | * or when the present bit is not set. Otherwise we would return a | |
370 | * pointer to a nonexisting mapping. | |
371 | */ | |
372 | pte_t *lookup_address(unsigned long address, unsigned int *level) | |
373 | { | |
426e34cc | 374 | return lookup_address_in_pgd(pgd_offset_k(address), address, level); |
0fd64c23 | 375 | } |
75bb8835 | 376 | EXPORT_SYMBOL_GPL(lookup_address); |
9f4c815c | 377 | |
0fd64c23 BP |
378 | static pte_t *_lookup_address_cpa(struct cpa_data *cpa, unsigned long address, |
379 | unsigned int *level) | |
380 | { | |
381 | if (cpa->pgd) | |
426e34cc | 382 | return lookup_address_in_pgd(cpa->pgd + pgd_index(address), |
0fd64c23 BP |
383 | address, level); |
384 | ||
385 | return lookup_address(address, level); | |
386 | } | |
387 | ||
792230c3 JG |
388 | /* |
389 | * Lookup the PMD entry for a virtual address. Return a pointer to the entry | |
390 | * or NULL if not present. | |
391 | */ | |
392 | pmd_t *lookup_pmd_address(unsigned long address) | |
393 | { | |
394 | pgd_t *pgd; | |
395 | pud_t *pud; | |
396 | ||
397 | pgd = pgd_offset_k(address); | |
398 | if (pgd_none(*pgd)) | |
399 | return NULL; | |
400 | ||
401 | pud = pud_offset(pgd, address); | |
402 | if (pud_none(*pud) || pud_large(*pud) || !pud_present(*pud)) | |
403 | return NULL; | |
404 | ||
405 | return pmd_offset(pud, address); | |
406 | } | |
407 | ||
d7656534 DH |
408 | /* |
409 | * This is necessary because __pa() does not work on some | |
410 | * kinds of memory, like vmalloc() or the alloc_remap() | |
411 | * areas on 32-bit NUMA systems. The percpu areas can | |
412 | * end up in this kind of memory, for instance. | |
413 | * | |
414 | * This could be optimized, but it is only intended to be | |
415 | * used at inititalization time, and keeping it | |
416 | * unoptimized should increase the testing coverage for | |
417 | * the more obscure platforms. | |
418 | */ | |
419 | phys_addr_t slow_virt_to_phys(void *__virt_addr) | |
420 | { | |
421 | unsigned long virt_addr = (unsigned long)__virt_addr; | |
34437e67 | 422 | unsigned long phys_addr, offset; |
d7656534 | 423 | enum pg_level level; |
d7656534 DH |
424 | pte_t *pte; |
425 | ||
426 | pte = lookup_address(virt_addr, &level); | |
427 | BUG_ON(!pte); | |
34437e67 TK |
428 | |
429 | switch (level) { | |
430 | case PG_LEVEL_1G: | |
431 | phys_addr = pud_pfn(*(pud_t *)pte) << PAGE_SHIFT; | |
432 | offset = virt_addr & ~PUD_PAGE_MASK; | |
433 | break; | |
434 | case PG_LEVEL_2M: | |
435 | phys_addr = pmd_pfn(*(pmd_t *)pte) << PAGE_SHIFT; | |
436 | offset = virt_addr & ~PMD_PAGE_MASK; | |
437 | break; | |
438 | default: | |
439 | phys_addr = pte_pfn(*pte) << PAGE_SHIFT; | |
440 | offset = virt_addr & ~PAGE_MASK; | |
441 | } | |
442 | ||
443 | return (phys_addr_t)(phys_addr | offset); | |
d7656534 DH |
444 | } |
445 | EXPORT_SYMBOL_GPL(slow_virt_to_phys); | |
446 | ||
9df84993 IM |
447 | /* |
448 | * Set the new pmd in all the pgds we know about: | |
449 | */ | |
9a3dc780 | 450 | static void __set_pmd_pte(pte_t *kpte, unsigned long address, pte_t pte) |
9f4c815c | 451 | { |
9f4c815c IM |
452 | /* change init_mm */ |
453 | set_pte_atomic(kpte, pte); | |
44af6c41 | 454 | #ifdef CONFIG_X86_32 |
e4b71dcf | 455 | if (!SHARED_KERNEL_PMD) { |
44af6c41 IM |
456 | struct page *page; |
457 | ||
e3ed910d | 458 | list_for_each_entry(page, &pgd_list, lru) { |
44af6c41 IM |
459 | pgd_t *pgd; |
460 | pud_t *pud; | |
461 | pmd_t *pmd; | |
462 | ||
463 | pgd = (pgd_t *)page_address(page) + pgd_index(address); | |
464 | pud = pud_offset(pgd, address); | |
465 | pmd = pmd_offset(pud, address); | |
466 | set_pte_atomic((pte_t *)pmd, pte); | |
467 | } | |
1da177e4 | 468 | } |
44af6c41 | 469 | #endif |
1da177e4 LT |
470 | } |
471 | ||
9df84993 IM |
472 | static int |
473 | try_preserve_large_page(pte_t *kpte, unsigned long address, | |
474 | struct cpa_data *cpa) | |
65e074df | 475 | { |
3a19109e | 476 | unsigned long nextpage_addr, numpages, pmask, psize, addr, pfn, old_pfn; |
65e074df | 477 | pte_t new_pte, old_pte, *tmp; |
64edc8ed | 478 | pgprot_t old_prot, new_prot, req_prot; |
fac84939 | 479 | int i, do_split = 1; |
f3c4fbb6 | 480 | enum pg_level level; |
65e074df | 481 | |
c9caa02c AK |
482 | if (cpa->force_split) |
483 | return 1; | |
484 | ||
a79e53d8 | 485 | spin_lock(&pgd_lock); |
65e074df TG |
486 | /* |
487 | * Check for races, another CPU might have split this page | |
488 | * up already: | |
489 | */ | |
82f0712c | 490 | tmp = _lookup_address_cpa(cpa, address, &level); |
65e074df TG |
491 | if (tmp != kpte) |
492 | goto out_unlock; | |
493 | ||
494 | switch (level) { | |
495 | case PG_LEVEL_2M: | |
3a19109e TK |
496 | old_prot = pmd_pgprot(*(pmd_t *)kpte); |
497 | old_pfn = pmd_pfn(*(pmd_t *)kpte); | |
498 | break; | |
65e074df | 499 | case PG_LEVEL_1G: |
3a19109e TK |
500 | old_prot = pud_pgprot(*(pud_t *)kpte); |
501 | old_pfn = pud_pfn(*(pud_t *)kpte); | |
f3c4fbb6 | 502 | break; |
65e074df | 503 | default: |
beaff633 | 504 | do_split = -EINVAL; |
65e074df TG |
505 | goto out_unlock; |
506 | } | |
507 | ||
3a19109e TK |
508 | psize = page_level_size(level); |
509 | pmask = page_level_mask(level); | |
510 | ||
65e074df TG |
511 | /* |
512 | * Calculate the number of pages, which fit into this large | |
513 | * page starting at address: | |
514 | */ | |
515 | nextpage_addr = (address + psize) & pmask; | |
516 | numpages = (nextpage_addr - address) >> PAGE_SHIFT; | |
9b5cf48b RW |
517 | if (numpages < cpa->numpages) |
518 | cpa->numpages = numpages; | |
65e074df TG |
519 | |
520 | /* | |
521 | * We are safe now. Check whether the new pgprot is the same: | |
f5b2831d JG |
522 | * Convert protection attributes to 4k-format, as cpa->mask* are set |
523 | * up accordingly. | |
65e074df TG |
524 | */ |
525 | old_pte = *kpte; | |
55696b1f | 526 | req_prot = pgprot_large_2_4k(old_prot); |
65e074df | 527 | |
64edc8ed MC |
528 | pgprot_val(req_prot) &= ~pgprot_val(cpa->mask_clr); |
529 | pgprot_val(req_prot) |= pgprot_val(cpa->mask_set); | |
c31c7d48 | 530 | |
f5b2831d JG |
531 | /* |
532 | * req_prot is in format of 4k pages. It must be converted to large | |
533 | * page format: the caching mode includes the PAT bit located at | |
534 | * different bit positions in the two formats. | |
535 | */ | |
536 | req_prot = pgprot_4k_2_large(req_prot); | |
537 | ||
a8aed3e0 AA |
538 | /* |
539 | * Set the PSE and GLOBAL flags only if the PRESENT flag is | |
540 | * set otherwise pmd_present/pmd_huge will return true even on | |
541 | * a non present pmd. The canon_pgprot will clear _PAGE_GLOBAL | |
542 | * for the ancient hardware that doesn't support it. | |
543 | */ | |
f76cfa3c AA |
544 | if (pgprot_val(req_prot) & _PAGE_PRESENT) |
545 | pgprot_val(req_prot) |= _PAGE_PSE | _PAGE_GLOBAL; | |
a8aed3e0 | 546 | else |
f76cfa3c | 547 | pgprot_val(req_prot) &= ~(_PAGE_PSE | _PAGE_GLOBAL); |
a8aed3e0 | 548 | |
f76cfa3c | 549 | req_prot = canon_pgprot(req_prot); |
a8aed3e0 | 550 | |
c31c7d48 | 551 | /* |
3a19109e | 552 | * old_pfn points to the large page base pfn. So we need |
c31c7d48 TG |
553 | * to add the offset of the virtual address: |
554 | */ | |
3a19109e | 555 | pfn = old_pfn + ((address & (psize - 1)) >> PAGE_SHIFT); |
c31c7d48 TG |
556 | cpa->pfn = pfn; |
557 | ||
64edc8ed | 558 | new_prot = static_protections(req_prot, address, pfn); |
65e074df | 559 | |
fac84939 TG |
560 | /* |
561 | * We need to check the full range, whether | |
562 | * static_protection() requires a different pgprot for one of | |
563 | * the pages in the range we try to preserve: | |
564 | */ | |
64edc8ed | 565 | addr = address & pmask; |
3a19109e | 566 | pfn = old_pfn; |
64edc8ed MC |
567 | for (i = 0; i < (psize >> PAGE_SHIFT); i++, addr += PAGE_SIZE, pfn++) { |
568 | pgprot_t chk_prot = static_protections(req_prot, addr, pfn); | |
fac84939 TG |
569 | |
570 | if (pgprot_val(chk_prot) != pgprot_val(new_prot)) | |
571 | goto out_unlock; | |
572 | } | |
573 | ||
65e074df TG |
574 | /* |
575 | * If there are no changes, return. maxpages has been updated | |
576 | * above: | |
577 | */ | |
578 | if (pgprot_val(new_prot) == pgprot_val(old_prot)) { | |
beaff633 | 579 | do_split = 0; |
65e074df TG |
580 | goto out_unlock; |
581 | } | |
582 | ||
583 | /* | |
584 | * We need to change the attributes. Check, whether we can | |
585 | * change the large page in one go. We request a split, when | |
586 | * the address is not aligned and the number of pages is | |
587 | * smaller than the number of pages in the large page. Note | |
588 | * that we limited the number of possible pages already to | |
589 | * the number of pages in the large page. | |
590 | */ | |
64edc8ed | 591 | if (address == (address & pmask) && cpa->numpages == (psize >> PAGE_SHIFT)) { |
65e074df TG |
592 | /* |
593 | * The address is aligned and the number of pages | |
594 | * covers the full page. | |
595 | */ | |
3a19109e | 596 | new_pte = pfn_pte(old_pfn, new_prot); |
65e074df | 597 | __set_pmd_pte(kpte, address, new_pte); |
d75586ad | 598 | cpa->flags |= CPA_FLUSHTLB; |
beaff633 | 599 | do_split = 0; |
65e074df TG |
600 | } |
601 | ||
602 | out_unlock: | |
a79e53d8 | 603 | spin_unlock(&pgd_lock); |
9df84993 | 604 | |
beaff633 | 605 | return do_split; |
65e074df TG |
606 | } |
607 | ||
5952886b | 608 | static int |
82f0712c BP |
609 | __split_large_page(struct cpa_data *cpa, pte_t *kpte, unsigned long address, |
610 | struct page *base) | |
bb5c2dbd | 611 | { |
5952886b | 612 | pte_t *pbase = (pte_t *)page_address(base); |
d551aaa2 | 613 | unsigned long ref_pfn, pfn, pfninc = 1; |
9df84993 | 614 | unsigned int i, level; |
ae9aae9e | 615 | pte_t *tmp; |
9df84993 | 616 | pgprot_t ref_prot; |
bb5c2dbd | 617 | |
a79e53d8 | 618 | spin_lock(&pgd_lock); |
bb5c2dbd IM |
619 | /* |
620 | * Check for races, another CPU might have split this page | |
621 | * up for us already: | |
622 | */ | |
82f0712c | 623 | tmp = _lookup_address_cpa(cpa, address, &level); |
ae9aae9e WC |
624 | if (tmp != kpte) { |
625 | spin_unlock(&pgd_lock); | |
626 | return 1; | |
627 | } | |
bb5c2dbd | 628 | |
6944a9c8 | 629 | paravirt_alloc_pte(&init_mm, page_to_pfn(base)); |
f5b2831d | 630 | |
d551aaa2 TK |
631 | switch (level) { |
632 | case PG_LEVEL_2M: | |
633 | ref_prot = pmd_pgprot(*(pmd_t *)kpte); | |
634 | /* clear PSE and promote PAT bit to correct position */ | |
f5b2831d | 635 | ref_prot = pgprot_large_2_4k(ref_prot); |
d551aaa2 TK |
636 | ref_pfn = pmd_pfn(*(pmd_t *)kpte); |
637 | break; | |
bb5c2dbd | 638 | |
d551aaa2 TK |
639 | case PG_LEVEL_1G: |
640 | ref_prot = pud_pgprot(*(pud_t *)kpte); | |
641 | ref_pfn = pud_pfn(*(pud_t *)kpte); | |
f07333fd | 642 | pfninc = PMD_PAGE_SIZE >> PAGE_SHIFT; |
d551aaa2 | 643 | |
a8aed3e0 | 644 | /* |
d551aaa2 | 645 | * Clear the PSE flags if the PRESENT flag is not set |
a8aed3e0 AA |
646 | * otherwise pmd_present/pmd_huge will return true |
647 | * even on a non present pmd. | |
648 | */ | |
d551aaa2 | 649 | if (!(pgprot_val(ref_prot) & _PAGE_PRESENT)) |
a8aed3e0 | 650 | pgprot_val(ref_prot) &= ~_PAGE_PSE; |
d551aaa2 TK |
651 | break; |
652 | ||
653 | default: | |
654 | spin_unlock(&pgd_lock); | |
655 | return 1; | |
f07333fd | 656 | } |
f07333fd | 657 | |
a8aed3e0 AA |
658 | /* |
659 | * Set the GLOBAL flags only if the PRESENT flag is set | |
660 | * otherwise pmd/pte_present will return true even on a non | |
661 | * present pmd/pte. The canon_pgprot will clear _PAGE_GLOBAL | |
662 | * for the ancient hardware that doesn't support it. | |
663 | */ | |
664 | if (pgprot_val(ref_prot) & _PAGE_PRESENT) | |
665 | pgprot_val(ref_prot) |= _PAGE_GLOBAL; | |
666 | else | |
667 | pgprot_val(ref_prot) &= ~_PAGE_GLOBAL; | |
668 | ||
63c1dcf4 TG |
669 | /* |
670 | * Get the target pfn from the original entry: | |
671 | */ | |
d551aaa2 | 672 | pfn = ref_pfn; |
f07333fd | 673 | for (i = 0; i < PTRS_PER_PTE; i++, pfn += pfninc) |
a8aed3e0 | 674 | set_pte(&pbase[i], pfn_pte(pfn, canon_pgprot(ref_prot))); |
bb5c2dbd | 675 | |
2c66e24d SP |
676 | if (virt_addr_valid(address)) { |
677 | unsigned long pfn = PFN_DOWN(__pa(address)); | |
678 | ||
679 | if (pfn_range_is_mapped(pfn, pfn + 1)) | |
680 | split_page_count(level); | |
681 | } | |
f361a450 | 682 | |
bb5c2dbd | 683 | /* |
07a66d7c | 684 | * Install the new, split up pagetable. |
4c881ca1 | 685 | * |
07a66d7c IM |
686 | * We use the standard kernel pagetable protections for the new |
687 | * pagetable protections, the actual ptes set above control the | |
688 | * primary protection behavior: | |
bb5c2dbd | 689 | */ |
07a66d7c | 690 | __set_pmd_pte(kpte, address, mk_pte(base, __pgprot(_KERNPG_TABLE))); |
211b3d03 IM |
691 | |
692 | /* | |
693 | * Intel Atom errata AAH41 workaround. | |
694 | * | |
695 | * The real fix should be in hw or in a microcode update, but | |
696 | * we also probabilistically try to reduce the window of having | |
697 | * a large TLB mixed with 4K TLBs while instruction fetches are | |
698 | * going on. | |
699 | */ | |
700 | __flush_tlb_all(); | |
ae9aae9e | 701 | spin_unlock(&pgd_lock); |
211b3d03 | 702 | |
ae9aae9e WC |
703 | return 0; |
704 | } | |
bb5c2dbd | 705 | |
82f0712c BP |
706 | static int split_large_page(struct cpa_data *cpa, pte_t *kpte, |
707 | unsigned long address) | |
ae9aae9e | 708 | { |
ae9aae9e WC |
709 | struct page *base; |
710 | ||
711 | if (!debug_pagealloc) | |
712 | spin_unlock(&cpa_lock); | |
713 | base = alloc_pages(GFP_KERNEL | __GFP_NOTRACK, 0); | |
714 | if (!debug_pagealloc) | |
715 | spin_lock(&cpa_lock); | |
716 | if (!base) | |
717 | return -ENOMEM; | |
718 | ||
82f0712c | 719 | if (__split_large_page(cpa, kpte, address, base)) |
8311eb84 | 720 | __free_page(base); |
bb5c2dbd | 721 | |
bb5c2dbd IM |
722 | return 0; |
723 | } | |
724 | ||
52a628fb BP |
725 | static bool try_to_free_pte_page(pte_t *pte) |
726 | { | |
727 | int i; | |
728 | ||
729 | for (i = 0; i < PTRS_PER_PTE; i++) | |
730 | if (!pte_none(pte[i])) | |
731 | return false; | |
732 | ||
733 | free_page((unsigned long)pte); | |
734 | return true; | |
735 | } | |
736 | ||
737 | static bool try_to_free_pmd_page(pmd_t *pmd) | |
738 | { | |
739 | int i; | |
740 | ||
741 | for (i = 0; i < PTRS_PER_PMD; i++) | |
742 | if (!pmd_none(pmd[i])) | |
743 | return false; | |
744 | ||
745 | free_page((unsigned long)pmd); | |
746 | return true; | |
747 | } | |
748 | ||
42a54772 BP |
749 | static bool try_to_free_pud_page(pud_t *pud) |
750 | { | |
751 | int i; | |
752 | ||
753 | for (i = 0; i < PTRS_PER_PUD; i++) | |
754 | if (!pud_none(pud[i])) | |
755 | return false; | |
756 | ||
757 | free_page((unsigned long)pud); | |
758 | return true; | |
759 | } | |
760 | ||
52a628fb BP |
761 | static bool unmap_pte_range(pmd_t *pmd, unsigned long start, unsigned long end) |
762 | { | |
763 | pte_t *pte = pte_offset_kernel(pmd, start); | |
764 | ||
765 | while (start < end) { | |
766 | set_pte(pte, __pte(0)); | |
767 | ||
768 | start += PAGE_SIZE; | |
769 | pte++; | |
770 | } | |
771 | ||
772 | if (try_to_free_pte_page((pte_t *)pmd_page_vaddr(*pmd))) { | |
773 | pmd_clear(pmd); | |
774 | return true; | |
775 | } | |
776 | return false; | |
777 | } | |
778 | ||
779 | static void __unmap_pmd_range(pud_t *pud, pmd_t *pmd, | |
780 | unsigned long start, unsigned long end) | |
781 | { | |
782 | if (unmap_pte_range(pmd, start, end)) | |
783 | if (try_to_free_pmd_page((pmd_t *)pud_page_vaddr(*pud))) | |
784 | pud_clear(pud); | |
785 | } | |
786 | ||
787 | static void unmap_pmd_range(pud_t *pud, unsigned long start, unsigned long end) | |
788 | { | |
789 | pmd_t *pmd = pmd_offset(pud, start); | |
790 | ||
791 | /* | |
792 | * Not on a 2MB page boundary? | |
793 | */ | |
794 | if (start & (PMD_SIZE - 1)) { | |
795 | unsigned long next_page = (start + PMD_SIZE) & PMD_MASK; | |
796 | unsigned long pre_end = min_t(unsigned long, end, next_page); | |
797 | ||
798 | __unmap_pmd_range(pud, pmd, start, pre_end); | |
799 | ||
800 | start = pre_end; | |
801 | pmd++; | |
802 | } | |
803 | ||
804 | /* | |
805 | * Try to unmap in 2M chunks. | |
806 | */ | |
807 | while (end - start >= PMD_SIZE) { | |
808 | if (pmd_large(*pmd)) | |
809 | pmd_clear(pmd); | |
810 | else | |
811 | __unmap_pmd_range(pud, pmd, start, start + PMD_SIZE); | |
812 | ||
813 | start += PMD_SIZE; | |
814 | pmd++; | |
815 | } | |
816 | ||
817 | /* | |
818 | * 4K leftovers? | |
819 | */ | |
820 | if (start < end) | |
821 | return __unmap_pmd_range(pud, pmd, start, end); | |
822 | ||
823 | /* | |
824 | * Try again to free the PMD page if haven't succeeded above. | |
825 | */ | |
826 | if (!pud_none(*pud)) | |
827 | if (try_to_free_pmd_page((pmd_t *)pud_page_vaddr(*pud))) | |
828 | pud_clear(pud); | |
829 | } | |
0bb8aeee BP |
830 | |
831 | static void unmap_pud_range(pgd_t *pgd, unsigned long start, unsigned long end) | |
832 | { | |
833 | pud_t *pud = pud_offset(pgd, start); | |
834 | ||
835 | /* | |
836 | * Not on a GB page boundary? | |
837 | */ | |
838 | if (start & (PUD_SIZE - 1)) { | |
839 | unsigned long next_page = (start + PUD_SIZE) & PUD_MASK; | |
840 | unsigned long pre_end = min_t(unsigned long, end, next_page); | |
841 | ||
842 | unmap_pmd_range(pud, start, pre_end); | |
843 | ||
844 | start = pre_end; | |
845 | pud++; | |
846 | } | |
847 | ||
848 | /* | |
849 | * Try to unmap in 1G chunks? | |
850 | */ | |
851 | while (end - start >= PUD_SIZE) { | |
852 | ||
853 | if (pud_large(*pud)) | |
854 | pud_clear(pud); | |
855 | else | |
856 | unmap_pmd_range(pud, start, start + PUD_SIZE); | |
857 | ||
858 | start += PUD_SIZE; | |
859 | pud++; | |
860 | } | |
861 | ||
862 | /* | |
863 | * 2M leftovers? | |
864 | */ | |
865 | if (start < end) | |
866 | unmap_pmd_range(pud, start, end); | |
867 | ||
868 | /* | |
869 | * No need to try to free the PUD page because we'll free it in | |
870 | * populate_pgd's error path | |
871 | */ | |
872 | } | |
873 | ||
42a54772 BP |
874 | static void unmap_pgd_range(pgd_t *root, unsigned long addr, unsigned long end) |
875 | { | |
876 | pgd_t *pgd_entry = root + pgd_index(addr); | |
877 | ||
878 | unmap_pud_range(pgd_entry, addr, end); | |
879 | ||
880 | if (try_to_free_pud_page((pud_t *)pgd_page_vaddr(*pgd_entry))) | |
881 | pgd_clear(pgd_entry); | |
882 | } | |
883 | ||
f900a4b8 BP |
884 | static int alloc_pte_page(pmd_t *pmd) |
885 | { | |
886 | pte_t *pte = (pte_t *)get_zeroed_page(GFP_KERNEL | __GFP_NOTRACK); | |
887 | if (!pte) | |
888 | return -1; | |
889 | ||
890 | set_pmd(pmd, __pmd(__pa(pte) | _KERNPG_TABLE)); | |
891 | return 0; | |
892 | } | |
893 | ||
4b23538d BP |
894 | static int alloc_pmd_page(pud_t *pud) |
895 | { | |
896 | pmd_t *pmd = (pmd_t *)get_zeroed_page(GFP_KERNEL | __GFP_NOTRACK); | |
897 | if (!pmd) | |
898 | return -1; | |
899 | ||
900 | set_pud(pud, __pud(__pa(pmd) | _KERNPG_TABLE)); | |
901 | return 0; | |
902 | } | |
903 | ||
c6b6f363 BP |
904 | static void populate_pte(struct cpa_data *cpa, |
905 | unsigned long start, unsigned long end, | |
906 | unsigned num_pages, pmd_t *pmd, pgprot_t pgprot) | |
907 | { | |
908 | pte_t *pte; | |
909 | ||
910 | pte = pte_offset_kernel(pmd, start); | |
911 | ||
912 | while (num_pages-- && start < end) { | |
913 | ||
914 | /* deal with the NX bit */ | |
915 | if (!(pgprot_val(pgprot) & _PAGE_NX)) | |
916 | cpa->pfn &= ~_PAGE_NX; | |
917 | ||
918 | set_pte(pte, pfn_pte(cpa->pfn >> PAGE_SHIFT, pgprot)); | |
919 | ||
920 | start += PAGE_SIZE; | |
921 | cpa->pfn += PAGE_SIZE; | |
922 | pte++; | |
923 | } | |
924 | } | |
f900a4b8 BP |
925 | |
926 | static int populate_pmd(struct cpa_data *cpa, | |
927 | unsigned long start, unsigned long end, | |
928 | unsigned num_pages, pud_t *pud, pgprot_t pgprot) | |
929 | { | |
930 | unsigned int cur_pages = 0; | |
931 | pmd_t *pmd; | |
f5b2831d | 932 | pgprot_t pmd_pgprot; |
f900a4b8 BP |
933 | |
934 | /* | |
935 | * Not on a 2M boundary? | |
936 | */ | |
937 | if (start & (PMD_SIZE - 1)) { | |
938 | unsigned long pre_end = start + (num_pages << PAGE_SHIFT); | |
939 | unsigned long next_page = (start + PMD_SIZE) & PMD_MASK; | |
940 | ||
941 | pre_end = min_t(unsigned long, pre_end, next_page); | |
942 | cur_pages = (pre_end - start) >> PAGE_SHIFT; | |
943 | cur_pages = min_t(unsigned int, num_pages, cur_pages); | |
944 | ||
945 | /* | |
946 | * Need a PTE page? | |
947 | */ | |
948 | pmd = pmd_offset(pud, start); | |
949 | if (pmd_none(*pmd)) | |
950 | if (alloc_pte_page(pmd)) | |
951 | return -1; | |
952 | ||
953 | populate_pte(cpa, start, pre_end, cur_pages, pmd, pgprot); | |
954 | ||
955 | start = pre_end; | |
956 | } | |
957 | ||
958 | /* | |
959 | * We mapped them all? | |
960 | */ | |
961 | if (num_pages == cur_pages) | |
962 | return cur_pages; | |
963 | ||
f5b2831d JG |
964 | pmd_pgprot = pgprot_4k_2_large(pgprot); |
965 | ||
f900a4b8 BP |
966 | while (end - start >= PMD_SIZE) { |
967 | ||
968 | /* | |
969 | * We cannot use a 1G page so allocate a PMD page if needed. | |
970 | */ | |
971 | if (pud_none(*pud)) | |
972 | if (alloc_pmd_page(pud)) | |
973 | return -1; | |
974 | ||
975 | pmd = pmd_offset(pud, start); | |
976 | ||
f5b2831d JG |
977 | set_pmd(pmd, __pmd(cpa->pfn | _PAGE_PSE | |
978 | massage_pgprot(pmd_pgprot))); | |
f900a4b8 BP |
979 | |
980 | start += PMD_SIZE; | |
981 | cpa->pfn += PMD_SIZE; | |
982 | cur_pages += PMD_SIZE >> PAGE_SHIFT; | |
983 | } | |
984 | ||
985 | /* | |
986 | * Map trailing 4K pages. | |
987 | */ | |
988 | if (start < end) { | |
989 | pmd = pmd_offset(pud, start); | |
990 | if (pmd_none(*pmd)) | |
991 | if (alloc_pte_page(pmd)) | |
992 | return -1; | |
993 | ||
994 | populate_pte(cpa, start, end, num_pages - cur_pages, | |
995 | pmd, pgprot); | |
996 | } | |
997 | return num_pages; | |
998 | } | |
4b23538d BP |
999 | |
1000 | static int populate_pud(struct cpa_data *cpa, unsigned long start, pgd_t *pgd, | |
1001 | pgprot_t pgprot) | |
1002 | { | |
1003 | pud_t *pud; | |
1004 | unsigned long end; | |
1005 | int cur_pages = 0; | |
f5b2831d | 1006 | pgprot_t pud_pgprot; |
4b23538d BP |
1007 | |
1008 | end = start + (cpa->numpages << PAGE_SHIFT); | |
1009 | ||
1010 | /* | |
1011 | * Not on a Gb page boundary? => map everything up to it with | |
1012 | * smaller pages. | |
1013 | */ | |
1014 | if (start & (PUD_SIZE - 1)) { | |
1015 | unsigned long pre_end; | |
1016 | unsigned long next_page = (start + PUD_SIZE) & PUD_MASK; | |
1017 | ||
1018 | pre_end = min_t(unsigned long, end, next_page); | |
1019 | cur_pages = (pre_end - start) >> PAGE_SHIFT; | |
1020 | cur_pages = min_t(int, (int)cpa->numpages, cur_pages); | |
1021 | ||
1022 | pud = pud_offset(pgd, start); | |
1023 | ||
1024 | /* | |
1025 | * Need a PMD page? | |
1026 | */ | |
1027 | if (pud_none(*pud)) | |
1028 | if (alloc_pmd_page(pud)) | |
1029 | return -1; | |
1030 | ||
1031 | cur_pages = populate_pmd(cpa, start, pre_end, cur_pages, | |
1032 | pud, pgprot); | |
1033 | if (cur_pages < 0) | |
1034 | return cur_pages; | |
1035 | ||
1036 | start = pre_end; | |
1037 | } | |
1038 | ||
1039 | /* We mapped them all? */ | |
1040 | if (cpa->numpages == cur_pages) | |
1041 | return cur_pages; | |
1042 | ||
1043 | pud = pud_offset(pgd, start); | |
f5b2831d | 1044 | pud_pgprot = pgprot_4k_2_large(pgprot); |
4b23538d BP |
1045 | |
1046 | /* | |
1047 | * Map everything starting from the Gb boundary, possibly with 1G pages | |
1048 | */ | |
1049 | while (end - start >= PUD_SIZE) { | |
f5b2831d JG |
1050 | set_pud(pud, __pud(cpa->pfn | _PAGE_PSE | |
1051 | massage_pgprot(pud_pgprot))); | |
4b23538d BP |
1052 | |
1053 | start += PUD_SIZE; | |
1054 | cpa->pfn += PUD_SIZE; | |
1055 | cur_pages += PUD_SIZE >> PAGE_SHIFT; | |
1056 | pud++; | |
1057 | } | |
1058 | ||
1059 | /* Map trailing leftover */ | |
1060 | if (start < end) { | |
1061 | int tmp; | |
1062 | ||
1063 | pud = pud_offset(pgd, start); | |
1064 | if (pud_none(*pud)) | |
1065 | if (alloc_pmd_page(pud)) | |
1066 | return -1; | |
1067 | ||
1068 | tmp = populate_pmd(cpa, start, end, cpa->numpages - cur_pages, | |
1069 | pud, pgprot); | |
1070 | if (tmp < 0) | |
1071 | return cur_pages; | |
1072 | ||
1073 | cur_pages += tmp; | |
1074 | } | |
1075 | return cur_pages; | |
1076 | } | |
f3f72966 BP |
1077 | |
1078 | /* | |
1079 | * Restrictions for kernel page table do not necessarily apply when mapping in | |
1080 | * an alternate PGD. | |
1081 | */ | |
1082 | static int populate_pgd(struct cpa_data *cpa, unsigned long addr) | |
1083 | { | |
1084 | pgprot_t pgprot = __pgprot(_KERNPG_TABLE); | |
f3f72966 | 1085 | pud_t *pud = NULL; /* shut up gcc */ |
42a54772 | 1086 | pgd_t *pgd_entry; |
f3f72966 BP |
1087 | int ret; |
1088 | ||
1089 | pgd_entry = cpa->pgd + pgd_index(addr); | |
1090 | ||
1091 | /* | |
1092 | * Allocate a PUD page and hand it down for mapping. | |
1093 | */ | |
1094 | if (pgd_none(*pgd_entry)) { | |
1095 | pud = (pud_t *)get_zeroed_page(GFP_KERNEL | __GFP_NOTRACK); | |
1096 | if (!pud) | |
1097 | return -1; | |
1098 | ||
1099 | set_pgd(pgd_entry, __pgd(__pa(pud) | _KERNPG_TABLE)); | |
f3f72966 BP |
1100 | } |
1101 | ||
1102 | pgprot_val(pgprot) &= ~pgprot_val(cpa->mask_clr); | |
1103 | pgprot_val(pgprot) |= pgprot_val(cpa->mask_set); | |
1104 | ||
1105 | ret = populate_pud(cpa, addr, pgd_entry, pgprot); | |
0bb8aeee | 1106 | if (ret < 0) { |
42a54772 | 1107 | unmap_pgd_range(cpa->pgd, addr, |
0bb8aeee | 1108 | addr + (cpa->numpages << PAGE_SHIFT)); |
f3f72966 | 1109 | return ret; |
0bb8aeee | 1110 | } |
42a54772 | 1111 | |
f3f72966 BP |
1112 | cpa->numpages = ret; |
1113 | return 0; | |
1114 | } | |
1115 | ||
a1e46212 SS |
1116 | static int __cpa_process_fault(struct cpa_data *cpa, unsigned long vaddr, |
1117 | int primary) | |
1118 | { | |
82f0712c BP |
1119 | if (cpa->pgd) |
1120 | return populate_pgd(cpa, vaddr); | |
1121 | ||
a1e46212 SS |
1122 | /* |
1123 | * Ignore all non primary paths. | |
1124 | */ | |
1125 | if (!primary) | |
1126 | return 0; | |
1127 | ||
1128 | /* | |
1129 | * Ignore the NULL PTE for kernel identity mapping, as it is expected | |
1130 | * to have holes. | |
1131 | * Also set numpages to '1' indicating that we processed cpa req for | |
1132 | * one virtual address page and its pfn. TBD: numpages can be set based | |
1133 | * on the initial value and the level returned by lookup_address(). | |
1134 | */ | |
1135 | if (within(vaddr, PAGE_OFFSET, | |
1136 | PAGE_OFFSET + (max_pfn_mapped << PAGE_SHIFT))) { | |
1137 | cpa->numpages = 1; | |
1138 | cpa->pfn = __pa(vaddr) >> PAGE_SHIFT; | |
1139 | return 0; | |
1140 | } else { | |
1141 | WARN(1, KERN_WARNING "CPA: called for zero pte. " | |
1142 | "vaddr = %lx cpa->vaddr = %lx\n", vaddr, | |
1143 | *cpa->vaddr); | |
1144 | ||
1145 | return -EFAULT; | |
1146 | } | |
1147 | } | |
1148 | ||
c31c7d48 | 1149 | static int __change_page_attr(struct cpa_data *cpa, int primary) |
9f4c815c | 1150 | { |
d75586ad | 1151 | unsigned long address; |
da7bfc50 HH |
1152 | int do_split, err; |
1153 | unsigned int level; | |
c31c7d48 | 1154 | pte_t *kpte, old_pte; |
1da177e4 | 1155 | |
8523acfe TH |
1156 | if (cpa->flags & CPA_PAGES_ARRAY) { |
1157 | struct page *page = cpa->pages[cpa->curpage]; | |
1158 | if (unlikely(PageHighMem(page))) | |
1159 | return 0; | |
1160 | address = (unsigned long)page_address(page); | |
1161 | } else if (cpa->flags & CPA_ARRAY) | |
d75586ad SL |
1162 | address = cpa->vaddr[cpa->curpage]; |
1163 | else | |
1164 | address = *cpa->vaddr; | |
97f99fed | 1165 | repeat: |
82f0712c | 1166 | kpte = _lookup_address_cpa(cpa, address, &level); |
1da177e4 | 1167 | if (!kpte) |
a1e46212 | 1168 | return __cpa_process_fault(cpa, address, primary); |
c31c7d48 TG |
1169 | |
1170 | old_pte = *kpte; | |
a1e46212 SS |
1171 | if (!pte_val(old_pte)) |
1172 | return __cpa_process_fault(cpa, address, primary); | |
9f4c815c | 1173 | |
30551bb3 | 1174 | if (level == PG_LEVEL_4K) { |
c31c7d48 | 1175 | pte_t new_pte; |
626c2c9d | 1176 | pgprot_t new_prot = pte_pgprot(old_pte); |
c31c7d48 | 1177 | unsigned long pfn = pte_pfn(old_pte); |
86f03989 | 1178 | |
72e458df TG |
1179 | pgprot_val(new_prot) &= ~pgprot_val(cpa->mask_clr); |
1180 | pgprot_val(new_prot) |= pgprot_val(cpa->mask_set); | |
86f03989 | 1181 | |
c31c7d48 | 1182 | new_prot = static_protections(new_prot, address, pfn); |
86f03989 | 1183 | |
a8aed3e0 AA |
1184 | /* |
1185 | * Set the GLOBAL flags only if the PRESENT flag is | |
1186 | * set otherwise pte_present will return true even on | |
1187 | * a non present pte. The canon_pgprot will clear | |
1188 | * _PAGE_GLOBAL for the ancient hardware that doesn't | |
1189 | * support it. | |
1190 | */ | |
1191 | if (pgprot_val(new_prot) & _PAGE_PRESENT) | |
1192 | pgprot_val(new_prot) |= _PAGE_GLOBAL; | |
1193 | else | |
1194 | pgprot_val(new_prot) &= ~_PAGE_GLOBAL; | |
1195 | ||
626c2c9d AV |
1196 | /* |
1197 | * We need to keep the pfn from the existing PTE, | |
1198 | * after all we're only going to change it's attributes | |
1199 | * not the memory it points to | |
1200 | */ | |
c31c7d48 TG |
1201 | new_pte = pfn_pte(pfn, canon_pgprot(new_prot)); |
1202 | cpa->pfn = pfn; | |
f4ae5da0 TG |
1203 | /* |
1204 | * Do we really change anything ? | |
1205 | */ | |
1206 | if (pte_val(old_pte) != pte_val(new_pte)) { | |
1207 | set_pte_atomic(kpte, new_pte); | |
d75586ad | 1208 | cpa->flags |= CPA_FLUSHTLB; |
f4ae5da0 | 1209 | } |
9b5cf48b | 1210 | cpa->numpages = 1; |
65e074df | 1211 | return 0; |
1da177e4 | 1212 | } |
65e074df TG |
1213 | |
1214 | /* | |
1215 | * Check, whether we can keep the large page intact | |
1216 | * and just change the pte: | |
1217 | */ | |
beaff633 | 1218 | do_split = try_preserve_large_page(kpte, address, cpa); |
65e074df TG |
1219 | /* |
1220 | * When the range fits into the existing large page, | |
9b5cf48b | 1221 | * return. cp->numpages and cpa->tlbflush have been updated in |
65e074df TG |
1222 | * try_large_page: |
1223 | */ | |
87f7f8fe IM |
1224 | if (do_split <= 0) |
1225 | return do_split; | |
65e074df TG |
1226 | |
1227 | /* | |
1228 | * We have to split the large page: | |
1229 | */ | |
82f0712c | 1230 | err = split_large_page(cpa, kpte, address); |
87f7f8fe | 1231 | if (!err) { |
ad5ca55f SS |
1232 | /* |
1233 | * Do a global flush tlb after splitting the large page | |
1234 | * and before we do the actual change page attribute in the PTE. | |
1235 | * | |
1236 | * With out this, we violate the TLB application note, that says | |
1237 | * "The TLBs may contain both ordinary and large-page | |
1238 | * translations for a 4-KByte range of linear addresses. This | |
1239 | * may occur if software modifies the paging structures so that | |
1240 | * the page size used for the address range changes. If the two | |
1241 | * translations differ with respect to page frame or attributes | |
1242 | * (e.g., permissions), processor behavior is undefined and may | |
1243 | * be implementation-specific." | |
1244 | * | |
1245 | * We do this global tlb flush inside the cpa_lock, so that we | |
1246 | * don't allow any other cpu, with stale tlb entries change the | |
1247 | * page attribute in parallel, that also falls into the | |
1248 | * just split large page entry. | |
1249 | */ | |
1250 | flush_tlb_all(); | |
87f7f8fe IM |
1251 | goto repeat; |
1252 | } | |
beaff633 | 1253 | |
87f7f8fe | 1254 | return err; |
9f4c815c | 1255 | } |
1da177e4 | 1256 | |
c31c7d48 TG |
1257 | static int __change_page_attr_set_clr(struct cpa_data *cpa, int checkalias); |
1258 | ||
1259 | static int cpa_process_alias(struct cpa_data *cpa) | |
1da177e4 | 1260 | { |
c31c7d48 | 1261 | struct cpa_data alias_cpa; |
992f4c1c | 1262 | unsigned long laddr = (unsigned long)__va(cpa->pfn << PAGE_SHIFT); |
e933a73f | 1263 | unsigned long vaddr; |
992f4c1c | 1264 | int ret; |
44af6c41 | 1265 | |
8eb5779f | 1266 | if (!pfn_range_is_mapped(cpa->pfn, cpa->pfn + 1)) |
c31c7d48 | 1267 | return 0; |
626c2c9d | 1268 | |
f34b439f TG |
1269 | /* |
1270 | * No need to redo, when the primary call touched the direct | |
1271 | * mapping already: | |
1272 | */ | |
8523acfe TH |
1273 | if (cpa->flags & CPA_PAGES_ARRAY) { |
1274 | struct page *page = cpa->pages[cpa->curpage]; | |
1275 | if (unlikely(PageHighMem(page))) | |
1276 | return 0; | |
1277 | vaddr = (unsigned long)page_address(page); | |
1278 | } else if (cpa->flags & CPA_ARRAY) | |
d75586ad SL |
1279 | vaddr = cpa->vaddr[cpa->curpage]; |
1280 | else | |
1281 | vaddr = *cpa->vaddr; | |
1282 | ||
1283 | if (!(within(vaddr, PAGE_OFFSET, | |
a1e46212 | 1284 | PAGE_OFFSET + (max_pfn_mapped << PAGE_SHIFT)))) { |
44af6c41 | 1285 | |
f34b439f | 1286 | alias_cpa = *cpa; |
992f4c1c | 1287 | alias_cpa.vaddr = &laddr; |
9ae28475 | 1288 | alias_cpa.flags &= ~(CPA_PAGES_ARRAY | CPA_ARRAY); |
d75586ad | 1289 | |
f34b439f | 1290 | ret = __change_page_attr_set_clr(&alias_cpa, 0); |
992f4c1c TH |
1291 | if (ret) |
1292 | return ret; | |
f34b439f | 1293 | } |
44af6c41 | 1294 | |
44af6c41 | 1295 | #ifdef CONFIG_X86_64 |
488fd995 | 1296 | /* |
992f4c1c TH |
1297 | * If the primary call didn't touch the high mapping already |
1298 | * and the physical address is inside the kernel map, we need | |
0879750f | 1299 | * to touch the high mapped kernel as well: |
488fd995 | 1300 | */ |
992f4c1c TH |
1301 | if (!within(vaddr, (unsigned long)_text, _brk_end) && |
1302 | within(cpa->pfn, highmap_start_pfn(), highmap_end_pfn())) { | |
1303 | unsigned long temp_cpa_vaddr = (cpa->pfn << PAGE_SHIFT) + | |
1304 | __START_KERNEL_map - phys_base; | |
1305 | alias_cpa = *cpa; | |
1306 | alias_cpa.vaddr = &temp_cpa_vaddr; | |
1307 | alias_cpa.flags &= ~(CPA_PAGES_ARRAY | CPA_ARRAY); | |
c31c7d48 | 1308 | |
992f4c1c TH |
1309 | /* |
1310 | * The high mapping range is imprecise, so ignore the | |
1311 | * return value. | |
1312 | */ | |
1313 | __change_page_attr_set_clr(&alias_cpa, 0); | |
1314 | } | |
488fd995 | 1315 | #endif |
992f4c1c TH |
1316 | |
1317 | return 0; | |
1da177e4 LT |
1318 | } |
1319 | ||
c31c7d48 | 1320 | static int __change_page_attr_set_clr(struct cpa_data *cpa, int checkalias) |
ff31452b | 1321 | { |
65e074df | 1322 | int ret, numpages = cpa->numpages; |
ff31452b | 1323 | |
65e074df TG |
1324 | while (numpages) { |
1325 | /* | |
1326 | * Store the remaining nr of pages for the large page | |
1327 | * preservation check. | |
1328 | */ | |
9b5cf48b | 1329 | cpa->numpages = numpages; |
d75586ad | 1330 | /* for array changes, we can't use large page */ |
9ae28475 | 1331 | if (cpa->flags & (CPA_ARRAY | CPA_PAGES_ARRAY)) |
d75586ad | 1332 | cpa->numpages = 1; |
c31c7d48 | 1333 | |
ad5ca55f SS |
1334 | if (!debug_pagealloc) |
1335 | spin_lock(&cpa_lock); | |
c31c7d48 | 1336 | ret = __change_page_attr(cpa, checkalias); |
ad5ca55f SS |
1337 | if (!debug_pagealloc) |
1338 | spin_unlock(&cpa_lock); | |
ff31452b TG |
1339 | if (ret) |
1340 | return ret; | |
ff31452b | 1341 | |
c31c7d48 TG |
1342 | if (checkalias) { |
1343 | ret = cpa_process_alias(cpa); | |
1344 | if (ret) | |
1345 | return ret; | |
1346 | } | |
1347 | ||
65e074df TG |
1348 | /* |
1349 | * Adjust the number of pages with the result of the | |
1350 | * CPA operation. Either a large page has been | |
1351 | * preserved or a single page update happened. | |
1352 | */ | |
74256377 | 1353 | BUG_ON(cpa->numpages > numpages || !cpa->numpages); |
9b5cf48b | 1354 | numpages -= cpa->numpages; |
9ae28475 | 1355 | if (cpa->flags & (CPA_PAGES_ARRAY | CPA_ARRAY)) |
d75586ad SL |
1356 | cpa->curpage++; |
1357 | else | |
1358 | *cpa->vaddr += cpa->numpages * PAGE_SIZE; | |
1359 | ||
65e074df | 1360 | } |
ff31452b TG |
1361 | return 0; |
1362 | } | |
1363 | ||
d75586ad | 1364 | static int change_page_attr_set_clr(unsigned long *addr, int numpages, |
c9caa02c | 1365 | pgprot_t mask_set, pgprot_t mask_clr, |
9ae28475 | 1366 | int force_split, int in_flag, |
1367 | struct page **pages) | |
ff31452b | 1368 | { |
72e458df | 1369 | struct cpa_data cpa; |
cacf8906 | 1370 | int ret, cache, checkalias; |
fa526d0d | 1371 | unsigned long baddr = 0; |
331e4065 | 1372 | |
82f0712c BP |
1373 | memset(&cpa, 0, sizeof(cpa)); |
1374 | ||
331e4065 TG |
1375 | /* |
1376 | * Check, if we are requested to change a not supported | |
1377 | * feature: | |
1378 | */ | |
1379 | mask_set = canon_pgprot(mask_set); | |
1380 | mask_clr = canon_pgprot(mask_clr); | |
c9caa02c | 1381 | if (!pgprot_val(mask_set) && !pgprot_val(mask_clr) && !force_split) |
331e4065 TG |
1382 | return 0; |
1383 | ||
69b1415e | 1384 | /* Ensure we are PAGE_SIZE aligned */ |
9ae28475 | 1385 | if (in_flag & CPA_ARRAY) { |
d75586ad SL |
1386 | int i; |
1387 | for (i = 0; i < numpages; i++) { | |
1388 | if (addr[i] & ~PAGE_MASK) { | |
1389 | addr[i] &= PAGE_MASK; | |
1390 | WARN_ON_ONCE(1); | |
1391 | } | |
1392 | } | |
9ae28475 | 1393 | } else if (!(in_flag & CPA_PAGES_ARRAY)) { |
1394 | /* | |
1395 | * in_flag of CPA_PAGES_ARRAY implies it is aligned. | |
1396 | * No need to cehck in that case | |
1397 | */ | |
1398 | if (*addr & ~PAGE_MASK) { | |
1399 | *addr &= PAGE_MASK; | |
1400 | /* | |
1401 | * People should not be passing in unaligned addresses: | |
1402 | */ | |
1403 | WARN_ON_ONCE(1); | |
1404 | } | |
fa526d0d JS |
1405 | /* |
1406 | * Save address for cache flush. *addr is modified in the call | |
1407 | * to __change_page_attr_set_clr() below. | |
1408 | */ | |
1409 | baddr = *addr; | |
69b1415e TG |
1410 | } |
1411 | ||
5843d9a4 NP |
1412 | /* Must avoid aliasing mappings in the highmem code */ |
1413 | kmap_flush_unused(); | |
1414 | ||
db64fe02 NP |
1415 | vm_unmap_aliases(); |
1416 | ||
72e458df | 1417 | cpa.vaddr = addr; |
9ae28475 | 1418 | cpa.pages = pages; |
72e458df TG |
1419 | cpa.numpages = numpages; |
1420 | cpa.mask_set = mask_set; | |
1421 | cpa.mask_clr = mask_clr; | |
d75586ad SL |
1422 | cpa.flags = 0; |
1423 | cpa.curpage = 0; | |
c9caa02c | 1424 | cpa.force_split = force_split; |
72e458df | 1425 | |
9ae28475 | 1426 | if (in_flag & (CPA_ARRAY | CPA_PAGES_ARRAY)) |
1427 | cpa.flags |= in_flag; | |
d75586ad | 1428 | |
af96e443 TG |
1429 | /* No alias checking for _NX bit modifications */ |
1430 | checkalias = (pgprot_val(mask_set) | pgprot_val(mask_clr)) != _PAGE_NX; | |
1431 | ||
1432 | ret = __change_page_attr_set_clr(&cpa, checkalias); | |
ff31452b | 1433 | |
f4ae5da0 TG |
1434 | /* |
1435 | * Check whether we really changed something: | |
1436 | */ | |
d75586ad | 1437 | if (!(cpa.flags & CPA_FLUSHTLB)) |
1ac2f7d5 | 1438 | goto out; |
cacf8906 | 1439 | |
6bb8383b AK |
1440 | /* |
1441 | * No need to flush, when we did not set any of the caching | |
1442 | * attributes: | |
1443 | */ | |
c06814d8 | 1444 | cache = !!pgprot2cachemode(mask_set); |
6bb8383b | 1445 | |
57a6a46a | 1446 | /* |
b82ad3d3 BP |
1447 | * On success we use CLFLUSH, when the CPU supports it to |
1448 | * avoid the WBINVD. If the CPU does not support it and in the | |
f026cfa8 | 1449 | * error case we fall back to cpa_flush_all (which uses |
b82ad3d3 | 1450 | * WBINVD): |
57a6a46a | 1451 | */ |
f026cfa8 | 1452 | if (!ret && cpu_has_clflush) { |
9ae28475 | 1453 | if (cpa.flags & (CPA_PAGES_ARRAY | CPA_ARRAY)) { |
1454 | cpa_flush_array(addr, numpages, cache, | |
1455 | cpa.flags, pages); | |
1456 | } else | |
fa526d0d | 1457 | cpa_flush_range(baddr, numpages, cache); |
d75586ad | 1458 | } else |
6bb8383b | 1459 | cpa_flush_all(cache); |
cacf8906 | 1460 | |
76ebd054 | 1461 | out: |
ff31452b TG |
1462 | return ret; |
1463 | } | |
1464 | ||
d75586ad SL |
1465 | static inline int change_page_attr_set(unsigned long *addr, int numpages, |
1466 | pgprot_t mask, int array) | |
75cbade8 | 1467 | { |
d75586ad | 1468 | return change_page_attr_set_clr(addr, numpages, mask, __pgprot(0), 0, |
9ae28475 | 1469 | (array ? CPA_ARRAY : 0), NULL); |
75cbade8 AV |
1470 | } |
1471 | ||
d75586ad SL |
1472 | static inline int change_page_attr_clear(unsigned long *addr, int numpages, |
1473 | pgprot_t mask, int array) | |
72932c7a | 1474 | { |
d75586ad | 1475 | return change_page_attr_set_clr(addr, numpages, __pgprot(0), mask, 0, |
9ae28475 | 1476 | (array ? CPA_ARRAY : 0), NULL); |
72932c7a TG |
1477 | } |
1478 | ||
0f350755 | 1479 | static inline int cpa_set_pages_array(struct page **pages, int numpages, |
1480 | pgprot_t mask) | |
1481 | { | |
1482 | return change_page_attr_set_clr(NULL, numpages, mask, __pgprot(0), 0, | |
1483 | CPA_PAGES_ARRAY, pages); | |
1484 | } | |
1485 | ||
1486 | static inline int cpa_clear_pages_array(struct page **pages, int numpages, | |
1487 | pgprot_t mask) | |
1488 | { | |
1489 | return change_page_attr_set_clr(NULL, numpages, __pgprot(0), mask, 0, | |
1490 | CPA_PAGES_ARRAY, pages); | |
1491 | } | |
1492 | ||
1219333d | 1493 | int _set_memory_uc(unsigned long addr, int numpages) |
72932c7a | 1494 | { |
de33c442 SS |
1495 | /* |
1496 | * for now UC MINUS. see comments in ioremap_nocache() | |
e4b6be33 LR |
1497 | * If you really need strong UC use ioremap_uc(), but note |
1498 | * that you cannot override IO areas with set_memory_*() as | |
1499 | * these helpers cannot work with IO memory. | |
de33c442 | 1500 | */ |
d75586ad | 1501 | return change_page_attr_set(&addr, numpages, |
c06814d8 JG |
1502 | cachemode2pgprot(_PAGE_CACHE_MODE_UC_MINUS), |
1503 | 0); | |
75cbade8 | 1504 | } |
1219333d | 1505 | |
1506 | int set_memory_uc(unsigned long addr, int numpages) | |
1507 | { | |
9fa3ab39 | 1508 | int ret; |
1509 | ||
de33c442 SS |
1510 | /* |
1511 | * for now UC MINUS. see comments in ioremap_nocache() | |
1512 | */ | |
9fa3ab39 | 1513 | ret = reserve_memtype(__pa(addr), __pa(addr) + numpages * PAGE_SIZE, |
e00c8cc9 | 1514 | _PAGE_CACHE_MODE_UC_MINUS, NULL); |
9fa3ab39 | 1515 | if (ret) |
1516 | goto out_err; | |
1517 | ||
1518 | ret = _set_memory_uc(addr, numpages); | |
1519 | if (ret) | |
1520 | goto out_free; | |
1521 | ||
1522 | return 0; | |
1219333d | 1523 | |
9fa3ab39 | 1524 | out_free: |
1525 | free_memtype(__pa(addr), __pa(addr) + numpages * PAGE_SIZE); | |
1526 | out_err: | |
1527 | return ret; | |
1219333d | 1528 | } |
75cbade8 AV |
1529 | EXPORT_SYMBOL(set_memory_uc); |
1530 | ||
2d070eff | 1531 | static int _set_memory_array(unsigned long *addr, int addrinarray, |
c06814d8 | 1532 | enum page_cache_mode new_type) |
d75586ad | 1533 | { |
623dffb2 | 1534 | enum page_cache_mode set_type; |
9fa3ab39 | 1535 | int i, j; |
1536 | int ret; | |
1537 | ||
d75586ad | 1538 | for (i = 0; i < addrinarray; i++) { |
9fa3ab39 | 1539 | ret = reserve_memtype(__pa(addr[i]), __pa(addr[i]) + PAGE_SIZE, |
4f646254 | 1540 | new_type, NULL); |
9fa3ab39 | 1541 | if (ret) |
1542 | goto out_free; | |
d75586ad SL |
1543 | } |
1544 | ||
623dffb2 TK |
1545 | /* If WC, set to UC- first and then WC */ |
1546 | set_type = (new_type == _PAGE_CACHE_MODE_WC) ? | |
1547 | _PAGE_CACHE_MODE_UC_MINUS : new_type; | |
1548 | ||
9fa3ab39 | 1549 | ret = change_page_attr_set(addr, addrinarray, |
623dffb2 | 1550 | cachemode2pgprot(set_type), 1); |
4f646254 | 1551 | |
c06814d8 | 1552 | if (!ret && new_type == _PAGE_CACHE_MODE_WC) |
4f646254 | 1553 | ret = change_page_attr_set_clr(addr, addrinarray, |
c06814d8 JG |
1554 | cachemode2pgprot( |
1555 | _PAGE_CACHE_MODE_WC), | |
4f646254 PN |
1556 | __pgprot(_PAGE_CACHE_MASK), |
1557 | 0, CPA_ARRAY, NULL); | |
9fa3ab39 | 1558 | if (ret) |
1559 | goto out_free; | |
1560 | ||
1561 | return 0; | |
1562 | ||
1563 | out_free: | |
1564 | for (j = 0; j < i; j++) | |
1565 | free_memtype(__pa(addr[j]), __pa(addr[j]) + PAGE_SIZE); | |
1566 | ||
1567 | return ret; | |
d75586ad | 1568 | } |
4f646254 PN |
1569 | |
1570 | int set_memory_array_uc(unsigned long *addr, int addrinarray) | |
1571 | { | |
c06814d8 | 1572 | return _set_memory_array(addr, addrinarray, _PAGE_CACHE_MODE_UC_MINUS); |
4f646254 | 1573 | } |
d75586ad SL |
1574 | EXPORT_SYMBOL(set_memory_array_uc); |
1575 | ||
4f646254 PN |
1576 | int set_memory_array_wc(unsigned long *addr, int addrinarray) |
1577 | { | |
c06814d8 | 1578 | return _set_memory_array(addr, addrinarray, _PAGE_CACHE_MODE_WC); |
4f646254 PN |
1579 | } |
1580 | EXPORT_SYMBOL(set_memory_array_wc); | |
1581 | ||
623dffb2 TK |
1582 | int set_memory_array_wt(unsigned long *addr, int addrinarray) |
1583 | { | |
1584 | return _set_memory_array(addr, addrinarray, _PAGE_CACHE_MODE_WT); | |
1585 | } | |
1586 | EXPORT_SYMBOL_GPL(set_memory_array_wt); | |
1587 | ||
ef354af4 | 1588 | int _set_memory_wc(unsigned long addr, int numpages) |
1589 | { | |
3869c4aa | 1590 | int ret; |
bdc6340f PV |
1591 | unsigned long addr_copy = addr; |
1592 | ||
3869c4aa | 1593 | ret = change_page_attr_set(&addr, numpages, |
c06814d8 JG |
1594 | cachemode2pgprot(_PAGE_CACHE_MODE_UC_MINUS), |
1595 | 0); | |
3869c4aa | 1596 | if (!ret) { |
bdc6340f | 1597 | ret = change_page_attr_set_clr(&addr_copy, numpages, |
c06814d8 JG |
1598 | cachemode2pgprot( |
1599 | _PAGE_CACHE_MODE_WC), | |
bdc6340f PV |
1600 | __pgprot(_PAGE_CACHE_MASK), |
1601 | 0, 0, NULL); | |
3869c4aa | 1602 | } |
1603 | return ret; | |
ef354af4 | 1604 | } |
1605 | ||
1606 | int set_memory_wc(unsigned long addr, int numpages) | |
1607 | { | |
9fa3ab39 | 1608 | int ret; |
1609 | ||
9fa3ab39 | 1610 | ret = reserve_memtype(__pa(addr), __pa(addr) + numpages * PAGE_SIZE, |
e00c8cc9 | 1611 | _PAGE_CACHE_MODE_WC, NULL); |
9fa3ab39 | 1612 | if (ret) |
623dffb2 | 1613 | return ret; |
ef354af4 | 1614 | |
9fa3ab39 | 1615 | ret = _set_memory_wc(addr, numpages); |
1616 | if (ret) | |
623dffb2 | 1617 | free_memtype(__pa(addr), __pa(addr) + numpages * PAGE_SIZE); |
9fa3ab39 | 1618 | |
9fa3ab39 | 1619 | return ret; |
ef354af4 | 1620 | } |
1621 | EXPORT_SYMBOL(set_memory_wc); | |
1622 | ||
623dffb2 TK |
1623 | int _set_memory_wt(unsigned long addr, int numpages) |
1624 | { | |
1625 | return change_page_attr_set(&addr, numpages, | |
1626 | cachemode2pgprot(_PAGE_CACHE_MODE_WT), 0); | |
1627 | } | |
1628 | ||
1629 | int set_memory_wt(unsigned long addr, int numpages) | |
1630 | { | |
1631 | int ret; | |
1632 | ||
1633 | ret = reserve_memtype(__pa(addr), __pa(addr) + numpages * PAGE_SIZE, | |
1634 | _PAGE_CACHE_MODE_WT, NULL); | |
1635 | if (ret) | |
1636 | return ret; | |
1637 | ||
1638 | ret = _set_memory_wt(addr, numpages); | |
1639 | if (ret) | |
1640 | free_memtype(__pa(addr), __pa(addr) + numpages * PAGE_SIZE); | |
1641 | ||
1642 | return ret; | |
1643 | } | |
1644 | EXPORT_SYMBOL_GPL(set_memory_wt); | |
1645 | ||
1219333d | 1646 | int _set_memory_wb(unsigned long addr, int numpages) |
75cbade8 | 1647 | { |
c06814d8 | 1648 | /* WB cache mode is hard wired to all cache attribute bits being 0 */ |
d75586ad SL |
1649 | return change_page_attr_clear(&addr, numpages, |
1650 | __pgprot(_PAGE_CACHE_MASK), 0); | |
75cbade8 | 1651 | } |
1219333d | 1652 | |
1653 | int set_memory_wb(unsigned long addr, int numpages) | |
1654 | { | |
9fa3ab39 | 1655 | int ret; |
1656 | ||
1657 | ret = _set_memory_wb(addr, numpages); | |
1658 | if (ret) | |
1659 | return ret; | |
1660 | ||
c15238df | 1661 | free_memtype(__pa(addr), __pa(addr) + numpages * PAGE_SIZE); |
9fa3ab39 | 1662 | return 0; |
1219333d | 1663 | } |
75cbade8 AV |
1664 | EXPORT_SYMBOL(set_memory_wb); |
1665 | ||
d75586ad SL |
1666 | int set_memory_array_wb(unsigned long *addr, int addrinarray) |
1667 | { | |
1668 | int i; | |
a5593e0b | 1669 | int ret; |
1670 | ||
c06814d8 | 1671 | /* WB cache mode is hard wired to all cache attribute bits being 0 */ |
a5593e0b | 1672 | ret = change_page_attr_clear(addr, addrinarray, |
1673 | __pgprot(_PAGE_CACHE_MASK), 1); | |
9fa3ab39 | 1674 | if (ret) |
1675 | return ret; | |
d75586ad | 1676 | |
9fa3ab39 | 1677 | for (i = 0; i < addrinarray; i++) |
1678 | free_memtype(__pa(addr[i]), __pa(addr[i]) + PAGE_SIZE); | |
c5e147cf | 1679 | |
9fa3ab39 | 1680 | return 0; |
d75586ad SL |
1681 | } |
1682 | EXPORT_SYMBOL(set_memory_array_wb); | |
1683 | ||
75cbade8 AV |
1684 | int set_memory_x(unsigned long addr, int numpages) |
1685 | { | |
583140af PA |
1686 | if (!(__supported_pte_mask & _PAGE_NX)) |
1687 | return 0; | |
1688 | ||
d75586ad | 1689 | return change_page_attr_clear(&addr, numpages, __pgprot(_PAGE_NX), 0); |
75cbade8 AV |
1690 | } |
1691 | EXPORT_SYMBOL(set_memory_x); | |
1692 | ||
1693 | int set_memory_nx(unsigned long addr, int numpages) | |
1694 | { | |
583140af PA |
1695 | if (!(__supported_pte_mask & _PAGE_NX)) |
1696 | return 0; | |
1697 | ||
d75586ad | 1698 | return change_page_attr_set(&addr, numpages, __pgprot(_PAGE_NX), 0); |
75cbade8 AV |
1699 | } |
1700 | EXPORT_SYMBOL(set_memory_nx); | |
1701 | ||
1702 | int set_memory_ro(unsigned long addr, int numpages) | |
1703 | { | |
d75586ad | 1704 | return change_page_attr_clear(&addr, numpages, __pgprot(_PAGE_RW), 0); |
75cbade8 | 1705 | } |
75cbade8 AV |
1706 | |
1707 | int set_memory_rw(unsigned long addr, int numpages) | |
1708 | { | |
d75586ad | 1709 | return change_page_attr_set(&addr, numpages, __pgprot(_PAGE_RW), 0); |
75cbade8 | 1710 | } |
f62d0f00 IM |
1711 | |
1712 | int set_memory_np(unsigned long addr, int numpages) | |
1713 | { | |
d75586ad | 1714 | return change_page_attr_clear(&addr, numpages, __pgprot(_PAGE_PRESENT), 0); |
f62d0f00 | 1715 | } |
75cbade8 | 1716 | |
c9caa02c AK |
1717 | int set_memory_4k(unsigned long addr, int numpages) |
1718 | { | |
d75586ad | 1719 | return change_page_attr_set_clr(&addr, numpages, __pgprot(0), |
9ae28475 | 1720 | __pgprot(0), 1, 0, NULL); |
c9caa02c AK |
1721 | } |
1722 | ||
75cbade8 AV |
1723 | int set_pages_uc(struct page *page, int numpages) |
1724 | { | |
1725 | unsigned long addr = (unsigned long)page_address(page); | |
75cbade8 | 1726 | |
d7c8f21a | 1727 | return set_memory_uc(addr, numpages); |
75cbade8 AV |
1728 | } |
1729 | EXPORT_SYMBOL(set_pages_uc); | |
1730 | ||
4f646254 | 1731 | static int _set_pages_array(struct page **pages, int addrinarray, |
c06814d8 | 1732 | enum page_cache_mode new_type) |
0f350755 | 1733 | { |
1734 | unsigned long start; | |
1735 | unsigned long end; | |
623dffb2 | 1736 | enum page_cache_mode set_type; |
0f350755 | 1737 | int i; |
1738 | int free_idx; | |
4f646254 | 1739 | int ret; |
0f350755 | 1740 | |
1741 | for (i = 0; i < addrinarray; i++) { | |
8523acfe TH |
1742 | if (PageHighMem(pages[i])) |
1743 | continue; | |
1744 | start = page_to_pfn(pages[i]) << PAGE_SHIFT; | |
0f350755 | 1745 | end = start + PAGE_SIZE; |
4f646254 | 1746 | if (reserve_memtype(start, end, new_type, NULL)) |
0f350755 | 1747 | goto err_out; |
1748 | } | |
1749 | ||
623dffb2 TK |
1750 | /* If WC, set to UC- first and then WC */ |
1751 | set_type = (new_type == _PAGE_CACHE_MODE_WC) ? | |
1752 | _PAGE_CACHE_MODE_UC_MINUS : new_type; | |
1753 | ||
4f646254 | 1754 | ret = cpa_set_pages_array(pages, addrinarray, |
623dffb2 | 1755 | cachemode2pgprot(set_type)); |
c06814d8 | 1756 | if (!ret && new_type == _PAGE_CACHE_MODE_WC) |
4f646254 | 1757 | ret = change_page_attr_set_clr(NULL, addrinarray, |
c06814d8 JG |
1758 | cachemode2pgprot( |
1759 | _PAGE_CACHE_MODE_WC), | |
4f646254 PN |
1760 | __pgprot(_PAGE_CACHE_MASK), |
1761 | 0, CPA_PAGES_ARRAY, pages); | |
1762 | if (ret) | |
1763 | goto err_out; | |
1764 | return 0; /* Success */ | |
0f350755 | 1765 | err_out: |
1766 | free_idx = i; | |
1767 | for (i = 0; i < free_idx; i++) { | |
8523acfe TH |
1768 | if (PageHighMem(pages[i])) |
1769 | continue; | |
1770 | start = page_to_pfn(pages[i]) << PAGE_SHIFT; | |
0f350755 | 1771 | end = start + PAGE_SIZE; |
1772 | free_memtype(start, end); | |
1773 | } | |
1774 | return -EINVAL; | |
1775 | } | |
4f646254 PN |
1776 | |
1777 | int set_pages_array_uc(struct page **pages, int addrinarray) | |
1778 | { | |
c06814d8 | 1779 | return _set_pages_array(pages, addrinarray, _PAGE_CACHE_MODE_UC_MINUS); |
4f646254 | 1780 | } |
0f350755 | 1781 | EXPORT_SYMBOL(set_pages_array_uc); |
1782 | ||
4f646254 PN |
1783 | int set_pages_array_wc(struct page **pages, int addrinarray) |
1784 | { | |
c06814d8 | 1785 | return _set_pages_array(pages, addrinarray, _PAGE_CACHE_MODE_WC); |
4f646254 PN |
1786 | } |
1787 | EXPORT_SYMBOL(set_pages_array_wc); | |
1788 | ||
623dffb2 TK |
1789 | int set_pages_array_wt(struct page **pages, int addrinarray) |
1790 | { | |
1791 | return _set_pages_array(pages, addrinarray, _PAGE_CACHE_MODE_WT); | |
1792 | } | |
1793 | EXPORT_SYMBOL_GPL(set_pages_array_wt); | |
1794 | ||
75cbade8 AV |
1795 | int set_pages_wb(struct page *page, int numpages) |
1796 | { | |
1797 | unsigned long addr = (unsigned long)page_address(page); | |
75cbade8 | 1798 | |
d7c8f21a | 1799 | return set_memory_wb(addr, numpages); |
75cbade8 AV |
1800 | } |
1801 | EXPORT_SYMBOL(set_pages_wb); | |
1802 | ||
0f350755 | 1803 | int set_pages_array_wb(struct page **pages, int addrinarray) |
1804 | { | |
1805 | int retval; | |
1806 | unsigned long start; | |
1807 | unsigned long end; | |
1808 | int i; | |
1809 | ||
c06814d8 | 1810 | /* WB cache mode is hard wired to all cache attribute bits being 0 */ |
0f350755 | 1811 | retval = cpa_clear_pages_array(pages, addrinarray, |
1812 | __pgprot(_PAGE_CACHE_MASK)); | |
9fa3ab39 | 1813 | if (retval) |
1814 | return retval; | |
0f350755 | 1815 | |
1816 | for (i = 0; i < addrinarray; i++) { | |
8523acfe TH |
1817 | if (PageHighMem(pages[i])) |
1818 | continue; | |
1819 | start = page_to_pfn(pages[i]) << PAGE_SHIFT; | |
0f350755 | 1820 | end = start + PAGE_SIZE; |
1821 | free_memtype(start, end); | |
1822 | } | |
1823 | ||
9fa3ab39 | 1824 | return 0; |
0f350755 | 1825 | } |
1826 | EXPORT_SYMBOL(set_pages_array_wb); | |
1827 | ||
75cbade8 AV |
1828 | int set_pages_x(struct page *page, int numpages) |
1829 | { | |
1830 | unsigned long addr = (unsigned long)page_address(page); | |
75cbade8 | 1831 | |
d7c8f21a | 1832 | return set_memory_x(addr, numpages); |
75cbade8 AV |
1833 | } |
1834 | EXPORT_SYMBOL(set_pages_x); | |
1835 | ||
1836 | int set_pages_nx(struct page *page, int numpages) | |
1837 | { | |
1838 | unsigned long addr = (unsigned long)page_address(page); | |
75cbade8 | 1839 | |
d7c8f21a | 1840 | return set_memory_nx(addr, numpages); |
75cbade8 AV |
1841 | } |
1842 | EXPORT_SYMBOL(set_pages_nx); | |
1843 | ||
1844 | int set_pages_ro(struct page *page, int numpages) | |
1845 | { | |
1846 | unsigned long addr = (unsigned long)page_address(page); | |
75cbade8 | 1847 | |
d7c8f21a | 1848 | return set_memory_ro(addr, numpages); |
75cbade8 | 1849 | } |
75cbade8 AV |
1850 | |
1851 | int set_pages_rw(struct page *page, int numpages) | |
1852 | { | |
1853 | unsigned long addr = (unsigned long)page_address(page); | |
e81d5dc4 | 1854 | |
d7c8f21a | 1855 | return set_memory_rw(addr, numpages); |
78c94aba IM |
1856 | } |
1857 | ||
1da177e4 | 1858 | #ifdef CONFIG_DEBUG_PAGEALLOC |
f62d0f00 IM |
1859 | |
1860 | static int __set_pages_p(struct page *page, int numpages) | |
1861 | { | |
d75586ad SL |
1862 | unsigned long tempaddr = (unsigned long) page_address(page); |
1863 | struct cpa_data cpa = { .vaddr = &tempaddr, | |
82f0712c | 1864 | .pgd = NULL, |
72e458df TG |
1865 | .numpages = numpages, |
1866 | .mask_set = __pgprot(_PAGE_PRESENT | _PAGE_RW), | |
d75586ad SL |
1867 | .mask_clr = __pgprot(0), |
1868 | .flags = 0}; | |
72932c7a | 1869 | |
55121b43 SS |
1870 | /* |
1871 | * No alias checking needed for setting present flag. otherwise, | |
1872 | * we may need to break large pages for 64-bit kernel text | |
1873 | * mappings (this adds to complexity if we want to do this from | |
1874 | * atomic context especially). Let's keep it simple! | |
1875 | */ | |
1876 | return __change_page_attr_set_clr(&cpa, 0); | |
f62d0f00 IM |
1877 | } |
1878 | ||
1879 | static int __set_pages_np(struct page *page, int numpages) | |
1880 | { | |
d75586ad SL |
1881 | unsigned long tempaddr = (unsigned long) page_address(page); |
1882 | struct cpa_data cpa = { .vaddr = &tempaddr, | |
82f0712c | 1883 | .pgd = NULL, |
72e458df TG |
1884 | .numpages = numpages, |
1885 | .mask_set = __pgprot(0), | |
d75586ad SL |
1886 | .mask_clr = __pgprot(_PAGE_PRESENT | _PAGE_RW), |
1887 | .flags = 0}; | |
72932c7a | 1888 | |
55121b43 SS |
1889 | /* |
1890 | * No alias checking needed for setting not present flag. otherwise, | |
1891 | * we may need to break large pages for 64-bit kernel text | |
1892 | * mappings (this adds to complexity if we want to do this from | |
1893 | * atomic context especially). Let's keep it simple! | |
1894 | */ | |
1895 | return __change_page_attr_set_clr(&cpa, 0); | |
f62d0f00 IM |
1896 | } |
1897 | ||
031bc574 | 1898 | void __kernel_map_pages(struct page *page, int numpages, int enable) |
1da177e4 LT |
1899 | { |
1900 | if (PageHighMem(page)) | |
1901 | return; | |
9f4c815c | 1902 | if (!enable) { |
f9b8404c IM |
1903 | debug_check_no_locks_freed(page_address(page), |
1904 | numpages * PAGE_SIZE); | |
9f4c815c | 1905 | } |
de5097c2 | 1906 | |
9f4c815c | 1907 | /* |
f8d8406b | 1908 | * The return value is ignored as the calls cannot fail. |
55121b43 SS |
1909 | * Large pages for identity mappings are not used at boot time |
1910 | * and hence no memory allocations during large page split. | |
1da177e4 | 1911 | */ |
f62d0f00 IM |
1912 | if (enable) |
1913 | __set_pages_p(page, numpages); | |
1914 | else | |
1915 | __set_pages_np(page, numpages); | |
9f4c815c IM |
1916 | |
1917 | /* | |
e4b71dcf IM |
1918 | * We should perform an IPI and flush all tlbs, |
1919 | * but that can deadlock->flush only current cpu: | |
1da177e4 LT |
1920 | */ |
1921 | __flush_tlb_all(); | |
26564600 BO |
1922 | |
1923 | arch_flush_lazy_mmu_mode(); | |
ee7ae7a1 TG |
1924 | } |
1925 | ||
8a235efa RW |
1926 | #ifdef CONFIG_HIBERNATION |
1927 | ||
1928 | bool kernel_page_present(struct page *page) | |
1929 | { | |
1930 | unsigned int level; | |
1931 | pte_t *pte; | |
1932 | ||
1933 | if (PageHighMem(page)) | |
1934 | return false; | |
1935 | ||
1936 | pte = lookup_address((unsigned long)page_address(page), &level); | |
1937 | return (pte_val(*pte) & _PAGE_PRESENT); | |
1938 | } | |
1939 | ||
1940 | #endif /* CONFIG_HIBERNATION */ | |
1941 | ||
1942 | #endif /* CONFIG_DEBUG_PAGEALLOC */ | |
d1028a15 | 1943 | |
82f0712c BP |
1944 | int kernel_map_pages_in_pgd(pgd_t *pgd, u64 pfn, unsigned long address, |
1945 | unsigned numpages, unsigned long page_flags) | |
1946 | { | |
1947 | int retval = -EINVAL; | |
1948 | ||
1949 | struct cpa_data cpa = { | |
1950 | .vaddr = &address, | |
1951 | .pfn = pfn, | |
1952 | .pgd = pgd, | |
1953 | .numpages = numpages, | |
1954 | .mask_set = __pgprot(0), | |
1955 | .mask_clr = __pgprot(0), | |
1956 | .flags = 0, | |
1957 | }; | |
1958 | ||
1959 | if (!(__supported_pte_mask & _PAGE_NX)) | |
1960 | goto out; | |
1961 | ||
1962 | if (!(page_flags & _PAGE_NX)) | |
1963 | cpa.mask_clr = __pgprot(_PAGE_NX); | |
1964 | ||
1965 | cpa.mask_set = __pgprot(_PAGE_PRESENT | page_flags); | |
1966 | ||
1967 | retval = __change_page_attr_set_clr(&cpa, 0); | |
1968 | __flush_tlb_all(); | |
1969 | ||
1970 | out: | |
1971 | return retval; | |
1972 | } | |
1973 | ||
42a54772 BP |
1974 | void kernel_unmap_pages_in_pgd(pgd_t *root, unsigned long address, |
1975 | unsigned numpages) | |
1976 | { | |
1977 | unmap_pgd_range(root, address, address + (numpages << PAGE_SHIFT)); | |
1978 | } | |
1979 | ||
d1028a15 AV |
1980 | /* |
1981 | * The testcases use internal knowledge of the implementation that shouldn't | |
1982 | * be exposed to the rest of the kernel. Include these directly here. | |
1983 | */ | |
1984 | #ifdef CONFIG_CPA_DEBUG | |
1985 | #include "pageattr-test.c" | |
1986 | #endif |