Commit | Line | Data |
---|---|---|
2e5d9c85 | 1 | /* |
2 | * Handle caching attributes in page tables (PAT) | |
3 | * | |
4 | * Authors: Venkatesh Pallipadi <venkatesh.pallipadi@intel.com> | |
5 | * Suresh B Siddha <suresh.b.siddha@intel.com> | |
6 | * | |
7 | * Loosely based on earlier PAT patchset from Eric Biederman and Andi Kleen. | |
8 | */ | |
9 | ||
ad2cde16 IM |
10 | #include <linux/seq_file.h> |
11 | #include <linux/bootmem.h> | |
12 | #include <linux/debugfs.h> | |
2e5d9c85 | 13 | #include <linux/kernel.h> |
92b9af9e | 14 | #include <linux/module.h> |
5a0e3ad6 | 15 | #include <linux/slab.h> |
ad2cde16 | 16 | #include <linux/mm.h> |
2e5d9c85 | 17 | #include <linux/fs.h> |
335ef896 | 18 | #include <linux/rbtree.h> |
2e5d9c85 | 19 | |
ad2cde16 | 20 | #include <asm/cacheflush.h> |
2e5d9c85 | 21 | #include <asm/processor.h> |
ad2cde16 | 22 | #include <asm/tlbflush.h> |
fd12a0d6 | 23 | #include <asm/x86_init.h> |
2e5d9c85 | 24 | #include <asm/pgtable.h> |
2e5d9c85 | 25 | #include <asm/fcntl.h> |
ad2cde16 | 26 | #include <asm/e820.h> |
2e5d9c85 | 27 | #include <asm/mtrr.h> |
ad2cde16 IM |
28 | #include <asm/page.h> |
29 | #include <asm/msr.h> | |
30 | #include <asm/pat.h> | |
e7f260a2 | 31 | #include <asm/io.h> |
2e5d9c85 | 32 | |
be5a0c12 | 33 | #include "pat_internal.h" |
34 | ||
8d4a4300 | 35 | #ifdef CONFIG_X86_PAT |
499f8f84 | 36 | int __read_mostly pat_enabled = 1; |
2e5d9c85 | 37 | |
1ee4bd92 | 38 | static inline void pat_disable(const char *reason) |
2e5d9c85 | 39 | { |
499f8f84 | 40 | pat_enabled = 0; |
8d4a4300 | 41 | printk(KERN_INFO "%s\n", reason); |
2e5d9c85 | 42 | } |
2e5d9c85 | 43 | |
be524fb9 | 44 | static int __init nopat(char *str) |
2e5d9c85 | 45 | { |
8d4a4300 | 46 | pat_disable("PAT support disabled."); |
2e5d9c85 | 47 | return 0; |
48 | } | |
8d4a4300 | 49 | early_param("nopat", nopat); |
75a04811 PA |
50 | #else |
51 | static inline void pat_disable(const char *reason) | |
52 | { | |
53 | (void)reason; | |
54 | } | |
8d4a4300 TG |
55 | #endif |
56 | ||
77b52b4c | 57 | |
be5a0c12 | 58 | int pat_debug_enable; |
ad2cde16 | 59 | |
77b52b4c VP |
60 | static int __init pat_debug_setup(char *str) |
61 | { | |
be5a0c12 | 62 | pat_debug_enable = 1; |
77b52b4c VP |
63 | return 0; |
64 | } | |
65 | __setup("debugpat", pat_debug_setup); | |
66 | ||
8d4a4300 | 67 | static u64 __read_mostly boot_pat_state; |
2e5d9c85 | 68 | |
69 | enum { | |
70 | PAT_UC = 0, /* uncached */ | |
71 | PAT_WC = 1, /* Write combining */ | |
72 | PAT_WT = 4, /* Write Through */ | |
73 | PAT_WP = 5, /* Write Protected */ | |
74 | PAT_WB = 6, /* Write Back (default) */ | |
75 | PAT_UC_MINUS = 7, /* UC, but can be overriden by MTRR */ | |
76 | }; | |
77 | ||
cd7a4e93 | 78 | #define PAT(x, y) ((u64)PAT_ ## y << ((x)*8)) |
2e5d9c85 | 79 | |
80 | void pat_init(void) | |
81 | { | |
82 | u64 pat; | |
e23a8b6a | 83 | bool boot_cpu = !boot_pat_state; |
2e5d9c85 | 84 | |
499f8f84 | 85 | if (!pat_enabled) |
2e5d9c85 | 86 | return; |
87 | ||
75a04811 PA |
88 | if (!cpu_has_pat) { |
89 | if (!boot_pat_state) { | |
90 | pat_disable("PAT not supported by CPU."); | |
91 | return; | |
92 | } else { | |
93 | /* | |
94 | * If this happens we are on a secondary CPU, but | |
95 | * switched to PAT on the boot CPU. We have no way to | |
96 | * undo PAT. | |
97 | */ | |
98 | printk(KERN_ERR "PAT enabled, " | |
99 | "but not supported by secondary CPU\n"); | |
100 | BUG(); | |
101 | } | |
8d4a4300 | 102 | } |
2e5d9c85 | 103 | |
104 | /* Set PWT to Write-Combining. All other bits stay the same */ | |
105 | /* | |
106 | * PTE encoding used in Linux: | |
107 | * PAT | |
108 | * |PCD | |
109 | * ||PWT | |
110 | * ||| | |
111 | * 000 WB _PAGE_CACHE_WB | |
112 | * 001 WC _PAGE_CACHE_WC | |
113 | * 010 UC- _PAGE_CACHE_UC_MINUS | |
114 | * 011 UC _PAGE_CACHE_UC | |
115 | * PAT bit unused | |
116 | */ | |
cd7a4e93 AH |
117 | pat = PAT(0, WB) | PAT(1, WC) | PAT(2, UC_MINUS) | PAT(3, UC) | |
118 | PAT(4, WB) | PAT(5, WC) | PAT(6, UC_MINUS) | PAT(7, UC); | |
2e5d9c85 | 119 | |
120 | /* Boot CPU check */ | |
8d4a4300 | 121 | if (!boot_pat_state) |
2e5d9c85 | 122 | rdmsrl(MSR_IA32_CR_PAT, boot_pat_state); |
2e5d9c85 | 123 | |
124 | wrmsrl(MSR_IA32_CR_PAT, pat); | |
e23a8b6a RD |
125 | |
126 | if (boot_cpu) | |
127 | printk(KERN_INFO "x86 PAT enabled: cpu %d, old 0x%Lx, new 0x%Lx\n", | |
128 | smp_processor_id(), boot_pat_state, pat); | |
2e5d9c85 | 129 | } |
130 | ||
131 | #undef PAT | |
132 | ||
9e41a49a | 133 | static DEFINE_SPINLOCK(memtype_lock); /* protects memtype accesses */ |
335ef896 | 134 | |
2e5d9c85 | 135 | /* |
136 | * Does intersection of PAT memory type and MTRR memory type and returns | |
137 | * the resulting memory type as PAT understands it. | |
138 | * (Type in pat and mtrr will not have same value) | |
139 | * The intersection is based on "Effective Memory Type" tables in IA-32 | |
140 | * SDM vol 3a | |
141 | */ | |
e00c8cc9 JG |
142 | static unsigned long pat_x_mtrr_type(u64 start, u64 end, |
143 | enum page_cache_mode req_type) | |
2e5d9c85 | 144 | { |
c26421d0 VP |
145 | /* |
146 | * Look for MTRR hint to get the effective type in case where PAT | |
147 | * request is for WB. | |
148 | */ | |
e00c8cc9 | 149 | if (req_type == _PAGE_CACHE_MODE_WB) { |
dd0c7c49 AH |
150 | u8 mtrr_type; |
151 | ||
152 | mtrr_type = mtrr_type_lookup(start, end); | |
b6ff32d9 | 153 | if (mtrr_type != MTRR_TYPE_WRBACK) |
e00c8cc9 | 154 | return _PAGE_CACHE_MODE_UC_MINUS; |
b6ff32d9 | 155 | |
e00c8cc9 | 156 | return _PAGE_CACHE_MODE_WB; |
dd0c7c49 AH |
157 | } |
158 | ||
159 | return req_type; | |
2e5d9c85 | 160 | } |
161 | ||
fa83523f JD |
162 | struct pagerange_state { |
163 | unsigned long cur_pfn; | |
164 | int ram; | |
165 | int not_ram; | |
166 | }; | |
167 | ||
168 | static int | |
169 | pagerange_is_ram_callback(unsigned long initial_pfn, unsigned long total_nr_pages, void *arg) | |
170 | { | |
171 | struct pagerange_state *state = arg; | |
172 | ||
173 | state->not_ram |= initial_pfn > state->cur_pfn; | |
174 | state->ram |= total_nr_pages > 0; | |
175 | state->cur_pfn = initial_pfn + total_nr_pages; | |
176 | ||
177 | return state->ram && state->not_ram; | |
178 | } | |
179 | ||
3709c857 | 180 | static int pat_pagerange_is_ram(resource_size_t start, resource_size_t end) |
be03d9e8 | 181 | { |
fa83523f JD |
182 | int ret = 0; |
183 | unsigned long start_pfn = start >> PAGE_SHIFT; | |
184 | unsigned long end_pfn = (end + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
185 | struct pagerange_state state = {start_pfn, 0, 0}; | |
186 | ||
187 | /* | |
188 | * For legacy reasons, physical address range in the legacy ISA | |
189 | * region is tracked as non-RAM. This will allow users of | |
190 | * /dev/mem to map portions of legacy ISA region, even when | |
191 | * some of those portions are listed(or not even listed) with | |
192 | * different e820 types(RAM/reserved/..) | |
193 | */ | |
194 | if (start_pfn < ISA_END_ADDRESS >> PAGE_SHIFT) | |
195 | start_pfn = ISA_END_ADDRESS >> PAGE_SHIFT; | |
196 | ||
197 | if (start_pfn < end_pfn) { | |
198 | ret = walk_system_ram_range(start_pfn, end_pfn - start_pfn, | |
199 | &state, pagerange_is_ram_callback); | |
be03d9e8 SS |
200 | } |
201 | ||
fa83523f | 202 | return (ret > 0) ? -1 : (state.ram ? 1 : 0); |
be03d9e8 SS |
203 | } |
204 | ||
9542ada8 | 205 | /* |
f5841740 VP |
206 | * For RAM pages, we use page flags to mark the pages with appropriate type. |
207 | * Here we do two pass: | |
208 | * - Find the memtype of all the pages in the range, look for any conflicts | |
209 | * - In case of no conflicts, set the new memtype for pages in the range | |
9542ada8 | 210 | */ |
e00c8cc9 JG |
211 | static int reserve_ram_pages_type(u64 start, u64 end, |
212 | enum page_cache_mode req_type, | |
213 | enum page_cache_mode *new_type) | |
9542ada8 SS |
214 | { |
215 | struct page *page; | |
f5841740 VP |
216 | u64 pfn; |
217 | ||
e00c8cc9 | 218 | if (req_type == _PAGE_CACHE_MODE_UC) { |
f5841740 VP |
219 | /* We do not support strong UC */ |
220 | WARN_ON_ONCE(1); | |
e00c8cc9 | 221 | req_type = _PAGE_CACHE_MODE_UC_MINUS; |
f5841740 | 222 | } |
9542ada8 SS |
223 | |
224 | for (pfn = (start >> PAGE_SHIFT); pfn < (end >> PAGE_SHIFT); ++pfn) { | |
e00c8cc9 | 225 | enum page_cache_mode type; |
9542ada8 | 226 | |
f5841740 VP |
227 | page = pfn_to_page(pfn); |
228 | type = get_page_memtype(page); | |
229 | if (type != -1) { | |
e00c8cc9 | 230 | pr_info("reserve_ram_pages_type failed [mem %#010Lx-%#010Lx], track 0x%x, req 0x%x\n", |
365811d6 | 231 | start, end - 1, type, req_type); |
f5841740 VP |
232 | if (new_type) |
233 | *new_type = type; | |
234 | ||
235 | return -EBUSY; | |
236 | } | |
9542ada8 | 237 | } |
9542ada8 | 238 | |
f5841740 VP |
239 | if (new_type) |
240 | *new_type = req_type; | |
241 | ||
242 | for (pfn = (start >> PAGE_SHIFT); pfn < (end >> PAGE_SHIFT); ++pfn) { | |
9542ada8 | 243 | page = pfn_to_page(pfn); |
f5841740 | 244 | set_page_memtype(page, req_type); |
9542ada8 | 245 | } |
f5841740 | 246 | return 0; |
9542ada8 SS |
247 | } |
248 | ||
249 | static int free_ram_pages_type(u64 start, u64 end) | |
250 | { | |
251 | struct page *page; | |
f5841740 | 252 | u64 pfn; |
9542ada8 SS |
253 | |
254 | for (pfn = (start >> PAGE_SHIFT); pfn < (end >> PAGE_SHIFT); ++pfn) { | |
255 | page = pfn_to_page(pfn); | |
f5841740 | 256 | set_page_memtype(page, -1); |
9542ada8 SS |
257 | } |
258 | return 0; | |
9542ada8 SS |
259 | } |
260 | ||
e7f260a2 | 261 | /* |
262 | * req_type typically has one of the: | |
e00c8cc9 JG |
263 | * - _PAGE_CACHE_MODE_WB |
264 | * - _PAGE_CACHE_MODE_WC | |
265 | * - _PAGE_CACHE_MODE_UC_MINUS | |
266 | * - _PAGE_CACHE_MODE_UC | |
e7f260a2 | 267 | * |
ac97991e AH |
268 | * If new_type is NULL, function will return an error if it cannot reserve the |
269 | * region with req_type. If new_type is non-NULL, function will return | |
270 | * available type in new_type in case of no error. In case of any error | |
e7f260a2 | 271 | * it will return a negative return value. |
272 | */ | |
e00c8cc9 JG |
273 | int reserve_memtype(u64 start, u64 end, enum page_cache_mode req_type, |
274 | enum page_cache_mode *new_type) | |
2e5d9c85 | 275 | { |
be5a0c12 | 276 | struct memtype *new; |
e00c8cc9 | 277 | enum page_cache_mode actual_type; |
9542ada8 | 278 | int is_range_ram; |
ad2cde16 | 279 | int err = 0; |
2e5d9c85 | 280 | |
ad2cde16 | 281 | BUG_ON(start >= end); /* end is exclusive */ |
69e26be9 | 282 | |
499f8f84 | 283 | if (!pat_enabled) { |
e7f260a2 | 284 | /* This is identical to page table setting without PAT */ |
ac97991e | 285 | if (new_type) { |
e00c8cc9 JG |
286 | if (req_type == _PAGE_CACHE_MODE_WC) |
287 | *new_type = _PAGE_CACHE_MODE_UC_MINUS; | |
ac97991e | 288 | else |
e00c8cc9 | 289 | *new_type = req_type; |
e7f260a2 | 290 | } |
2e5d9c85 | 291 | return 0; |
292 | } | |
293 | ||
294 | /* Low ISA region is always mapped WB in page table. No need to track */ | |
8a271389 | 295 | if (x86_platform.is_untracked_pat_range(start, end)) { |
ac97991e | 296 | if (new_type) |
e00c8cc9 | 297 | *new_type = _PAGE_CACHE_MODE_WB; |
2e5d9c85 | 298 | return 0; |
299 | } | |
300 | ||
b6ff32d9 SS |
301 | /* |
302 | * Call mtrr_lookup to get the type hint. This is an | |
303 | * optimization for /dev/mem mmap'ers into WB memory (BIOS | |
304 | * tools and ACPI tools). Use WB request for WB memory and use | |
305 | * UC_MINUS otherwise. | |
306 | */ | |
e00c8cc9 | 307 | actual_type = pat_x_mtrr_type(start, end, req_type); |
2e5d9c85 | 308 | |
95971342 SS |
309 | if (new_type) |
310 | *new_type = actual_type; | |
311 | ||
be03d9e8 | 312 | is_range_ram = pat_pagerange_is_ram(start, end); |
f5841740 VP |
313 | if (is_range_ram == 1) { |
314 | ||
f5841740 | 315 | err = reserve_ram_pages_type(start, end, req_type, new_type); |
f5841740 VP |
316 | |
317 | return err; | |
318 | } else if (is_range_ram < 0) { | |
9542ada8 | 319 | return -EINVAL; |
f5841740 | 320 | } |
9542ada8 | 321 | |
6a4f3b52 | 322 | new = kzalloc(sizeof(struct memtype), GFP_KERNEL); |
ac97991e | 323 | if (!new) |
2e5d9c85 | 324 | return -ENOMEM; |
325 | ||
ad2cde16 IM |
326 | new->start = start; |
327 | new->end = end; | |
328 | new->type = actual_type; | |
2e5d9c85 | 329 | |
2e5d9c85 | 330 | spin_lock(&memtype_lock); |
331 | ||
9e41a49a | 332 | err = rbt_memtype_check_insert(new, new_type); |
2e5d9c85 | 333 | if (err) { |
365811d6 BH |
334 | printk(KERN_INFO "reserve_memtype failed [mem %#010Lx-%#010Lx], track %s, req %s\n", |
335 | start, end - 1, | |
336 | cattr_name(new->type), cattr_name(req_type)); | |
ac97991e | 337 | kfree(new); |
2e5d9c85 | 338 | spin_unlock(&memtype_lock); |
ad2cde16 | 339 | |
2e5d9c85 | 340 | return err; |
341 | } | |
342 | ||
2e5d9c85 | 343 | spin_unlock(&memtype_lock); |
3e9c83b3 | 344 | |
365811d6 BH |
345 | dprintk("reserve_memtype added [mem %#010Lx-%#010Lx], track %s, req %s, ret %s\n", |
346 | start, end - 1, cattr_name(new->type), cattr_name(req_type), | |
3e9c83b3 AH |
347 | new_type ? cattr_name(*new_type) : "-"); |
348 | ||
2e5d9c85 | 349 | return err; |
350 | } | |
351 | ||
352 | int free_memtype(u64 start, u64 end) | |
353 | { | |
2e5d9c85 | 354 | int err = -EINVAL; |
9542ada8 | 355 | int is_range_ram; |
20413f27 | 356 | struct memtype *entry; |
2e5d9c85 | 357 | |
69e26be9 | 358 | if (!pat_enabled) |
2e5d9c85 | 359 | return 0; |
2e5d9c85 | 360 | |
361 | /* Low ISA region is always mapped WB. No need to track */ | |
8a271389 | 362 | if (x86_platform.is_untracked_pat_range(start, end)) |
2e5d9c85 | 363 | return 0; |
2e5d9c85 | 364 | |
be03d9e8 | 365 | is_range_ram = pat_pagerange_is_ram(start, end); |
f5841740 VP |
366 | if (is_range_ram == 1) { |
367 | ||
f5841740 | 368 | err = free_ram_pages_type(start, end); |
f5841740 VP |
369 | |
370 | return err; | |
371 | } else if (is_range_ram < 0) { | |
9542ada8 | 372 | return -EINVAL; |
f5841740 | 373 | } |
9542ada8 | 374 | |
2e5d9c85 | 375 | spin_lock(&memtype_lock); |
20413f27 | 376 | entry = rbt_memtype_erase(start, end); |
2e5d9c85 | 377 | spin_unlock(&memtype_lock); |
378 | ||
20413f27 | 379 | if (!entry) { |
365811d6 BH |
380 | printk(KERN_INFO "%s:%d freeing invalid memtype [mem %#010Lx-%#010Lx]\n", |
381 | current->comm, current->pid, start, end - 1); | |
20413f27 | 382 | return -EINVAL; |
2e5d9c85 | 383 | } |
6997ab49 | 384 | |
20413f27 XF |
385 | kfree(entry); |
386 | ||
365811d6 | 387 | dprintk("free_memtype request [mem %#010Lx-%#010Lx]\n", start, end - 1); |
ad2cde16 | 388 | |
20413f27 | 389 | return 0; |
2e5d9c85 | 390 | } |
391 | ||
f0970c13 | 392 | |
637b86e7 VP |
393 | /** |
394 | * lookup_memtype - Looksup the memory type for a physical address | |
395 | * @paddr: physical address of which memory type needs to be looked up | |
396 | * | |
397 | * Only to be called when PAT is enabled | |
398 | * | |
2a374698 JG |
399 | * Returns _PAGE_CACHE_MODE_WB, _PAGE_CACHE_MODE_WC, _PAGE_CACHE_MODE_UC_MINUS |
400 | * or _PAGE_CACHE_MODE_UC | |
637b86e7 | 401 | */ |
2a374698 | 402 | static enum page_cache_mode lookup_memtype(u64 paddr) |
637b86e7 | 403 | { |
2a374698 | 404 | enum page_cache_mode rettype = _PAGE_CACHE_MODE_WB; |
637b86e7 VP |
405 | struct memtype *entry; |
406 | ||
8a271389 | 407 | if (x86_platform.is_untracked_pat_range(paddr, paddr + PAGE_SIZE)) |
637b86e7 VP |
408 | return rettype; |
409 | ||
410 | if (pat_pagerange_is_ram(paddr, paddr + PAGE_SIZE)) { | |
411 | struct page *page; | |
637b86e7 | 412 | page = pfn_to_page(paddr >> PAGE_SHIFT); |
e00c8cc9 | 413 | rettype = get_page_memtype(page); |
637b86e7 VP |
414 | /* |
415 | * -1 from get_page_memtype() implies RAM page is in its | |
416 | * default state and not reserved, and hence of type WB | |
417 | */ | |
418 | if (rettype == -1) | |
2a374698 | 419 | rettype = _PAGE_CACHE_MODE_WB; |
637b86e7 VP |
420 | |
421 | return rettype; | |
422 | } | |
423 | ||
424 | spin_lock(&memtype_lock); | |
425 | ||
9e41a49a | 426 | entry = rbt_memtype_lookup(paddr); |
637b86e7 | 427 | if (entry != NULL) |
e00c8cc9 | 428 | rettype = entry->type; |
637b86e7 | 429 | else |
2a374698 | 430 | rettype = _PAGE_CACHE_MODE_UC_MINUS; |
637b86e7 VP |
431 | |
432 | spin_unlock(&memtype_lock); | |
433 | return rettype; | |
434 | } | |
435 | ||
9fd126bc VP |
436 | /** |
437 | * io_reserve_memtype - Request a memory type mapping for a region of memory | |
438 | * @start: start (physical address) of the region | |
439 | * @end: end (physical address) of the region | |
440 | * @type: A pointer to memtype, with requested type. On success, requested | |
441 | * or any other compatible type that was available for the region is returned | |
442 | * | |
443 | * On success, returns 0 | |
444 | * On failure, returns non-zero | |
445 | */ | |
446 | int io_reserve_memtype(resource_size_t start, resource_size_t end, | |
49a3b3cb | 447 | enum page_cache_mode *type) |
9fd126bc | 448 | { |
b855192c | 449 | resource_size_t size = end - start; |
49a3b3cb JG |
450 | enum page_cache_mode req_type = *type; |
451 | enum page_cache_mode new_type; | |
9fd126bc VP |
452 | int ret; |
453 | ||
b855192c | 454 | WARN_ON_ONCE(iomem_map_sanity_check(start, size)); |
9fd126bc | 455 | |
e00c8cc9 | 456 | ret = reserve_memtype(start, end, req_type, &new_type); |
9fd126bc VP |
457 | if (ret) |
458 | goto out_err; | |
459 | ||
49a3b3cb | 460 | if (!is_new_memtype_allowed(start, size, req_type, new_type)) |
9fd126bc VP |
461 | goto out_free; |
462 | ||
b14097bd | 463 | if (kernel_map_sync_memtype(start, size, new_type) < 0) |
9fd126bc VP |
464 | goto out_free; |
465 | ||
466 | *type = new_type; | |
467 | return 0; | |
468 | ||
469 | out_free: | |
470 | free_memtype(start, end); | |
471 | ret = -EBUSY; | |
472 | out_err: | |
473 | return ret; | |
474 | } | |
475 | ||
476 | /** | |
477 | * io_free_memtype - Release a memory type mapping for a region of memory | |
478 | * @start: start (physical address) of the region | |
479 | * @end: end (physical address) of the region | |
480 | */ | |
481 | void io_free_memtype(resource_size_t start, resource_size_t end) | |
482 | { | |
483 | free_memtype(start, end); | |
484 | } | |
485 | ||
f0970c13 | 486 | pgprot_t phys_mem_access_prot(struct file *file, unsigned long pfn, |
487 | unsigned long size, pgprot_t vma_prot) | |
488 | { | |
489 | return vma_prot; | |
490 | } | |
491 | ||
d092633b IM |
492 | #ifdef CONFIG_STRICT_DEVMEM |
493 | /* This check is done in drivers/char/mem.c in case of STRICT_DEVMEM*/ | |
0124cecf VP |
494 | static inline int range_is_allowed(unsigned long pfn, unsigned long size) |
495 | { | |
496 | return 1; | |
497 | } | |
498 | #else | |
9e41bff2 | 499 | /* This check is needed to avoid cache aliasing when PAT is enabled */ |
0124cecf VP |
500 | static inline int range_is_allowed(unsigned long pfn, unsigned long size) |
501 | { | |
502 | u64 from = ((u64)pfn) << PAGE_SHIFT; | |
503 | u64 to = from + size; | |
504 | u64 cursor = from; | |
505 | ||
9e41bff2 RT |
506 | if (!pat_enabled) |
507 | return 1; | |
508 | ||
0124cecf VP |
509 | while (cursor < to) { |
510 | if (!devmem_is_allowed(pfn)) { | |
365811d6 BH |
511 | printk(KERN_INFO "Program %s tried to access /dev/mem between [mem %#010Lx-%#010Lx]\n", |
512 | current->comm, from, to - 1); | |
0124cecf VP |
513 | return 0; |
514 | } | |
515 | cursor += PAGE_SIZE; | |
516 | pfn++; | |
517 | } | |
518 | return 1; | |
519 | } | |
d092633b | 520 | #endif /* CONFIG_STRICT_DEVMEM */ |
0124cecf | 521 | |
f0970c13 | 522 | int phys_mem_access_prot_allowed(struct file *file, unsigned long pfn, |
523 | unsigned long size, pgprot_t *vma_prot) | |
524 | { | |
e00c8cc9 | 525 | enum page_cache_mode pcm = _PAGE_CACHE_MODE_WB; |
f0970c13 | 526 | |
0124cecf VP |
527 | if (!range_is_allowed(pfn, size)) |
528 | return 0; | |
529 | ||
6b2f3d1f | 530 | if (file->f_flags & O_DSYNC) |
e00c8cc9 | 531 | pcm = _PAGE_CACHE_MODE_UC_MINUS; |
f0970c13 | 532 | |
533 | #ifdef CONFIG_X86_32 | |
534 | /* | |
535 | * On the PPro and successors, the MTRRs are used to set | |
536 | * memory types for physical addresses outside main memory, | |
537 | * so blindly setting UC or PWT on those pages is wrong. | |
538 | * For Pentiums and earlier, the surround logic should disable | |
539 | * caching for the high addresses through the KEN pin, but | |
540 | * we maintain the tradition of paranoia in this code. | |
541 | */ | |
499f8f84 | 542 | if (!pat_enabled && |
cd7a4e93 AH |
543 | !(boot_cpu_has(X86_FEATURE_MTRR) || |
544 | boot_cpu_has(X86_FEATURE_K6_MTRR) || | |
545 | boot_cpu_has(X86_FEATURE_CYRIX_ARR) || | |
546 | boot_cpu_has(X86_FEATURE_CENTAUR_MCR)) && | |
547 | (pfn << PAGE_SHIFT) >= __pa(high_memory)) { | |
e00c8cc9 | 548 | pcm = _PAGE_CACHE_MODE_UC; |
f0970c13 | 549 | } |
550 | #endif | |
551 | ||
e7f260a2 | 552 | *vma_prot = __pgprot((pgprot_val(*vma_prot) & ~_PAGE_CACHE_MASK) | |
e00c8cc9 | 553 | cachemode2protval(pcm)); |
f0970c13 | 554 | return 1; |
555 | } | |
e7f260a2 | 556 | |
7880f746 VP |
557 | /* |
558 | * Change the memory type for the physial address range in kernel identity | |
559 | * mapping space if that range is a part of identity map. | |
560 | */ | |
b14097bd JG |
561 | int kernel_map_sync_memtype(u64 base, unsigned long size, |
562 | enum page_cache_mode pcm) | |
7880f746 VP |
563 | { |
564 | unsigned long id_sz; | |
565 | ||
a25b9316 | 566 | if (base > __pa(high_memory-1)) |
7880f746 VP |
567 | return 0; |
568 | ||
60f583d5 DH |
569 | /* |
570 | * some areas in the middle of the kernel identity range | |
571 | * are not mapped, like the PCI space. | |
572 | */ | |
573 | if (!page_is_ram(base >> PAGE_SHIFT)) | |
574 | return 0; | |
575 | ||
a25b9316 | 576 | id_sz = (__pa(high_memory-1) <= base + size) ? |
7880f746 VP |
577 | __pa(high_memory) - base : |
578 | size; | |
579 | ||
b14097bd | 580 | if (ioremap_change_attr((unsigned long)__va(base), id_sz, pcm) < 0) { |
365811d6 BH |
581 | printk(KERN_INFO "%s:%d ioremap_change_attr failed %s " |
582 | "for [mem %#010Lx-%#010Lx]\n", | |
7880f746 | 583 | current->comm, current->pid, |
e00c8cc9 | 584 | cattr_name(pcm), |
365811d6 | 585 | base, (unsigned long long)(base + size-1)); |
7880f746 VP |
586 | return -EINVAL; |
587 | } | |
588 | return 0; | |
589 | } | |
590 | ||
5899329b | 591 | /* |
592 | * Internal interface to reserve a range of physical memory with prot. | |
593 | * Reserved non RAM regions only and after successful reserve_memtype, | |
594 | * this func also keeps identity mapping (if any) in sync with this new prot. | |
595 | */ | |
cdecff68 | 596 | static int reserve_pfn_range(u64 paddr, unsigned long size, pgprot_t *vma_prot, |
597 | int strict_prot) | |
5899329b | 598 | { |
599 | int is_ram = 0; | |
7880f746 | 600 | int ret; |
e00c8cc9 JG |
601 | enum page_cache_mode want_pcm = pgprot2cachemode(*vma_prot); |
602 | enum page_cache_mode pcm = want_pcm; | |
5899329b | 603 | |
be03d9e8 | 604 | is_ram = pat_pagerange_is_ram(paddr, paddr + size); |
5899329b | 605 | |
be03d9e8 | 606 | /* |
d886c73c VP |
607 | * reserve_pfn_range() for RAM pages. We do not refcount to keep |
608 | * track of number of mappings of RAM pages. We can assert that | |
609 | * the type requested matches the type of first page in the range. | |
be03d9e8 | 610 | */ |
d886c73c VP |
611 | if (is_ram) { |
612 | if (!pat_enabled) | |
613 | return 0; | |
614 | ||
e00c8cc9 JG |
615 | pcm = lookup_memtype(paddr); |
616 | if (want_pcm != pcm) { | |
365811d6 | 617 | printk(KERN_WARNING "%s:%d map pfn RAM range req %s for [mem %#010Lx-%#010Lx], got %s\n", |
d886c73c | 618 | current->comm, current->pid, |
e00c8cc9 | 619 | cattr_name(want_pcm), |
d886c73c | 620 | (unsigned long long)paddr, |
365811d6 | 621 | (unsigned long long)(paddr + size - 1), |
e00c8cc9 | 622 | cattr_name(pcm)); |
d886c73c | 623 | *vma_prot = __pgprot((pgprot_val(*vma_prot) & |
e00c8cc9 JG |
624 | (~_PAGE_CACHE_MASK)) | |
625 | cachemode2protval(pcm)); | |
d886c73c | 626 | } |
4bb9c5c0 | 627 | return 0; |
d886c73c | 628 | } |
5899329b | 629 | |
e00c8cc9 | 630 | ret = reserve_memtype(paddr, paddr + size, want_pcm, &pcm); |
5899329b | 631 | if (ret) |
632 | return ret; | |
633 | ||
e00c8cc9 | 634 | if (pcm != want_pcm) { |
1adcaafe | 635 | if (strict_prot || |
e00c8cc9 | 636 | !is_new_memtype_allowed(paddr, size, want_pcm, pcm)) { |
cdecff68 | 637 | free_memtype(paddr, paddr + size); |
638 | printk(KERN_ERR "%s:%d map pfn expected mapping type %s" | |
365811d6 | 639 | " for [mem %#010Lx-%#010Lx], got %s\n", |
cdecff68 | 640 | current->comm, current->pid, |
e00c8cc9 | 641 | cattr_name(want_pcm), |
cdecff68 | 642 | (unsigned long long)paddr, |
365811d6 | 643 | (unsigned long long)(paddr + size - 1), |
e00c8cc9 | 644 | cattr_name(pcm)); |
cdecff68 | 645 | return -EINVAL; |
646 | } | |
647 | /* | |
648 | * We allow returning different type than the one requested in | |
649 | * non strict case. | |
650 | */ | |
651 | *vma_prot = __pgprot((pgprot_val(*vma_prot) & | |
652 | (~_PAGE_CACHE_MASK)) | | |
e00c8cc9 | 653 | cachemode2protval(pcm)); |
5899329b | 654 | } |
655 | ||
e00c8cc9 | 656 | if (kernel_map_sync_memtype(paddr, size, pcm) < 0) { |
5899329b | 657 | free_memtype(paddr, paddr + size); |
5899329b | 658 | return -EINVAL; |
659 | } | |
660 | return 0; | |
661 | } | |
662 | ||
663 | /* | |
664 | * Internal interface to free a range of physical memory. | |
665 | * Frees non RAM regions only. | |
666 | */ | |
667 | static void free_pfn_range(u64 paddr, unsigned long size) | |
668 | { | |
669 | int is_ram; | |
670 | ||
be03d9e8 | 671 | is_ram = pat_pagerange_is_ram(paddr, paddr + size); |
5899329b | 672 | if (is_ram == 0) |
673 | free_memtype(paddr, paddr + size); | |
674 | } | |
675 | ||
676 | /* | |
5180da41 | 677 | * track_pfn_copy is called when vma that is covering the pfnmap gets |
5899329b | 678 | * copied through copy_page_range(). |
679 | * | |
680 | * If the vma has a linear pfn mapping for the entire range, we get the prot | |
681 | * from pte and reserve the entire vma range with single reserve_pfn_range call. | |
5899329b | 682 | */ |
5180da41 | 683 | int track_pfn_copy(struct vm_area_struct *vma) |
5899329b | 684 | { |
c1c15b65 | 685 | resource_size_t paddr; |
982d789a | 686 | unsigned long prot; |
4b065046 | 687 | unsigned long vma_size = vma->vm_end - vma->vm_start; |
cdecff68 | 688 | pgprot_t pgprot; |
5899329b | 689 | |
b3b9c293 | 690 | if (vma->vm_flags & VM_PAT) { |
5899329b | 691 | /* |
982d789a | 692 | * reserve the whole chunk covered by vma. We need the |
693 | * starting address and protection from pte. | |
5899329b | 694 | */ |
4b065046 | 695 | if (follow_phys(vma, vma->vm_start, 0, &prot, &paddr)) { |
5899329b | 696 | WARN_ON_ONCE(1); |
982d789a | 697 | return -EINVAL; |
5899329b | 698 | } |
cdecff68 | 699 | pgprot = __pgprot(prot); |
700 | return reserve_pfn_range(paddr, vma_size, &pgprot, 1); | |
5899329b | 701 | } |
702 | ||
5899329b | 703 | return 0; |
5899329b | 704 | } |
705 | ||
706 | /* | |
5899329b | 707 | * prot is passed in as a parameter for the new mapping. If the vma has a |
708 | * linear pfn mapping for the entire range reserve the entire vma range with | |
709 | * single reserve_pfn_range call. | |
5899329b | 710 | */ |
5180da41 | 711 | int track_pfn_remap(struct vm_area_struct *vma, pgprot_t *prot, |
b3b9c293 | 712 | unsigned long pfn, unsigned long addr, unsigned long size) |
5899329b | 713 | { |
b1a86e15 | 714 | resource_size_t paddr = (resource_size_t)pfn << PAGE_SHIFT; |
2a374698 | 715 | enum page_cache_mode pcm; |
5899329b | 716 | |
b1a86e15 | 717 | /* reserve the whole chunk starting from paddr */ |
b3b9c293 KK |
718 | if (addr == vma->vm_start && size == (vma->vm_end - vma->vm_start)) { |
719 | int ret; | |
720 | ||
721 | ret = reserve_pfn_range(paddr, size, prot, 0); | |
722 | if (!ret) | |
723 | vma->vm_flags |= VM_PAT; | |
724 | return ret; | |
725 | } | |
5899329b | 726 | |
10876376 VP |
727 | if (!pat_enabled) |
728 | return 0; | |
729 | ||
5180da41 SS |
730 | /* |
731 | * For anything smaller than the vma size we set prot based on the | |
732 | * lookup. | |
733 | */ | |
2a374698 | 734 | pcm = lookup_memtype(paddr); |
5180da41 SS |
735 | |
736 | /* Check memtype for the remaining pages */ | |
737 | while (size > PAGE_SIZE) { | |
738 | size -= PAGE_SIZE; | |
739 | paddr += PAGE_SIZE; | |
2a374698 | 740 | if (pcm != lookup_memtype(paddr)) |
5180da41 SS |
741 | return -EINVAL; |
742 | } | |
743 | ||
744 | *prot = __pgprot((pgprot_val(vma->vm_page_prot) & (~_PAGE_CACHE_MASK)) | | |
2a374698 | 745 | cachemode2protval(pcm)); |
5180da41 SS |
746 | |
747 | return 0; | |
748 | } | |
749 | ||
750 | int track_pfn_insert(struct vm_area_struct *vma, pgprot_t *prot, | |
751 | unsigned long pfn) | |
752 | { | |
2a374698 | 753 | enum page_cache_mode pcm; |
5180da41 SS |
754 | |
755 | if (!pat_enabled) | |
756 | return 0; | |
757 | ||
758 | /* Set prot based on lookup */ | |
2a374698 | 759 | pcm = lookup_memtype((resource_size_t)pfn << PAGE_SHIFT); |
10876376 | 760 | *prot = __pgprot((pgprot_val(vma->vm_page_prot) & (~_PAGE_CACHE_MASK)) | |
2a374698 | 761 | cachemode2protval(pcm)); |
10876376 | 762 | |
5899329b | 763 | return 0; |
5899329b | 764 | } |
765 | ||
766 | /* | |
5180da41 | 767 | * untrack_pfn is called while unmapping a pfnmap for a region. |
5899329b | 768 | * untrack can be called for a specific region indicated by pfn and size or |
b1a86e15 | 769 | * can be for the entire vma (in which case pfn, size are zero). |
5899329b | 770 | */ |
5180da41 SS |
771 | void untrack_pfn(struct vm_area_struct *vma, unsigned long pfn, |
772 | unsigned long size) | |
5899329b | 773 | { |
c1c15b65 | 774 | resource_size_t paddr; |
b1a86e15 | 775 | unsigned long prot; |
5899329b | 776 | |
b3b9c293 | 777 | if (!(vma->vm_flags & VM_PAT)) |
5899329b | 778 | return; |
b1a86e15 SS |
779 | |
780 | /* free the chunk starting from pfn or the whole chunk */ | |
781 | paddr = (resource_size_t)pfn << PAGE_SHIFT; | |
782 | if (!paddr && !size) { | |
783 | if (follow_phys(vma, vma->vm_start, 0, &prot, &paddr)) { | |
784 | WARN_ON_ONCE(1); | |
785 | return; | |
786 | } | |
787 | ||
788 | size = vma->vm_end - vma->vm_start; | |
5899329b | 789 | } |
b1a86e15 | 790 | free_pfn_range(paddr, size); |
b3b9c293 | 791 | vma->vm_flags &= ~VM_PAT; |
5899329b | 792 | } |
793 | ||
2520bd31 | 794 | pgprot_t pgprot_writecombine(pgprot_t prot) |
795 | { | |
796 | if (pat_enabled) | |
e00c8cc9 JG |
797 | return __pgprot(pgprot_val(prot) | |
798 | cachemode2protval(_PAGE_CACHE_MODE_WC)); | |
2520bd31 | 799 | else |
800 | return pgprot_noncached(prot); | |
801 | } | |
92b9af9e | 802 | EXPORT_SYMBOL_GPL(pgprot_writecombine); |
2520bd31 | 803 | |
012f09e7 | 804 | #if defined(CONFIG_DEBUG_FS) && defined(CONFIG_X86_PAT) |
fec0962e | 805 | |
fec0962e | 806 | static struct memtype *memtype_get_idx(loff_t pos) |
807 | { | |
be5a0c12 | 808 | struct memtype *print_entry; |
809 | int ret; | |
fec0962e | 810 | |
be5a0c12 | 811 | print_entry = kzalloc(sizeof(struct memtype), GFP_KERNEL); |
fec0962e | 812 | if (!print_entry) |
813 | return NULL; | |
814 | ||
815 | spin_lock(&memtype_lock); | |
9e41a49a | 816 | ret = rbt_memtype_copy_nth_element(print_entry, pos); |
fec0962e | 817 | spin_unlock(&memtype_lock); |
ad2cde16 | 818 | |
be5a0c12 | 819 | if (!ret) { |
820 | return print_entry; | |
821 | } else { | |
822 | kfree(print_entry); | |
823 | return NULL; | |
824 | } | |
fec0962e | 825 | } |
826 | ||
827 | static void *memtype_seq_start(struct seq_file *seq, loff_t *pos) | |
828 | { | |
829 | if (*pos == 0) { | |
830 | ++*pos; | |
831 | seq_printf(seq, "PAT memtype list:\n"); | |
832 | } | |
833 | ||
834 | return memtype_get_idx(*pos); | |
835 | } | |
836 | ||
837 | static void *memtype_seq_next(struct seq_file *seq, void *v, loff_t *pos) | |
838 | { | |
839 | ++*pos; | |
840 | return memtype_get_idx(*pos); | |
841 | } | |
842 | ||
843 | static void memtype_seq_stop(struct seq_file *seq, void *v) | |
844 | { | |
845 | } | |
846 | ||
847 | static int memtype_seq_show(struct seq_file *seq, void *v) | |
848 | { | |
849 | struct memtype *print_entry = (struct memtype *)v; | |
850 | ||
851 | seq_printf(seq, "%s @ 0x%Lx-0x%Lx\n", cattr_name(print_entry->type), | |
852 | print_entry->start, print_entry->end); | |
853 | kfree(print_entry); | |
ad2cde16 | 854 | |
fec0962e | 855 | return 0; |
856 | } | |
857 | ||
d535e431 | 858 | static const struct seq_operations memtype_seq_ops = { |
fec0962e | 859 | .start = memtype_seq_start, |
860 | .next = memtype_seq_next, | |
861 | .stop = memtype_seq_stop, | |
862 | .show = memtype_seq_show, | |
863 | }; | |
864 | ||
865 | static int memtype_seq_open(struct inode *inode, struct file *file) | |
866 | { | |
867 | return seq_open(file, &memtype_seq_ops); | |
868 | } | |
869 | ||
870 | static const struct file_operations memtype_fops = { | |
871 | .open = memtype_seq_open, | |
872 | .read = seq_read, | |
873 | .llseek = seq_lseek, | |
874 | .release = seq_release, | |
875 | }; | |
876 | ||
877 | static int __init pat_memtype_list_init(void) | |
878 | { | |
dd4377b0 XF |
879 | if (pat_enabled) { |
880 | debugfs_create_file("pat_memtype_list", S_IRUSR, | |
881 | arch_debugfs_dir, NULL, &memtype_fops); | |
882 | } | |
fec0962e | 883 | return 0; |
884 | } | |
885 | ||
886 | late_initcall(pat_memtype_list_init); | |
887 | ||
012f09e7 | 888 | #endif /* CONFIG_DEBUG_FS && CONFIG_X86_PAT */ |