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> |
14 | #include <linux/gfp.h> | |
ad2cde16 | 15 | #include <linux/mm.h> |
2e5d9c85 | 16 | #include <linux/fs.h> |
17 | ||
ad2cde16 | 18 | #include <asm/cacheflush.h> |
2e5d9c85 | 19 | #include <asm/processor.h> |
ad2cde16 | 20 | #include <asm/tlbflush.h> |
2e5d9c85 | 21 | #include <asm/pgtable.h> |
2e5d9c85 | 22 | #include <asm/fcntl.h> |
ad2cde16 | 23 | #include <asm/e820.h> |
2e5d9c85 | 24 | #include <asm/mtrr.h> |
ad2cde16 IM |
25 | #include <asm/page.h> |
26 | #include <asm/msr.h> | |
27 | #include <asm/pat.h> | |
e7f260a2 | 28 | #include <asm/io.h> |
2e5d9c85 | 29 | |
8d4a4300 | 30 | #ifdef CONFIG_X86_PAT |
499f8f84 | 31 | int __read_mostly pat_enabled = 1; |
2e5d9c85 | 32 | |
31f4d870 | 33 | void __cpuinit pat_disable(char *reason) |
2e5d9c85 | 34 | { |
499f8f84 | 35 | pat_enabled = 0; |
8d4a4300 | 36 | printk(KERN_INFO "%s\n", reason); |
2e5d9c85 | 37 | } |
2e5d9c85 | 38 | |
be524fb9 | 39 | static int __init nopat(char *str) |
2e5d9c85 | 40 | { |
8d4a4300 | 41 | pat_disable("PAT support disabled."); |
2e5d9c85 | 42 | return 0; |
43 | } | |
8d4a4300 TG |
44 | early_param("nopat", nopat); |
45 | #endif | |
46 | ||
77b52b4c VP |
47 | |
48 | static int debug_enable; | |
ad2cde16 | 49 | |
77b52b4c VP |
50 | static int __init pat_debug_setup(char *str) |
51 | { | |
52 | debug_enable = 1; | |
53 | return 0; | |
54 | } | |
55 | __setup("debugpat", pat_debug_setup); | |
56 | ||
57 | #define dprintk(fmt, arg...) \ | |
58 | do { if (debug_enable) printk(KERN_INFO fmt, ##arg); } while (0) | |
59 | ||
60 | ||
8d4a4300 | 61 | static u64 __read_mostly boot_pat_state; |
2e5d9c85 | 62 | |
63 | enum { | |
64 | PAT_UC = 0, /* uncached */ | |
65 | PAT_WC = 1, /* Write combining */ | |
66 | PAT_WT = 4, /* Write Through */ | |
67 | PAT_WP = 5, /* Write Protected */ | |
68 | PAT_WB = 6, /* Write Back (default) */ | |
69 | PAT_UC_MINUS = 7, /* UC, but can be overriden by MTRR */ | |
70 | }; | |
71 | ||
cd7a4e93 | 72 | #define PAT(x, y) ((u64)PAT_ ## y << ((x)*8)) |
2e5d9c85 | 73 | |
74 | void pat_init(void) | |
75 | { | |
76 | u64 pat; | |
77 | ||
499f8f84 | 78 | if (!pat_enabled) |
2e5d9c85 | 79 | return; |
80 | ||
8d4a4300 | 81 | /* Paranoia check. */ |
97cfab6a | 82 | if (!cpu_has_pat && boot_pat_state) { |
8d4a4300 | 83 | /* |
97cfab6a | 84 | * If this happens we are on a secondary CPU, but |
8d4a4300 TG |
85 | * switched to PAT on the boot CPU. We have no way to |
86 | * undo PAT. | |
97cfab6a AH |
87 | */ |
88 | printk(KERN_ERR "PAT enabled, " | |
89 | "but not supported by secondary CPU\n"); | |
90 | BUG(); | |
8d4a4300 | 91 | } |
2e5d9c85 | 92 | |
93 | /* Set PWT to Write-Combining. All other bits stay the same */ | |
94 | /* | |
95 | * PTE encoding used in Linux: | |
96 | * PAT | |
97 | * |PCD | |
98 | * ||PWT | |
99 | * ||| | |
100 | * 000 WB _PAGE_CACHE_WB | |
101 | * 001 WC _PAGE_CACHE_WC | |
102 | * 010 UC- _PAGE_CACHE_UC_MINUS | |
103 | * 011 UC _PAGE_CACHE_UC | |
104 | * PAT bit unused | |
105 | */ | |
cd7a4e93 AH |
106 | pat = PAT(0, WB) | PAT(1, WC) | PAT(2, UC_MINUS) | PAT(3, UC) | |
107 | PAT(4, WB) | PAT(5, WC) | PAT(6, UC_MINUS) | PAT(7, UC); | |
2e5d9c85 | 108 | |
109 | /* Boot CPU check */ | |
8d4a4300 | 110 | if (!boot_pat_state) |
2e5d9c85 | 111 | rdmsrl(MSR_IA32_CR_PAT, boot_pat_state); |
2e5d9c85 | 112 | |
113 | wrmsrl(MSR_IA32_CR_PAT, pat); | |
114 | printk(KERN_INFO "x86 PAT enabled: cpu %d, old 0x%Lx, new 0x%Lx\n", | |
115 | smp_processor_id(), boot_pat_state, pat); | |
116 | } | |
117 | ||
118 | #undef PAT | |
119 | ||
120 | static char *cattr_name(unsigned long flags) | |
121 | { | |
122 | switch (flags & _PAGE_CACHE_MASK) { | |
cd7a4e93 AH |
123 | case _PAGE_CACHE_UC: return "uncached"; |
124 | case _PAGE_CACHE_UC_MINUS: return "uncached-minus"; | |
125 | case _PAGE_CACHE_WB: return "write-back"; | |
126 | case _PAGE_CACHE_WC: return "write-combining"; | |
127 | default: return "broken"; | |
2e5d9c85 | 128 | } |
129 | } | |
130 | ||
131 | /* | |
132 | * The global memtype list keeps track of memory type for specific | |
133 | * physical memory areas. Conflicting memory types in different | |
134 | * mappings can cause CPU cache corruption. To avoid this we keep track. | |
135 | * | |
136 | * The list is sorted based on starting address and can contain multiple | |
137 | * entries for each address (this allows reference counting for overlapping | |
138 | * areas). All the aliases have the same cache attributes of course. | |
139 | * Zero attributes are represented as holes. | |
140 | * | |
141 | * Currently the data structure is a list because the number of mappings | |
142 | * are expected to be relatively small. If this should be a problem | |
143 | * it could be changed to a rbtree or similar. | |
144 | * | |
145 | * memtype_lock protects the whole list. | |
146 | */ | |
147 | ||
148 | struct memtype { | |
ad2cde16 IM |
149 | u64 start; |
150 | u64 end; | |
151 | unsigned long type; | |
152 | struct list_head nd; | |
2e5d9c85 | 153 | }; |
154 | ||
155 | static LIST_HEAD(memtype_list); | |
ad2cde16 | 156 | static DEFINE_SPINLOCK(memtype_lock); /* protects memtype list */ |
2e5d9c85 | 157 | |
158 | /* | |
159 | * Does intersection of PAT memory type and MTRR memory type and returns | |
160 | * the resulting memory type as PAT understands it. | |
161 | * (Type in pat and mtrr will not have same value) | |
162 | * The intersection is based on "Effective Memory Type" tables in IA-32 | |
163 | * SDM vol 3a | |
164 | */ | |
6cf514fc | 165 | static unsigned long pat_x_mtrr_type(u64 start, u64 end, unsigned long req_type) |
2e5d9c85 | 166 | { |
c26421d0 VP |
167 | /* |
168 | * Look for MTRR hint to get the effective type in case where PAT | |
169 | * request is for WB. | |
170 | */ | |
dd0c7c49 AH |
171 | if (req_type == _PAGE_CACHE_WB) { |
172 | u8 mtrr_type; | |
173 | ||
174 | mtrr_type = mtrr_type_lookup(start, end); | |
175 | if (mtrr_type == MTRR_TYPE_UNCACHABLE) | |
176 | return _PAGE_CACHE_UC; | |
177 | if (mtrr_type == MTRR_TYPE_WRCOMB) | |
178 | return _PAGE_CACHE_WC; | |
179 | } | |
180 | ||
181 | return req_type; | |
2e5d9c85 | 182 | } |
183 | ||
ad2cde16 IM |
184 | static int |
185 | chk_conflict(struct memtype *new, struct memtype *entry, unsigned long *type) | |
64fe44c3 AH |
186 | { |
187 | if (new->type != entry->type) { | |
188 | if (type) { | |
189 | new->type = entry->type; | |
190 | *type = entry->type; | |
191 | } else | |
192 | goto conflict; | |
193 | } | |
194 | ||
195 | /* check overlaps with more than one entry in the list */ | |
196 | list_for_each_entry_continue(entry, &memtype_list, nd) { | |
197 | if (new->end <= entry->start) | |
198 | break; | |
199 | else if (new->type != entry->type) | |
200 | goto conflict; | |
201 | } | |
202 | return 0; | |
203 | ||
204 | conflict: | |
205 | printk(KERN_INFO "%s:%d conflicting memory types " | |
206 | "%Lx-%Lx %s<->%s\n", current->comm, current->pid, new->start, | |
207 | new->end, cattr_name(new->type), cattr_name(entry->type)); | |
208 | return -EBUSY; | |
209 | } | |
210 | ||
80c5e73d VP |
211 | static struct memtype *cached_entry; |
212 | static u64 cached_start; | |
213 | ||
be03d9e8 SS |
214 | static int pat_pagerange_is_ram(unsigned long start, unsigned long end) |
215 | { | |
216 | int ram_page = 0, not_rampage = 0; | |
217 | unsigned long page_nr; | |
218 | ||
219 | for (page_nr = (start >> PAGE_SHIFT); page_nr < (end >> PAGE_SHIFT); | |
220 | ++page_nr) { | |
221 | /* | |
222 | * For legacy reasons, physical address range in the legacy ISA | |
223 | * region is tracked as non-RAM. This will allow users of | |
224 | * /dev/mem to map portions of legacy ISA region, even when | |
225 | * some of those portions are listed(or not even listed) with | |
226 | * different e820 types(RAM/reserved/..) | |
227 | */ | |
228 | if (page_nr >= (ISA_END_ADDRESS >> PAGE_SHIFT) && | |
229 | page_is_ram(page_nr)) | |
230 | ram_page = 1; | |
231 | else | |
232 | not_rampage = 1; | |
233 | ||
234 | if (ram_page == not_rampage) | |
235 | return -1; | |
236 | } | |
237 | ||
238 | return ram_page; | |
239 | } | |
240 | ||
9542ada8 SS |
241 | /* |
242 | * For RAM pages, mark the pages as non WB memory type using | |
243 | * PageNonWB (PG_arch_1). We allow only one set_memory_uc() or | |
244 | * set_memory_wc() on a RAM page at a time before marking it as WB again. | |
245 | * This is ok, because only one driver will be owning the page and | |
246 | * doing set_memory_*() calls. | |
247 | * | |
248 | * For now, we use PageNonWB to track that the RAM page is being mapped | |
249 | * as non WB. In future, we will have to use one more flag | |
250 | * (or some other mechanism in page_struct) to distinguish between | |
251 | * UC and WC mapping. | |
252 | */ | |
253 | static int reserve_ram_pages_type(u64 start, u64 end, unsigned long req_type, | |
ad2cde16 | 254 | unsigned long *new_type) |
9542ada8 SS |
255 | { |
256 | struct page *page; | |
257 | u64 pfn, end_pfn; | |
258 | ||
259 | for (pfn = (start >> PAGE_SHIFT); pfn < (end >> PAGE_SHIFT); ++pfn) { | |
260 | page = pfn_to_page(pfn); | |
261 | if (page_mapped(page) || PageNonWB(page)) | |
262 | goto out; | |
263 | ||
264 | SetPageNonWB(page); | |
265 | } | |
266 | return 0; | |
267 | ||
268 | out: | |
269 | end_pfn = pfn; | |
270 | for (pfn = (start >> PAGE_SHIFT); pfn < end_pfn; ++pfn) { | |
271 | page = pfn_to_page(pfn); | |
272 | ClearPageNonWB(page); | |
273 | } | |
274 | ||
275 | return -EINVAL; | |
276 | } | |
277 | ||
278 | static int free_ram_pages_type(u64 start, u64 end) | |
279 | { | |
280 | struct page *page; | |
281 | u64 pfn, end_pfn; | |
282 | ||
283 | for (pfn = (start >> PAGE_SHIFT); pfn < (end >> PAGE_SHIFT); ++pfn) { | |
284 | page = pfn_to_page(pfn); | |
285 | if (page_mapped(page) || !PageNonWB(page)) | |
286 | goto out; | |
287 | ||
288 | ClearPageNonWB(page); | |
289 | } | |
290 | return 0; | |
291 | ||
292 | out: | |
293 | end_pfn = pfn; | |
294 | for (pfn = (start >> PAGE_SHIFT); pfn < end_pfn; ++pfn) { | |
295 | page = pfn_to_page(pfn); | |
296 | SetPageNonWB(page); | |
297 | } | |
298 | return -EINVAL; | |
299 | } | |
300 | ||
e7f260a2 | 301 | /* |
302 | * req_type typically has one of the: | |
303 | * - _PAGE_CACHE_WB | |
304 | * - _PAGE_CACHE_WC | |
305 | * - _PAGE_CACHE_UC_MINUS | |
306 | * - _PAGE_CACHE_UC | |
307 | * | |
308 | * req_type will have a special case value '-1', when requester want to inherit | |
309 | * the memory type from mtrr (if WB), existing PAT, defaulting to UC_MINUS. | |
310 | * | |
ac97991e AH |
311 | * If new_type is NULL, function will return an error if it cannot reserve the |
312 | * region with req_type. If new_type is non-NULL, function will return | |
313 | * available type in new_type in case of no error. In case of any error | |
e7f260a2 | 314 | * it will return a negative return value. |
315 | */ | |
2e5d9c85 | 316 | int reserve_memtype(u64 start, u64 end, unsigned long req_type, |
ad2cde16 | 317 | unsigned long *new_type) |
2e5d9c85 | 318 | { |
ac97991e | 319 | struct memtype *new, *entry; |
2e5d9c85 | 320 | unsigned long actual_type; |
f6887264 | 321 | struct list_head *where; |
9542ada8 | 322 | int is_range_ram; |
ad2cde16 | 323 | int err = 0; |
2e5d9c85 | 324 | |
ad2cde16 | 325 | BUG_ON(start >= end); /* end is exclusive */ |
69e26be9 | 326 | |
499f8f84 | 327 | if (!pat_enabled) { |
e7f260a2 | 328 | /* This is identical to page table setting without PAT */ |
ac97991e AH |
329 | if (new_type) { |
330 | if (req_type == -1) | |
331 | *new_type = _PAGE_CACHE_WB; | |
332 | else | |
333 | *new_type = req_type & _PAGE_CACHE_MASK; | |
e7f260a2 | 334 | } |
2e5d9c85 | 335 | return 0; |
336 | } | |
337 | ||
338 | /* Low ISA region is always mapped WB in page table. No need to track */ | |
bcc643dc | 339 | if (is_ISA_range(start, end - 1)) { |
ac97991e AH |
340 | if (new_type) |
341 | *new_type = _PAGE_CACHE_WB; | |
2e5d9c85 | 342 | return 0; |
343 | } | |
344 | ||
e7f260a2 | 345 | if (req_type == -1) { |
346 | /* | |
c26421d0 VP |
347 | * Call mtrr_lookup to get the type hint. This is an |
348 | * optimization for /dev/mem mmap'ers into WB memory (BIOS | |
349 | * tools and ACPI tools). Use WB request for WB memory and use | |
350 | * UC_MINUS otherwise. | |
e7f260a2 | 351 | */ |
352 | u8 mtrr_type = mtrr_type_lookup(start, end); | |
e7f260a2 | 353 | |
69e26be9 | 354 | if (mtrr_type == MTRR_TYPE_WRBACK) |
e7f260a2 | 355 | actual_type = _PAGE_CACHE_WB; |
69e26be9 | 356 | else |
e7f260a2 | 357 | actual_type = _PAGE_CACHE_UC_MINUS; |
ad2cde16 | 358 | } else { |
69e26be9 AH |
359 | actual_type = pat_x_mtrr_type(start, end, |
360 | req_type & _PAGE_CACHE_MASK); | |
ad2cde16 | 361 | } |
2e5d9c85 | 362 | |
95971342 SS |
363 | if (new_type) |
364 | *new_type = actual_type; | |
365 | ||
be03d9e8 SS |
366 | is_range_ram = pat_pagerange_is_ram(start, end); |
367 | if (is_range_ram == 1) | |
368 | return reserve_ram_pages_type(start, end, req_type, | |
369 | new_type); | |
370 | else if (is_range_ram < 0) | |
371 | return -EINVAL; | |
9542ada8 | 372 | |
ac97991e AH |
373 | new = kmalloc(sizeof(struct memtype), GFP_KERNEL); |
374 | if (!new) | |
2e5d9c85 | 375 | return -ENOMEM; |
376 | ||
ad2cde16 IM |
377 | new->start = start; |
378 | new->end = end; | |
379 | new->type = actual_type; | |
2e5d9c85 | 380 | |
2e5d9c85 | 381 | spin_lock(&memtype_lock); |
382 | ||
80c5e73d VP |
383 | if (cached_entry && start >= cached_start) |
384 | entry = cached_entry; | |
385 | else | |
386 | entry = list_entry(&memtype_list, struct memtype, nd); | |
387 | ||
2e5d9c85 | 388 | /* Search for existing mapping that overlaps the current range */ |
f6887264 | 389 | where = NULL; |
80c5e73d | 390 | list_for_each_entry_continue(entry, &memtype_list, nd) { |
33af9039 | 391 | if (end <= entry->start) { |
f6887264 | 392 | where = entry->nd.prev; |
80c5e73d | 393 | cached_entry = list_entry(where, struct memtype, nd); |
2e5d9c85 | 394 | break; |
33af9039 | 395 | } else if (start <= entry->start) { /* end > entry->start */ |
64fe44c3 | 396 | err = chk_conflict(new, entry, new_type); |
33af9039 AH |
397 | if (!err) { |
398 | dprintk("Overlap at 0x%Lx-0x%Lx\n", | |
399 | entry->start, entry->end); | |
400 | where = entry->nd.prev; | |
80c5e73d VP |
401 | cached_entry = list_entry(where, |
402 | struct memtype, nd); | |
2e5d9c85 | 403 | } |
2e5d9c85 | 404 | break; |
33af9039 | 405 | } else if (start < entry->end) { /* start > entry->start */ |
64fe44c3 | 406 | err = chk_conflict(new, entry, new_type); |
33af9039 AH |
407 | if (!err) { |
408 | dprintk("Overlap at 0x%Lx-0x%Lx\n", | |
409 | entry->start, entry->end); | |
80c5e73d VP |
410 | cached_entry = list_entry(entry->nd.prev, |
411 | struct memtype, nd); | |
412 | ||
413 | /* | |
414 | * Move to right position in the linked | |
415 | * list to add this new entry | |
416 | */ | |
417 | list_for_each_entry_continue(entry, | |
418 | &memtype_list, nd) { | |
419 | if (start <= entry->start) { | |
420 | where = entry->nd.prev; | |
421 | break; | |
422 | } | |
423 | } | |
2e5d9c85 | 424 | } |
2e5d9c85 | 425 | break; |
426 | } | |
427 | } | |
428 | ||
429 | if (err) { | |
3e9c83b3 AH |
430 | printk(KERN_INFO "reserve_memtype failed 0x%Lx-0x%Lx, " |
431 | "track %s, req %s\n", | |
432 | start, end, cattr_name(new->type), cattr_name(req_type)); | |
ac97991e | 433 | kfree(new); |
2e5d9c85 | 434 | spin_unlock(&memtype_lock); |
ad2cde16 | 435 | |
2e5d9c85 | 436 | return err; |
437 | } | |
438 | ||
80c5e73d VP |
439 | cached_start = start; |
440 | ||
f6887264 AH |
441 | if (where) |
442 | list_add(&new->nd, where); | |
443 | else | |
ac97991e | 444 | list_add_tail(&new->nd, &memtype_list); |
6997ab49 | 445 | |
2e5d9c85 | 446 | spin_unlock(&memtype_lock); |
3e9c83b3 AH |
447 | |
448 | dprintk("reserve_memtype added 0x%Lx-0x%Lx, track %s, req %s, ret %s\n", | |
449 | start, end, cattr_name(new->type), cattr_name(req_type), | |
450 | new_type ? cattr_name(*new_type) : "-"); | |
451 | ||
2e5d9c85 | 452 | return err; |
453 | } | |
454 | ||
455 | int free_memtype(u64 start, u64 end) | |
456 | { | |
ac97991e | 457 | struct memtype *entry; |
2e5d9c85 | 458 | int err = -EINVAL; |
9542ada8 | 459 | int is_range_ram; |
2e5d9c85 | 460 | |
69e26be9 | 461 | if (!pat_enabled) |
2e5d9c85 | 462 | return 0; |
2e5d9c85 | 463 | |
464 | /* Low ISA region is always mapped WB. No need to track */ | |
bcc643dc | 465 | if (is_ISA_range(start, end - 1)) |
2e5d9c85 | 466 | return 0; |
2e5d9c85 | 467 | |
be03d9e8 SS |
468 | is_range_ram = pat_pagerange_is_ram(start, end); |
469 | if (is_range_ram == 1) | |
470 | return free_ram_pages_type(start, end); | |
471 | else if (is_range_ram < 0) | |
472 | return -EINVAL; | |
9542ada8 | 473 | |
2e5d9c85 | 474 | spin_lock(&memtype_lock); |
ac97991e AH |
475 | list_for_each_entry(entry, &memtype_list, nd) { |
476 | if (entry->start == start && entry->end == end) { | |
80c5e73d VP |
477 | if (cached_entry == entry || cached_start == start) |
478 | cached_entry = NULL; | |
479 | ||
ac97991e AH |
480 | list_del(&entry->nd); |
481 | kfree(entry); | |
2e5d9c85 | 482 | err = 0; |
483 | break; | |
484 | } | |
485 | } | |
486 | spin_unlock(&memtype_lock); | |
487 | ||
488 | if (err) { | |
28eb559b | 489 | printk(KERN_INFO "%s:%d freeing invalid memtype %Lx-%Lx\n", |
2e5d9c85 | 490 | current->comm, current->pid, start, end); |
491 | } | |
6997ab49 | 492 | |
77b52b4c | 493 | dprintk("free_memtype request 0x%Lx-0x%Lx\n", start, end); |
ad2cde16 | 494 | |
2e5d9c85 | 495 | return err; |
496 | } | |
497 | ||
f0970c13 | 498 | |
f0970c13 | 499 | pgprot_t phys_mem_access_prot(struct file *file, unsigned long pfn, |
500 | unsigned long size, pgprot_t vma_prot) | |
501 | { | |
502 | return vma_prot; | |
503 | } | |
504 | ||
d092633b IM |
505 | #ifdef CONFIG_STRICT_DEVMEM |
506 | /* This check is done in drivers/char/mem.c in case of STRICT_DEVMEM*/ | |
0124cecf VP |
507 | static inline int range_is_allowed(unsigned long pfn, unsigned long size) |
508 | { | |
509 | return 1; | |
510 | } | |
511 | #else | |
9e41bff2 | 512 | /* This check is needed to avoid cache aliasing when PAT is enabled */ |
0124cecf VP |
513 | static inline int range_is_allowed(unsigned long pfn, unsigned long size) |
514 | { | |
515 | u64 from = ((u64)pfn) << PAGE_SHIFT; | |
516 | u64 to = from + size; | |
517 | u64 cursor = from; | |
518 | ||
9e41bff2 RT |
519 | if (!pat_enabled) |
520 | return 1; | |
521 | ||
0124cecf VP |
522 | while (cursor < to) { |
523 | if (!devmem_is_allowed(pfn)) { | |
524 | printk(KERN_INFO | |
525 | "Program %s tried to access /dev/mem between %Lx->%Lx.\n", | |
526 | current->comm, from, to); | |
527 | return 0; | |
528 | } | |
529 | cursor += PAGE_SIZE; | |
530 | pfn++; | |
531 | } | |
532 | return 1; | |
533 | } | |
d092633b | 534 | #endif /* CONFIG_STRICT_DEVMEM */ |
0124cecf | 535 | |
f0970c13 | 536 | int phys_mem_access_prot_allowed(struct file *file, unsigned long pfn, |
537 | unsigned long size, pgprot_t *vma_prot) | |
538 | { | |
e7f260a2 | 539 | u64 offset = ((u64) pfn) << PAGE_SHIFT; |
28df82eb | 540 | unsigned long flags = -1; |
e7f260a2 | 541 | int retval; |
f0970c13 | 542 | |
0124cecf VP |
543 | if (!range_is_allowed(pfn, size)) |
544 | return 0; | |
545 | ||
f0970c13 | 546 | if (file->f_flags & O_SYNC) { |
28df82eb | 547 | flags = _PAGE_CACHE_UC_MINUS; |
f0970c13 | 548 | } |
549 | ||
550 | #ifdef CONFIG_X86_32 | |
551 | /* | |
552 | * On the PPro and successors, the MTRRs are used to set | |
553 | * memory types for physical addresses outside main memory, | |
554 | * so blindly setting UC or PWT on those pages is wrong. | |
555 | * For Pentiums and earlier, the surround logic should disable | |
556 | * caching for the high addresses through the KEN pin, but | |
557 | * we maintain the tradition of paranoia in this code. | |
558 | */ | |
499f8f84 | 559 | if (!pat_enabled && |
cd7a4e93 AH |
560 | !(boot_cpu_has(X86_FEATURE_MTRR) || |
561 | boot_cpu_has(X86_FEATURE_K6_MTRR) || | |
562 | boot_cpu_has(X86_FEATURE_CYRIX_ARR) || | |
563 | boot_cpu_has(X86_FEATURE_CENTAUR_MCR)) && | |
564 | (pfn << PAGE_SHIFT) >= __pa(high_memory)) { | |
e7f260a2 | 565 | flags = _PAGE_CACHE_UC; |
f0970c13 | 566 | } |
567 | #endif | |
568 | ||
e7f260a2 | 569 | /* |
28df82eb | 570 | * With O_SYNC, we can only take UC_MINUS mapping. Fail if we cannot. |
571 | * | |
e7f260a2 | 572 | * Without O_SYNC, we want to get |
573 | * - WB for WB-able memory and no other conflicting mappings | |
574 | * - UC_MINUS for non-WB-able memory with no other conflicting mappings | |
575 | * - Inherit from confliting mappings otherwise | |
576 | */ | |
28df82eb | 577 | if (flags != -1) { |
e7f260a2 | 578 | retval = reserve_memtype(offset, offset + size, flags, NULL); |
579 | } else { | |
f022bfd5 | 580 | retval = reserve_memtype(offset, offset + size, -1, &flags); |
e7f260a2 | 581 | } |
582 | ||
583 | if (retval < 0) | |
584 | return 0; | |
585 | ||
b5db0e38 LT |
586 | if (((pfn < max_low_pfn_mapped) || |
587 | (pfn >= (1UL<<(32 - PAGE_SHIFT)) && pfn < max_pfn_mapped)) && | |
588 | ioremap_change_attr((unsigned long)__va(offset), size, flags) < 0) { | |
e7f260a2 | 589 | free_memtype(offset, offset + size); |
28eb559b | 590 | printk(KERN_INFO |
e7f260a2 | 591 | "%s:%d /dev/mem ioremap_change_attr failed %s for %Lx-%Lx\n", |
592 | current->comm, current->pid, | |
593 | cattr_name(flags), | |
afc85343 | 594 | offset, (unsigned long long)(offset + size)); |
e7f260a2 | 595 | return 0; |
596 | } | |
597 | ||
598 | *vma_prot = __pgprot((pgprot_val(*vma_prot) & ~_PAGE_CACHE_MASK) | | |
599 | flags); | |
f0970c13 | 600 | return 1; |
601 | } | |
e7f260a2 | 602 | |
603 | void map_devmem(unsigned long pfn, unsigned long size, pgprot_t vma_prot) | |
604 | { | |
ad2cde16 | 605 | unsigned long want_flags = (pgprot_val(vma_prot) & _PAGE_CACHE_MASK); |
e7f260a2 | 606 | u64 addr = (u64)pfn << PAGE_SHIFT; |
607 | unsigned long flags; | |
e7f260a2 | 608 | |
609 | reserve_memtype(addr, addr + size, want_flags, &flags); | |
610 | if (flags != want_flags) { | |
28eb559b | 611 | printk(KERN_INFO |
e7f260a2 | 612 | "%s:%d /dev/mem expected mapping type %s for %Lx-%Lx, got %s\n", |
613 | current->comm, current->pid, | |
614 | cattr_name(want_flags), | |
afc85343 | 615 | addr, (unsigned long long)(addr + size), |
e7f260a2 | 616 | cattr_name(flags)); |
617 | } | |
618 | } | |
619 | ||
620 | void unmap_devmem(unsigned long pfn, unsigned long size, pgprot_t vma_prot) | |
621 | { | |
622 | u64 addr = (u64)pfn << PAGE_SHIFT; | |
623 | ||
624 | free_memtype(addr, addr + size); | |
625 | } | |
626 | ||
5899329b | 627 | /* |
628 | * Internal interface to reserve a range of physical memory with prot. | |
629 | * Reserved non RAM regions only and after successful reserve_memtype, | |
630 | * this func also keeps identity mapping (if any) in sync with this new prot. | |
631 | */ | |
cdecff68 | 632 | static int reserve_pfn_range(u64 paddr, unsigned long size, pgprot_t *vma_prot, |
633 | int strict_prot) | |
5899329b | 634 | { |
635 | int is_ram = 0; | |
b5db0e38 | 636 | int id_sz, ret; |
5899329b | 637 | unsigned long flags; |
cdecff68 | 638 | unsigned long want_flags = (pgprot_val(*vma_prot) & _PAGE_CACHE_MASK); |
5899329b | 639 | |
be03d9e8 | 640 | is_ram = pat_pagerange_is_ram(paddr, paddr + size); |
5899329b | 641 | |
be03d9e8 SS |
642 | /* |
643 | * reserve_pfn_range() doesn't support RAM pages. | |
644 | */ | |
645 | if (is_ram != 0) | |
646 | return -EINVAL; | |
5899329b | 647 | |
648 | ret = reserve_memtype(paddr, paddr + size, want_flags, &flags); | |
649 | if (ret) | |
650 | return ret; | |
651 | ||
652 | if (flags != want_flags) { | |
cdecff68 | 653 | if (strict_prot || !is_new_memtype_allowed(want_flags, flags)) { |
654 | free_memtype(paddr, paddr + size); | |
655 | printk(KERN_ERR "%s:%d map pfn expected mapping type %s" | |
656 | " for %Lx-%Lx, got %s\n", | |
657 | current->comm, current->pid, | |
658 | cattr_name(want_flags), | |
659 | (unsigned long long)paddr, | |
660 | (unsigned long long)(paddr + size), | |
661 | cattr_name(flags)); | |
662 | return -EINVAL; | |
663 | } | |
664 | /* | |
665 | * We allow returning different type than the one requested in | |
666 | * non strict case. | |
667 | */ | |
668 | *vma_prot = __pgprot((pgprot_val(*vma_prot) & | |
669 | (~_PAGE_CACHE_MASK)) | | |
670 | flags); | |
5899329b | 671 | } |
672 | ||
b5db0e38 LT |
673 | /* Need to keep identity mapping in sync */ |
674 | if (paddr >= __pa(high_memory)) | |
675 | return 0; | |
676 | ||
677 | id_sz = (__pa(high_memory) < paddr + size) ? | |
678 | __pa(high_memory) - paddr : | |
679 | size; | |
680 | ||
681 | if (ioremap_change_attr((unsigned long)__va(paddr), id_sz, flags) < 0) { | |
5899329b | 682 | free_memtype(paddr, paddr + size); |
683 | printk(KERN_ERR | |
684 | "%s:%d reserve_pfn_range ioremap_change_attr failed %s " | |
685 | "for %Lx-%Lx\n", | |
686 | current->comm, current->pid, | |
687 | cattr_name(flags), | |
688 | (unsigned long long)paddr, | |
689 | (unsigned long long)(paddr + size)); | |
690 | return -EINVAL; | |
691 | } | |
692 | return 0; | |
693 | } | |
694 | ||
695 | /* | |
696 | * Internal interface to free a range of physical memory. | |
697 | * Frees non RAM regions only. | |
698 | */ | |
699 | static void free_pfn_range(u64 paddr, unsigned long size) | |
700 | { | |
701 | int is_ram; | |
702 | ||
be03d9e8 | 703 | is_ram = pat_pagerange_is_ram(paddr, paddr + size); |
5899329b | 704 | if (is_ram == 0) |
705 | free_memtype(paddr, paddr + size); | |
706 | } | |
707 | ||
708 | /* | |
709 | * track_pfn_vma_copy is called when vma that is covering the pfnmap gets | |
710 | * copied through copy_page_range(). | |
711 | * | |
712 | * If the vma has a linear pfn mapping for the entire range, we get the prot | |
713 | * from pte and reserve the entire vma range with single reserve_pfn_range call. | |
714 | * Otherwise, we reserve the entire vma range, my ging through the PTEs page | |
715 | * by page to get physical address and protection. | |
716 | */ | |
717 | int track_pfn_vma_copy(struct vm_area_struct *vma) | |
718 | { | |
719 | int retval = 0; | |
720 | unsigned long i, j; | |
c1c15b65 | 721 | resource_size_t paddr; |
982d789a | 722 | unsigned long prot; |
5899329b | 723 | unsigned long vma_start = vma->vm_start; |
724 | unsigned long vma_end = vma->vm_end; | |
725 | unsigned long vma_size = vma_end - vma_start; | |
cdecff68 | 726 | pgprot_t pgprot; |
5899329b | 727 | |
728 | if (!pat_enabled) | |
729 | return 0; | |
730 | ||
731 | if (is_linear_pfn_mapping(vma)) { | |
732 | /* | |
982d789a | 733 | * reserve the whole chunk covered by vma. We need the |
734 | * starting address and protection from pte. | |
5899329b | 735 | */ |
982d789a | 736 | if (follow_phys(vma, vma_start, 0, &prot, &paddr)) { |
5899329b | 737 | WARN_ON_ONCE(1); |
982d789a | 738 | return -EINVAL; |
5899329b | 739 | } |
cdecff68 | 740 | pgprot = __pgprot(prot); |
741 | return reserve_pfn_range(paddr, vma_size, &pgprot, 1); | |
5899329b | 742 | } |
743 | ||
744 | /* reserve entire vma page by page, using pfn and prot from pte */ | |
745 | for (i = 0; i < vma_size; i += PAGE_SIZE) { | |
982d789a | 746 | if (follow_phys(vma, vma_start + i, 0, &prot, &paddr)) |
5899329b | 747 | continue; |
748 | ||
cdecff68 | 749 | pgprot = __pgprot(prot); |
750 | retval = reserve_pfn_range(paddr, PAGE_SIZE, &pgprot, 1); | |
5899329b | 751 | if (retval) |
752 | goto cleanup_ret; | |
753 | } | |
754 | return 0; | |
755 | ||
756 | cleanup_ret: | |
757 | /* Reserve error: Cleanup partial reservation and return error */ | |
758 | for (j = 0; j < i; j += PAGE_SIZE) { | |
982d789a | 759 | if (follow_phys(vma, vma_start + j, 0, &prot, &paddr)) |
5899329b | 760 | continue; |
761 | ||
5899329b | 762 | free_pfn_range(paddr, PAGE_SIZE); |
763 | } | |
764 | ||
765 | return retval; | |
766 | } | |
767 | ||
768 | /* | |
769 | * track_pfn_vma_new is called when a _new_ pfn mapping is being established | |
770 | * for physical range indicated by pfn and size. | |
771 | * | |
772 | * prot is passed in as a parameter for the new mapping. If the vma has a | |
773 | * linear pfn mapping for the entire range reserve the entire vma range with | |
774 | * single reserve_pfn_range call. | |
775 | * Otherwise, we look t the pfn and size and reserve only the specified range | |
776 | * page by page. | |
777 | * | |
778 | * Note that this function can be called with caller trying to map only a | |
779 | * subrange/page inside the vma. | |
780 | */ | |
e4b866ed | 781 | int track_pfn_vma_new(struct vm_area_struct *vma, pgprot_t *prot, |
5899329b | 782 | unsigned long pfn, unsigned long size) |
783 | { | |
784 | int retval = 0; | |
785 | unsigned long i, j; | |
c1c15b65 PA |
786 | resource_size_t base_paddr; |
787 | resource_size_t paddr; | |
5899329b | 788 | unsigned long vma_start = vma->vm_start; |
789 | unsigned long vma_end = vma->vm_end; | |
790 | unsigned long vma_size = vma_end - vma_start; | |
791 | ||
792 | if (!pat_enabled) | |
793 | return 0; | |
794 | ||
795 | if (is_linear_pfn_mapping(vma)) { | |
796 | /* reserve the whole chunk starting from vm_pgoff */ | |
c1c15b65 | 797 | paddr = (resource_size_t)vma->vm_pgoff << PAGE_SHIFT; |
cdecff68 | 798 | return reserve_pfn_range(paddr, vma_size, prot, 0); |
5899329b | 799 | } |
800 | ||
801 | /* reserve page by page using pfn and size */ | |
c1c15b65 | 802 | base_paddr = (resource_size_t)pfn << PAGE_SHIFT; |
5899329b | 803 | for (i = 0; i < size; i += PAGE_SIZE) { |
804 | paddr = base_paddr + i; | |
cdecff68 | 805 | retval = reserve_pfn_range(paddr, PAGE_SIZE, prot, 0); |
5899329b | 806 | if (retval) |
807 | goto cleanup_ret; | |
808 | } | |
809 | return 0; | |
810 | ||
811 | cleanup_ret: | |
812 | /* Reserve error: Cleanup partial reservation and return error */ | |
813 | for (j = 0; j < i; j += PAGE_SIZE) { | |
814 | paddr = base_paddr + j; | |
815 | free_pfn_range(paddr, PAGE_SIZE); | |
816 | } | |
817 | ||
818 | return retval; | |
819 | } | |
820 | ||
821 | /* | |
822 | * untrack_pfn_vma is called while unmapping a pfnmap for a region. | |
823 | * untrack can be called for a specific region indicated by pfn and size or | |
824 | * can be for the entire vma (in which case size can be zero). | |
825 | */ | |
826 | void untrack_pfn_vma(struct vm_area_struct *vma, unsigned long pfn, | |
827 | unsigned long size) | |
828 | { | |
829 | unsigned long i; | |
c1c15b65 | 830 | resource_size_t paddr; |
982d789a | 831 | unsigned long prot; |
5899329b | 832 | unsigned long vma_start = vma->vm_start; |
833 | unsigned long vma_end = vma->vm_end; | |
834 | unsigned long vma_size = vma_end - vma_start; | |
835 | ||
836 | if (!pat_enabled) | |
837 | return; | |
838 | ||
839 | if (is_linear_pfn_mapping(vma)) { | |
840 | /* free the whole chunk starting from vm_pgoff */ | |
c1c15b65 | 841 | paddr = (resource_size_t)vma->vm_pgoff << PAGE_SHIFT; |
5899329b | 842 | free_pfn_range(paddr, vma_size); |
843 | return; | |
844 | } | |
845 | ||
846 | if (size != 0 && size != vma_size) { | |
847 | /* free page by page, using pfn and size */ | |
c1c15b65 | 848 | paddr = (resource_size_t)pfn << PAGE_SHIFT; |
5899329b | 849 | for (i = 0; i < size; i += PAGE_SIZE) { |
850 | paddr = paddr + i; | |
851 | free_pfn_range(paddr, PAGE_SIZE); | |
852 | } | |
853 | } else { | |
854 | /* free entire vma, page by page, using the pfn from pte */ | |
855 | for (i = 0; i < vma_size; i += PAGE_SIZE) { | |
982d789a | 856 | if (follow_phys(vma, vma_start + i, 0, &prot, &paddr)) |
5899329b | 857 | continue; |
858 | ||
5899329b | 859 | free_pfn_range(paddr, PAGE_SIZE); |
860 | } | |
861 | } | |
862 | } | |
863 | ||
2520bd31 | 864 | pgprot_t pgprot_writecombine(pgprot_t prot) |
865 | { | |
866 | if (pat_enabled) | |
867 | return __pgprot(pgprot_val(prot) | _PAGE_CACHE_WC); | |
868 | else | |
869 | return pgprot_noncached(prot); | |
870 | } | |
871 | ||
012f09e7 | 872 | #if defined(CONFIG_DEBUG_FS) && defined(CONFIG_X86_PAT) |
fec0962e | 873 | |
874 | /* get Nth element of the linked list */ | |
875 | static struct memtype *memtype_get_idx(loff_t pos) | |
876 | { | |
877 | struct memtype *list_node, *print_entry; | |
878 | int i = 1; | |
879 | ||
880 | print_entry = kmalloc(sizeof(struct memtype), GFP_KERNEL); | |
881 | if (!print_entry) | |
882 | return NULL; | |
883 | ||
884 | spin_lock(&memtype_lock); | |
885 | list_for_each_entry(list_node, &memtype_list, nd) { | |
886 | if (pos == i) { | |
887 | *print_entry = *list_node; | |
888 | spin_unlock(&memtype_lock); | |
889 | return print_entry; | |
890 | } | |
891 | ++i; | |
892 | } | |
893 | spin_unlock(&memtype_lock); | |
894 | kfree(print_entry); | |
ad2cde16 | 895 | |
fec0962e | 896 | return NULL; |
897 | } | |
898 | ||
899 | static void *memtype_seq_start(struct seq_file *seq, loff_t *pos) | |
900 | { | |
901 | if (*pos == 0) { | |
902 | ++*pos; | |
903 | seq_printf(seq, "PAT memtype list:\n"); | |
904 | } | |
905 | ||
906 | return memtype_get_idx(*pos); | |
907 | } | |
908 | ||
909 | static void *memtype_seq_next(struct seq_file *seq, void *v, loff_t *pos) | |
910 | { | |
911 | ++*pos; | |
912 | return memtype_get_idx(*pos); | |
913 | } | |
914 | ||
915 | static void memtype_seq_stop(struct seq_file *seq, void *v) | |
916 | { | |
917 | } | |
918 | ||
919 | static int memtype_seq_show(struct seq_file *seq, void *v) | |
920 | { | |
921 | struct memtype *print_entry = (struct memtype *)v; | |
922 | ||
923 | seq_printf(seq, "%s @ 0x%Lx-0x%Lx\n", cattr_name(print_entry->type), | |
924 | print_entry->start, print_entry->end); | |
925 | kfree(print_entry); | |
ad2cde16 | 926 | |
fec0962e | 927 | return 0; |
928 | } | |
929 | ||
930 | static struct seq_operations memtype_seq_ops = { | |
931 | .start = memtype_seq_start, | |
932 | .next = memtype_seq_next, | |
933 | .stop = memtype_seq_stop, | |
934 | .show = memtype_seq_show, | |
935 | }; | |
936 | ||
937 | static int memtype_seq_open(struct inode *inode, struct file *file) | |
938 | { | |
939 | return seq_open(file, &memtype_seq_ops); | |
940 | } | |
941 | ||
942 | static const struct file_operations memtype_fops = { | |
943 | .open = memtype_seq_open, | |
944 | .read = seq_read, | |
945 | .llseek = seq_lseek, | |
946 | .release = seq_release, | |
947 | }; | |
948 | ||
949 | static int __init pat_memtype_list_init(void) | |
950 | { | |
951 | debugfs_create_file("pat_memtype_list", S_IRUSR, arch_debugfs_dir, | |
952 | NULL, &memtype_fops); | |
953 | return 0; | |
954 | } | |
955 | ||
956 | late_initcall(pat_memtype_list_init); | |
957 | ||
012f09e7 | 958 | #endif /* CONFIG_DEBUG_FS && CONFIG_X86_PAT */ |