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> |
2e5d9c85 | 15 | #include <linux/gfp.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> |
2e5d9c85 | 23 | #include <asm/pgtable.h> |
2e5d9c85 | 24 | #include <asm/fcntl.h> |
ad2cde16 | 25 | #include <asm/e820.h> |
2e5d9c85 | 26 | #include <asm/mtrr.h> |
ad2cde16 IM |
27 | #include <asm/page.h> |
28 | #include <asm/msr.h> | |
29 | #include <asm/pat.h> | |
e7f260a2 | 30 | #include <asm/io.h> |
2e5d9c85 | 31 | |
8d4a4300 | 32 | #ifdef CONFIG_X86_PAT |
499f8f84 | 33 | int __read_mostly pat_enabled = 1; |
2e5d9c85 | 34 | |
1ee4bd92 | 35 | static inline void pat_disable(const char *reason) |
2e5d9c85 | 36 | { |
499f8f84 | 37 | pat_enabled = 0; |
8d4a4300 | 38 | printk(KERN_INFO "%s\n", reason); |
2e5d9c85 | 39 | } |
2e5d9c85 | 40 | |
be524fb9 | 41 | static int __init nopat(char *str) |
2e5d9c85 | 42 | { |
8d4a4300 | 43 | pat_disable("PAT support disabled."); |
2e5d9c85 | 44 | return 0; |
45 | } | |
8d4a4300 | 46 | early_param("nopat", nopat); |
75a04811 PA |
47 | #else |
48 | static inline void pat_disable(const char *reason) | |
49 | { | |
50 | (void)reason; | |
51 | } | |
8d4a4300 TG |
52 | #endif |
53 | ||
77b52b4c VP |
54 | |
55 | static int debug_enable; | |
ad2cde16 | 56 | |
77b52b4c VP |
57 | static int __init pat_debug_setup(char *str) |
58 | { | |
59 | debug_enable = 1; | |
60 | return 0; | |
61 | } | |
62 | __setup("debugpat", pat_debug_setup); | |
63 | ||
64 | #define dprintk(fmt, arg...) \ | |
65 | do { if (debug_enable) printk(KERN_INFO fmt, ##arg); } while (0) | |
66 | ||
67 | ||
8d4a4300 | 68 | static u64 __read_mostly boot_pat_state; |
2e5d9c85 | 69 | |
70 | enum { | |
71 | PAT_UC = 0, /* uncached */ | |
72 | PAT_WC = 1, /* Write combining */ | |
73 | PAT_WT = 4, /* Write Through */ | |
74 | PAT_WP = 5, /* Write Protected */ | |
75 | PAT_WB = 6, /* Write Back (default) */ | |
76 | PAT_UC_MINUS = 7, /* UC, but can be overriden by MTRR */ | |
77 | }; | |
78 | ||
cd7a4e93 | 79 | #define PAT(x, y) ((u64)PAT_ ## y << ((x)*8)) |
2e5d9c85 | 80 | |
81 | void pat_init(void) | |
82 | { | |
83 | u64 pat; | |
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); | |
125 | printk(KERN_INFO "x86 PAT enabled: cpu %d, old 0x%Lx, new 0x%Lx\n", | |
126 | smp_processor_id(), boot_pat_state, pat); | |
127 | } | |
128 | ||
129 | #undef PAT | |
130 | ||
131 | static char *cattr_name(unsigned long flags) | |
132 | { | |
133 | switch (flags & _PAGE_CACHE_MASK) { | |
cd7a4e93 AH |
134 | case _PAGE_CACHE_UC: return "uncached"; |
135 | case _PAGE_CACHE_UC_MINUS: return "uncached-minus"; | |
136 | case _PAGE_CACHE_WB: return "write-back"; | |
137 | case _PAGE_CACHE_WC: return "write-combining"; | |
138 | default: return "broken"; | |
2e5d9c85 | 139 | } |
140 | } | |
141 | ||
142 | /* | |
143 | * The global memtype list keeps track of memory type for specific | |
144 | * physical memory areas. Conflicting memory types in different | |
145 | * mappings can cause CPU cache corruption. To avoid this we keep track. | |
146 | * | |
147 | * The list is sorted based on starting address and can contain multiple | |
148 | * entries for each address (this allows reference counting for overlapping | |
149 | * areas). All the aliases have the same cache attributes of course. | |
150 | * Zero attributes are represented as holes. | |
151 | * | |
335ef896 VP |
152 | * The data structure is a list that is also organized as an rbtree |
153 | * sorted on the start address of memtype range. | |
2e5d9c85 | 154 | * |
335ef896 | 155 | * memtype_lock protects both the linear list and rbtree. |
2e5d9c85 | 156 | */ |
157 | ||
158 | struct memtype { | |
ad2cde16 IM |
159 | u64 start; |
160 | u64 end; | |
161 | unsigned long type; | |
162 | struct list_head nd; | |
335ef896 | 163 | struct rb_node rb; |
2e5d9c85 | 164 | }; |
165 | ||
335ef896 | 166 | static struct rb_root memtype_rbroot = RB_ROOT; |
2e5d9c85 | 167 | static LIST_HEAD(memtype_list); |
ad2cde16 | 168 | static DEFINE_SPINLOCK(memtype_lock); /* protects memtype list */ |
2e5d9c85 | 169 | |
335ef896 VP |
170 | static struct memtype *memtype_rb_search(struct rb_root *root, u64 start) |
171 | { | |
172 | struct rb_node *node = root->rb_node; | |
173 | struct memtype *last_lower = NULL; | |
174 | ||
175 | while (node) { | |
176 | struct memtype *data = container_of(node, struct memtype, rb); | |
177 | ||
178 | if (data->start < start) { | |
179 | last_lower = data; | |
180 | node = node->rb_right; | |
181 | } else if (data->start > start) { | |
182 | node = node->rb_left; | |
183 | } else | |
184 | return data; | |
185 | } | |
186 | ||
187 | /* Will return NULL if there is no entry with its start <= start */ | |
188 | return last_lower; | |
189 | } | |
190 | ||
191 | static void memtype_rb_insert(struct rb_root *root, struct memtype *data) | |
192 | { | |
193 | struct rb_node **new = &(root->rb_node); | |
194 | struct rb_node *parent = NULL; | |
195 | ||
196 | while (*new) { | |
197 | struct memtype *this = container_of(*new, struct memtype, rb); | |
198 | ||
199 | parent = *new; | |
200 | if (data->start <= this->start) | |
201 | new = &((*new)->rb_left); | |
202 | else if (data->start > this->start) | |
203 | new = &((*new)->rb_right); | |
204 | } | |
205 | ||
206 | rb_link_node(&data->rb, parent, new); | |
207 | rb_insert_color(&data->rb, root); | |
208 | } | |
209 | ||
2e5d9c85 | 210 | /* |
211 | * Does intersection of PAT memory type and MTRR memory type and returns | |
212 | * the resulting memory type as PAT understands it. | |
213 | * (Type in pat and mtrr will not have same value) | |
214 | * The intersection is based on "Effective Memory Type" tables in IA-32 | |
215 | * SDM vol 3a | |
216 | */ | |
6cf514fc | 217 | static unsigned long pat_x_mtrr_type(u64 start, u64 end, unsigned long req_type) |
2e5d9c85 | 218 | { |
c26421d0 VP |
219 | /* |
220 | * Look for MTRR hint to get the effective type in case where PAT | |
221 | * request is for WB. | |
222 | */ | |
dd0c7c49 AH |
223 | if (req_type == _PAGE_CACHE_WB) { |
224 | u8 mtrr_type; | |
225 | ||
226 | mtrr_type = mtrr_type_lookup(start, end); | |
b6ff32d9 SS |
227 | if (mtrr_type != MTRR_TYPE_WRBACK) |
228 | return _PAGE_CACHE_UC_MINUS; | |
229 | ||
230 | return _PAGE_CACHE_WB; | |
dd0c7c49 AH |
231 | } |
232 | ||
233 | return req_type; | |
2e5d9c85 | 234 | } |
235 | ||
ad2cde16 IM |
236 | static int |
237 | chk_conflict(struct memtype *new, struct memtype *entry, unsigned long *type) | |
64fe44c3 AH |
238 | { |
239 | if (new->type != entry->type) { | |
240 | if (type) { | |
241 | new->type = entry->type; | |
242 | *type = entry->type; | |
243 | } else | |
244 | goto conflict; | |
245 | } | |
246 | ||
247 | /* check overlaps with more than one entry in the list */ | |
248 | list_for_each_entry_continue(entry, &memtype_list, nd) { | |
249 | if (new->end <= entry->start) | |
250 | break; | |
251 | else if (new->type != entry->type) | |
252 | goto conflict; | |
253 | } | |
254 | return 0; | |
255 | ||
256 | conflict: | |
257 | printk(KERN_INFO "%s:%d conflicting memory types " | |
258 | "%Lx-%Lx %s<->%s\n", current->comm, current->pid, new->start, | |
259 | new->end, cattr_name(new->type), cattr_name(entry->type)); | |
260 | return -EBUSY; | |
261 | } | |
262 | ||
be03d9e8 SS |
263 | static int pat_pagerange_is_ram(unsigned long start, unsigned long end) |
264 | { | |
265 | int ram_page = 0, not_rampage = 0; | |
266 | unsigned long page_nr; | |
267 | ||
268 | for (page_nr = (start >> PAGE_SHIFT); page_nr < (end >> PAGE_SHIFT); | |
269 | ++page_nr) { | |
270 | /* | |
271 | * For legacy reasons, physical address range in the legacy ISA | |
272 | * region is tracked as non-RAM. This will allow users of | |
273 | * /dev/mem to map portions of legacy ISA region, even when | |
274 | * some of those portions are listed(or not even listed) with | |
275 | * different e820 types(RAM/reserved/..) | |
276 | */ | |
277 | if (page_nr >= (ISA_END_ADDRESS >> PAGE_SHIFT) && | |
278 | page_is_ram(page_nr)) | |
279 | ram_page = 1; | |
280 | else | |
281 | not_rampage = 1; | |
282 | ||
283 | if (ram_page == not_rampage) | |
284 | return -1; | |
285 | } | |
286 | ||
287 | return ram_page; | |
288 | } | |
289 | ||
9542ada8 | 290 | /* |
f5841740 VP |
291 | * For RAM pages, we use page flags to mark the pages with appropriate type. |
292 | * Here we do two pass: | |
293 | * - Find the memtype of all the pages in the range, look for any conflicts | |
294 | * - In case of no conflicts, set the new memtype for pages in the range | |
9542ada8 | 295 | * |
f5841740 | 296 | * Caller must hold memtype_lock for atomicity. |
9542ada8 SS |
297 | */ |
298 | static int reserve_ram_pages_type(u64 start, u64 end, unsigned long req_type, | |
ad2cde16 | 299 | unsigned long *new_type) |
9542ada8 SS |
300 | { |
301 | struct page *page; | |
f5841740 VP |
302 | u64 pfn; |
303 | ||
304 | if (req_type == _PAGE_CACHE_UC) { | |
305 | /* We do not support strong UC */ | |
306 | WARN_ON_ONCE(1); | |
307 | req_type = _PAGE_CACHE_UC_MINUS; | |
308 | } | |
9542ada8 SS |
309 | |
310 | for (pfn = (start >> PAGE_SHIFT); pfn < (end >> PAGE_SHIFT); ++pfn) { | |
f5841740 | 311 | unsigned long type; |
9542ada8 | 312 | |
f5841740 VP |
313 | page = pfn_to_page(pfn); |
314 | type = get_page_memtype(page); | |
315 | if (type != -1) { | |
316 | printk(KERN_INFO "reserve_ram_pages_type failed " | |
317 | "0x%Lx-0x%Lx, track 0x%lx, req 0x%lx\n", | |
318 | start, end, type, req_type); | |
319 | if (new_type) | |
320 | *new_type = type; | |
321 | ||
322 | return -EBUSY; | |
323 | } | |
9542ada8 | 324 | } |
9542ada8 | 325 | |
f5841740 VP |
326 | if (new_type) |
327 | *new_type = req_type; | |
328 | ||
329 | for (pfn = (start >> PAGE_SHIFT); pfn < (end >> PAGE_SHIFT); ++pfn) { | |
9542ada8 | 330 | page = pfn_to_page(pfn); |
f5841740 | 331 | set_page_memtype(page, req_type); |
9542ada8 | 332 | } |
f5841740 | 333 | return 0; |
9542ada8 SS |
334 | } |
335 | ||
336 | static int free_ram_pages_type(u64 start, u64 end) | |
337 | { | |
338 | struct page *page; | |
f5841740 | 339 | u64 pfn; |
9542ada8 SS |
340 | |
341 | for (pfn = (start >> PAGE_SHIFT); pfn < (end >> PAGE_SHIFT); ++pfn) { | |
342 | page = pfn_to_page(pfn); | |
f5841740 | 343 | set_page_memtype(page, -1); |
9542ada8 SS |
344 | } |
345 | return 0; | |
9542ada8 SS |
346 | } |
347 | ||
e7f260a2 | 348 | /* |
349 | * req_type typically has one of the: | |
350 | * - _PAGE_CACHE_WB | |
351 | * - _PAGE_CACHE_WC | |
352 | * - _PAGE_CACHE_UC_MINUS | |
353 | * - _PAGE_CACHE_UC | |
354 | * | |
355 | * req_type will have a special case value '-1', when requester want to inherit | |
356 | * the memory type from mtrr (if WB), existing PAT, defaulting to UC_MINUS. | |
357 | * | |
ac97991e AH |
358 | * If new_type is NULL, function will return an error if it cannot reserve the |
359 | * region with req_type. If new_type is non-NULL, function will return | |
360 | * available type in new_type in case of no error. In case of any error | |
e7f260a2 | 361 | * it will return a negative return value. |
362 | */ | |
2e5d9c85 | 363 | int reserve_memtype(u64 start, u64 end, unsigned long req_type, |
ad2cde16 | 364 | unsigned long *new_type) |
2e5d9c85 | 365 | { |
ac97991e | 366 | struct memtype *new, *entry; |
2e5d9c85 | 367 | unsigned long actual_type; |
f6887264 | 368 | struct list_head *where; |
9542ada8 | 369 | int is_range_ram; |
ad2cde16 | 370 | int err = 0; |
2e5d9c85 | 371 | |
ad2cde16 | 372 | BUG_ON(start >= end); /* end is exclusive */ |
69e26be9 | 373 | |
499f8f84 | 374 | if (!pat_enabled) { |
e7f260a2 | 375 | /* This is identical to page table setting without PAT */ |
ac97991e AH |
376 | if (new_type) { |
377 | if (req_type == -1) | |
378 | *new_type = _PAGE_CACHE_WB; | |
5fc51746 VP |
379 | else if (req_type == _PAGE_CACHE_WC) |
380 | *new_type = _PAGE_CACHE_UC_MINUS; | |
ac97991e AH |
381 | else |
382 | *new_type = req_type & _PAGE_CACHE_MASK; | |
e7f260a2 | 383 | } |
2e5d9c85 | 384 | return 0; |
385 | } | |
386 | ||
387 | /* Low ISA region is always mapped WB in page table. No need to track */ | |
bcc643dc | 388 | if (is_ISA_range(start, end - 1)) { |
ac97991e AH |
389 | if (new_type) |
390 | *new_type = _PAGE_CACHE_WB; | |
2e5d9c85 | 391 | return 0; |
392 | } | |
393 | ||
b6ff32d9 SS |
394 | /* |
395 | * Call mtrr_lookup to get the type hint. This is an | |
396 | * optimization for /dev/mem mmap'ers into WB memory (BIOS | |
397 | * tools and ACPI tools). Use WB request for WB memory and use | |
398 | * UC_MINUS otherwise. | |
399 | */ | |
400 | actual_type = pat_x_mtrr_type(start, end, req_type & _PAGE_CACHE_MASK); | |
2e5d9c85 | 401 | |
95971342 SS |
402 | if (new_type) |
403 | *new_type = actual_type; | |
404 | ||
be03d9e8 | 405 | is_range_ram = pat_pagerange_is_ram(start, end); |
f5841740 VP |
406 | if (is_range_ram == 1) { |
407 | ||
408 | spin_lock(&memtype_lock); | |
409 | err = reserve_ram_pages_type(start, end, req_type, new_type); | |
410 | spin_unlock(&memtype_lock); | |
411 | ||
412 | return err; | |
413 | } else if (is_range_ram < 0) { | |
9542ada8 | 414 | return -EINVAL; |
f5841740 | 415 | } |
9542ada8 | 416 | |
ac97991e AH |
417 | new = kmalloc(sizeof(struct memtype), GFP_KERNEL); |
418 | if (!new) | |
2e5d9c85 | 419 | return -ENOMEM; |
420 | ||
ad2cde16 IM |
421 | new->start = start; |
422 | new->end = end; | |
423 | new->type = actual_type; | |
2e5d9c85 | 424 | |
2e5d9c85 | 425 | spin_lock(&memtype_lock); |
426 | ||
427 | /* Search for existing mapping that overlaps the current range */ | |
f6887264 | 428 | where = NULL; |
dcb73bf4 | 429 | list_for_each_entry(entry, &memtype_list, nd) { |
33af9039 | 430 | if (end <= entry->start) { |
f6887264 | 431 | where = entry->nd.prev; |
2e5d9c85 | 432 | break; |
33af9039 | 433 | } else if (start <= entry->start) { /* end > entry->start */ |
64fe44c3 | 434 | err = chk_conflict(new, entry, new_type); |
33af9039 AH |
435 | if (!err) { |
436 | dprintk("Overlap at 0x%Lx-0x%Lx\n", | |
437 | entry->start, entry->end); | |
438 | where = entry->nd.prev; | |
2e5d9c85 | 439 | } |
2e5d9c85 | 440 | break; |
33af9039 | 441 | } else if (start < entry->end) { /* start > entry->start */ |
64fe44c3 | 442 | err = chk_conflict(new, entry, new_type); |
33af9039 AH |
443 | if (!err) { |
444 | dprintk("Overlap at 0x%Lx-0x%Lx\n", | |
445 | entry->start, entry->end); | |
80c5e73d VP |
446 | |
447 | /* | |
448 | * Move to right position in the linked | |
449 | * list to add this new entry | |
450 | */ | |
451 | list_for_each_entry_continue(entry, | |
452 | &memtype_list, nd) { | |
453 | if (start <= entry->start) { | |
454 | where = entry->nd.prev; | |
455 | break; | |
456 | } | |
457 | } | |
2e5d9c85 | 458 | } |
2e5d9c85 | 459 | break; |
460 | } | |
461 | } | |
462 | ||
463 | if (err) { | |
3e9c83b3 AH |
464 | printk(KERN_INFO "reserve_memtype failed 0x%Lx-0x%Lx, " |
465 | "track %s, req %s\n", | |
466 | start, end, cattr_name(new->type), cattr_name(req_type)); | |
ac97991e | 467 | kfree(new); |
2e5d9c85 | 468 | spin_unlock(&memtype_lock); |
ad2cde16 | 469 | |
2e5d9c85 | 470 | return err; |
471 | } | |
472 | ||
f6887264 AH |
473 | if (where) |
474 | list_add(&new->nd, where); | |
475 | else | |
ac97991e | 476 | list_add_tail(&new->nd, &memtype_list); |
6997ab49 | 477 | |
335ef896 VP |
478 | memtype_rb_insert(&memtype_rbroot, new); |
479 | ||
2e5d9c85 | 480 | spin_unlock(&memtype_lock); |
3e9c83b3 AH |
481 | |
482 | dprintk("reserve_memtype added 0x%Lx-0x%Lx, track %s, req %s, ret %s\n", | |
483 | start, end, cattr_name(new->type), cattr_name(req_type), | |
484 | new_type ? cattr_name(*new_type) : "-"); | |
485 | ||
2e5d9c85 | 486 | return err; |
487 | } | |
488 | ||
489 | int free_memtype(u64 start, u64 end) | |
490 | { | |
335ef896 | 491 | struct memtype *entry, *saved_entry; |
2e5d9c85 | 492 | int err = -EINVAL; |
9542ada8 | 493 | int is_range_ram; |
2e5d9c85 | 494 | |
69e26be9 | 495 | if (!pat_enabled) |
2e5d9c85 | 496 | return 0; |
2e5d9c85 | 497 | |
498 | /* Low ISA region is always mapped WB. No need to track */ | |
bcc643dc | 499 | if (is_ISA_range(start, end - 1)) |
2e5d9c85 | 500 | return 0; |
2e5d9c85 | 501 | |
be03d9e8 | 502 | is_range_ram = pat_pagerange_is_ram(start, end); |
f5841740 VP |
503 | if (is_range_ram == 1) { |
504 | ||
505 | spin_lock(&memtype_lock); | |
506 | err = free_ram_pages_type(start, end); | |
507 | spin_unlock(&memtype_lock); | |
508 | ||
509 | return err; | |
510 | } else if (is_range_ram < 0) { | |
9542ada8 | 511 | return -EINVAL; |
f5841740 | 512 | } |
9542ada8 | 513 | |
2e5d9c85 | 514 | spin_lock(&memtype_lock); |
335ef896 VP |
515 | |
516 | entry = memtype_rb_search(&memtype_rbroot, start); | |
517 | if (unlikely(entry == NULL)) | |
518 | goto unlock_ret; | |
519 | ||
520 | /* | |
521 | * Saved entry points to an entry with start same or less than what | |
522 | * we searched for. Now go through the list in both directions to look | |
523 | * for the entry that matches with both start and end, with list stored | |
524 | * in sorted start address | |
525 | */ | |
526 | saved_entry = entry; | |
dcb73bf4 | 527 | list_for_each_entry_from(entry, &memtype_list, nd) { |
ac97991e | 528 | if (entry->start == start && entry->end == end) { |
335ef896 VP |
529 | rb_erase(&entry->rb, &memtype_rbroot); |
530 | list_del(&entry->nd); | |
531 | kfree(entry); | |
532 | err = 0; | |
533 | break; | |
534 | } else if (entry->start > start) { | |
535 | break; | |
536 | } | |
537 | } | |
538 | ||
539 | if (!err) | |
540 | goto unlock_ret; | |
80c5e73d | 541 | |
335ef896 VP |
542 | entry = saved_entry; |
543 | list_for_each_entry_reverse(entry, &memtype_list, nd) { | |
544 | if (entry->start == start && entry->end == end) { | |
545 | rb_erase(&entry->rb, &memtype_rbroot); | |
ac97991e AH |
546 | list_del(&entry->nd); |
547 | kfree(entry); | |
2e5d9c85 | 548 | err = 0; |
549 | break; | |
335ef896 VP |
550 | } else if (entry->start < start) { |
551 | break; | |
2e5d9c85 | 552 | } |
553 | } | |
335ef896 | 554 | unlock_ret: |
2e5d9c85 | 555 | spin_unlock(&memtype_lock); |
556 | ||
557 | if (err) { | |
28eb559b | 558 | printk(KERN_INFO "%s:%d freeing invalid memtype %Lx-%Lx\n", |
2e5d9c85 | 559 | current->comm, current->pid, start, end); |
560 | } | |
6997ab49 | 561 | |
77b52b4c | 562 | dprintk("free_memtype request 0x%Lx-0x%Lx\n", start, end); |
ad2cde16 | 563 | |
2e5d9c85 | 564 | return err; |
565 | } | |
566 | ||
f0970c13 | 567 | |
637b86e7 VP |
568 | /** |
569 | * lookup_memtype - Looksup the memory type for a physical address | |
570 | * @paddr: physical address of which memory type needs to be looked up | |
571 | * | |
572 | * Only to be called when PAT is enabled | |
573 | * | |
574 | * Returns _PAGE_CACHE_WB, _PAGE_CACHE_WC, _PAGE_CACHE_UC_MINUS or | |
575 | * _PAGE_CACHE_UC | |
576 | */ | |
577 | static unsigned long lookup_memtype(u64 paddr) | |
578 | { | |
579 | int rettype = _PAGE_CACHE_WB; | |
580 | struct memtype *entry; | |
581 | ||
582 | if (is_ISA_range(paddr, paddr + PAGE_SIZE - 1)) | |
583 | return rettype; | |
584 | ||
585 | if (pat_pagerange_is_ram(paddr, paddr + PAGE_SIZE)) { | |
586 | struct page *page; | |
587 | spin_lock(&memtype_lock); | |
588 | page = pfn_to_page(paddr >> PAGE_SHIFT); | |
589 | rettype = get_page_memtype(page); | |
590 | spin_unlock(&memtype_lock); | |
591 | /* | |
592 | * -1 from get_page_memtype() implies RAM page is in its | |
593 | * default state and not reserved, and hence of type WB | |
594 | */ | |
595 | if (rettype == -1) | |
596 | rettype = _PAGE_CACHE_WB; | |
597 | ||
598 | return rettype; | |
599 | } | |
600 | ||
601 | spin_lock(&memtype_lock); | |
602 | ||
603 | entry = memtype_rb_search(&memtype_rbroot, paddr); | |
604 | if (entry != NULL) | |
605 | rettype = entry->type; | |
606 | else | |
607 | rettype = _PAGE_CACHE_UC_MINUS; | |
608 | ||
609 | spin_unlock(&memtype_lock); | |
610 | return rettype; | |
611 | } | |
612 | ||
9fd126bc VP |
613 | /** |
614 | * io_reserve_memtype - Request a memory type mapping for a region of memory | |
615 | * @start: start (physical address) of the region | |
616 | * @end: end (physical address) of the region | |
617 | * @type: A pointer to memtype, with requested type. On success, requested | |
618 | * or any other compatible type that was available for the region is returned | |
619 | * | |
620 | * On success, returns 0 | |
621 | * On failure, returns non-zero | |
622 | */ | |
623 | int io_reserve_memtype(resource_size_t start, resource_size_t end, | |
624 | unsigned long *type) | |
625 | { | |
b855192c | 626 | resource_size_t size = end - start; |
9fd126bc VP |
627 | unsigned long req_type = *type; |
628 | unsigned long new_type; | |
629 | int ret; | |
630 | ||
b855192c | 631 | WARN_ON_ONCE(iomem_map_sanity_check(start, size)); |
9fd126bc VP |
632 | |
633 | ret = reserve_memtype(start, end, req_type, &new_type); | |
634 | if (ret) | |
635 | goto out_err; | |
636 | ||
b855192c | 637 | if (!is_new_memtype_allowed(start, size, req_type, new_type)) |
9fd126bc VP |
638 | goto out_free; |
639 | ||
b855192c | 640 | if (kernel_map_sync_memtype(start, size, new_type) < 0) |
9fd126bc VP |
641 | goto out_free; |
642 | ||
643 | *type = new_type; | |
644 | return 0; | |
645 | ||
646 | out_free: | |
647 | free_memtype(start, end); | |
648 | ret = -EBUSY; | |
649 | out_err: | |
650 | return ret; | |
651 | } | |
652 | ||
653 | /** | |
654 | * io_free_memtype - Release a memory type mapping for a region of memory | |
655 | * @start: start (physical address) of the region | |
656 | * @end: end (physical address) of the region | |
657 | */ | |
658 | void io_free_memtype(resource_size_t start, resource_size_t end) | |
659 | { | |
660 | free_memtype(start, end); | |
661 | } | |
662 | ||
f0970c13 | 663 | pgprot_t phys_mem_access_prot(struct file *file, unsigned long pfn, |
664 | unsigned long size, pgprot_t vma_prot) | |
665 | { | |
666 | return vma_prot; | |
667 | } | |
668 | ||
d092633b IM |
669 | #ifdef CONFIG_STRICT_DEVMEM |
670 | /* This check is done in drivers/char/mem.c in case of STRICT_DEVMEM*/ | |
0124cecf VP |
671 | static inline int range_is_allowed(unsigned long pfn, unsigned long size) |
672 | { | |
673 | return 1; | |
674 | } | |
675 | #else | |
9e41bff2 | 676 | /* This check is needed to avoid cache aliasing when PAT is enabled */ |
0124cecf VP |
677 | static inline int range_is_allowed(unsigned long pfn, unsigned long size) |
678 | { | |
679 | u64 from = ((u64)pfn) << PAGE_SHIFT; | |
680 | u64 to = from + size; | |
681 | u64 cursor = from; | |
682 | ||
9e41bff2 RT |
683 | if (!pat_enabled) |
684 | return 1; | |
685 | ||
0124cecf VP |
686 | while (cursor < to) { |
687 | if (!devmem_is_allowed(pfn)) { | |
688 | printk(KERN_INFO | |
689 | "Program %s tried to access /dev/mem between %Lx->%Lx.\n", | |
690 | current->comm, from, to); | |
691 | return 0; | |
692 | } | |
693 | cursor += PAGE_SIZE; | |
694 | pfn++; | |
695 | } | |
696 | return 1; | |
697 | } | |
d092633b | 698 | #endif /* CONFIG_STRICT_DEVMEM */ |
0124cecf | 699 | |
f0970c13 | 700 | int phys_mem_access_prot_allowed(struct file *file, unsigned long pfn, |
701 | unsigned long size, pgprot_t *vma_prot) | |
702 | { | |
0c3c8a18 | 703 | unsigned long flags = _PAGE_CACHE_WB; |
f0970c13 | 704 | |
0124cecf VP |
705 | if (!range_is_allowed(pfn, size)) |
706 | return 0; | |
707 | ||
f0970c13 | 708 | if (file->f_flags & O_SYNC) { |
28df82eb | 709 | flags = _PAGE_CACHE_UC_MINUS; |
f0970c13 | 710 | } |
711 | ||
712 | #ifdef CONFIG_X86_32 | |
713 | /* | |
714 | * On the PPro and successors, the MTRRs are used to set | |
715 | * memory types for physical addresses outside main memory, | |
716 | * so blindly setting UC or PWT on those pages is wrong. | |
717 | * For Pentiums and earlier, the surround logic should disable | |
718 | * caching for the high addresses through the KEN pin, but | |
719 | * we maintain the tradition of paranoia in this code. | |
720 | */ | |
499f8f84 | 721 | if (!pat_enabled && |
cd7a4e93 AH |
722 | !(boot_cpu_has(X86_FEATURE_MTRR) || |
723 | boot_cpu_has(X86_FEATURE_K6_MTRR) || | |
724 | boot_cpu_has(X86_FEATURE_CYRIX_ARR) || | |
725 | boot_cpu_has(X86_FEATURE_CENTAUR_MCR)) && | |
726 | (pfn << PAGE_SHIFT) >= __pa(high_memory)) { | |
e7f260a2 | 727 | flags = _PAGE_CACHE_UC; |
f0970c13 | 728 | } |
729 | #endif | |
730 | ||
e7f260a2 | 731 | *vma_prot = __pgprot((pgprot_val(*vma_prot) & ~_PAGE_CACHE_MASK) | |
732 | flags); | |
f0970c13 | 733 | return 1; |
734 | } | |
e7f260a2 | 735 | |
7880f746 VP |
736 | /* |
737 | * Change the memory type for the physial address range in kernel identity | |
738 | * mapping space if that range is a part of identity map. | |
739 | */ | |
740 | int kernel_map_sync_memtype(u64 base, unsigned long size, unsigned long flags) | |
741 | { | |
742 | unsigned long id_sz; | |
743 | ||
5fc51746 | 744 | if (base >= __pa(high_memory)) |
7880f746 VP |
745 | return 0; |
746 | ||
747 | id_sz = (__pa(high_memory) < base + size) ? | |
748 | __pa(high_memory) - base : | |
749 | size; | |
750 | ||
751 | if (ioremap_change_attr((unsigned long)__va(base), id_sz, flags) < 0) { | |
752 | printk(KERN_INFO | |
753 | "%s:%d ioremap_change_attr failed %s " | |
754 | "for %Lx-%Lx\n", | |
755 | current->comm, current->pid, | |
756 | cattr_name(flags), | |
757 | base, (unsigned long long)(base + size)); | |
758 | return -EINVAL; | |
759 | } | |
760 | return 0; | |
761 | } | |
762 | ||
5899329b | 763 | /* |
764 | * Internal interface to reserve a range of physical memory with prot. | |
765 | * Reserved non RAM regions only and after successful reserve_memtype, | |
766 | * this func also keeps identity mapping (if any) in sync with this new prot. | |
767 | */ | |
cdecff68 | 768 | static int reserve_pfn_range(u64 paddr, unsigned long size, pgprot_t *vma_prot, |
769 | int strict_prot) | |
5899329b | 770 | { |
771 | int is_ram = 0; | |
7880f746 | 772 | int ret; |
cdecff68 | 773 | unsigned long want_flags = (pgprot_val(*vma_prot) & _PAGE_CACHE_MASK); |
0c3c8a18 | 774 | unsigned long flags = want_flags; |
5899329b | 775 | |
be03d9e8 | 776 | is_ram = pat_pagerange_is_ram(paddr, paddr + size); |
5899329b | 777 | |
be03d9e8 | 778 | /* |
d886c73c VP |
779 | * reserve_pfn_range() for RAM pages. We do not refcount to keep |
780 | * track of number of mappings of RAM pages. We can assert that | |
781 | * the type requested matches the type of first page in the range. | |
be03d9e8 | 782 | */ |
d886c73c VP |
783 | if (is_ram) { |
784 | if (!pat_enabled) | |
785 | return 0; | |
786 | ||
787 | flags = lookup_memtype(paddr); | |
788 | if (want_flags != flags) { | |
789 | printk(KERN_WARNING | |
790 | "%s:%d map pfn RAM range req %s for %Lx-%Lx, got %s\n", | |
791 | current->comm, current->pid, | |
792 | cattr_name(want_flags), | |
793 | (unsigned long long)paddr, | |
794 | (unsigned long long)(paddr + size), | |
795 | cattr_name(flags)); | |
796 | *vma_prot = __pgprot((pgprot_val(*vma_prot) & | |
797 | (~_PAGE_CACHE_MASK)) | | |
798 | flags); | |
799 | } | |
4bb9c5c0 | 800 | return 0; |
d886c73c | 801 | } |
5899329b | 802 | |
803 | ret = reserve_memtype(paddr, paddr + size, want_flags, &flags); | |
804 | if (ret) | |
805 | return ret; | |
806 | ||
807 | if (flags != want_flags) { | |
1adcaafe SS |
808 | if (strict_prot || |
809 | !is_new_memtype_allowed(paddr, size, want_flags, flags)) { | |
cdecff68 | 810 | free_memtype(paddr, paddr + size); |
811 | printk(KERN_ERR "%s:%d map pfn expected mapping type %s" | |
812 | " for %Lx-%Lx, got %s\n", | |
813 | current->comm, current->pid, | |
814 | cattr_name(want_flags), | |
815 | (unsigned long long)paddr, | |
816 | (unsigned long long)(paddr + size), | |
817 | cattr_name(flags)); | |
818 | return -EINVAL; | |
819 | } | |
820 | /* | |
821 | * We allow returning different type than the one requested in | |
822 | * non strict case. | |
823 | */ | |
824 | *vma_prot = __pgprot((pgprot_val(*vma_prot) & | |
825 | (~_PAGE_CACHE_MASK)) | | |
826 | flags); | |
5899329b | 827 | } |
828 | ||
7880f746 | 829 | if (kernel_map_sync_memtype(paddr, size, flags) < 0) { |
5899329b | 830 | free_memtype(paddr, paddr + size); |
5899329b | 831 | return -EINVAL; |
832 | } | |
833 | return 0; | |
834 | } | |
835 | ||
836 | /* | |
837 | * Internal interface to free a range of physical memory. | |
838 | * Frees non RAM regions only. | |
839 | */ | |
840 | static void free_pfn_range(u64 paddr, unsigned long size) | |
841 | { | |
842 | int is_ram; | |
843 | ||
be03d9e8 | 844 | is_ram = pat_pagerange_is_ram(paddr, paddr + size); |
5899329b | 845 | if (is_ram == 0) |
846 | free_memtype(paddr, paddr + size); | |
847 | } | |
848 | ||
849 | /* | |
850 | * track_pfn_vma_copy is called when vma that is covering the pfnmap gets | |
851 | * copied through copy_page_range(). | |
852 | * | |
853 | * If the vma has a linear pfn mapping for the entire range, we get the prot | |
854 | * from pte and reserve the entire vma range with single reserve_pfn_range call. | |
5899329b | 855 | */ |
856 | int track_pfn_vma_copy(struct vm_area_struct *vma) | |
857 | { | |
c1c15b65 | 858 | resource_size_t paddr; |
982d789a | 859 | unsigned long prot; |
4b065046 | 860 | unsigned long vma_size = vma->vm_end - vma->vm_start; |
cdecff68 | 861 | pgprot_t pgprot; |
5899329b | 862 | |
5899329b | 863 | if (is_linear_pfn_mapping(vma)) { |
864 | /* | |
982d789a | 865 | * reserve the whole chunk covered by vma. We need the |
866 | * starting address and protection from pte. | |
5899329b | 867 | */ |
4b065046 | 868 | if (follow_phys(vma, vma->vm_start, 0, &prot, &paddr)) { |
5899329b | 869 | WARN_ON_ONCE(1); |
982d789a | 870 | return -EINVAL; |
5899329b | 871 | } |
cdecff68 | 872 | pgprot = __pgprot(prot); |
873 | return reserve_pfn_range(paddr, vma_size, &pgprot, 1); | |
5899329b | 874 | } |
875 | ||
5899329b | 876 | return 0; |
5899329b | 877 | } |
878 | ||
879 | /* | |
880 | * track_pfn_vma_new is called when a _new_ pfn mapping is being established | |
881 | * for physical range indicated by pfn and size. | |
882 | * | |
883 | * prot is passed in as a parameter for the new mapping. If the vma has a | |
884 | * linear pfn mapping for the entire range reserve the entire vma range with | |
885 | * single reserve_pfn_range call. | |
5899329b | 886 | */ |
e4b866ed | 887 | int track_pfn_vma_new(struct vm_area_struct *vma, pgprot_t *prot, |
5899329b | 888 | unsigned long pfn, unsigned long size) |
889 | { | |
10876376 | 890 | unsigned long flags; |
c1c15b65 | 891 | resource_size_t paddr; |
4b065046 | 892 | unsigned long vma_size = vma->vm_end - vma->vm_start; |
5899329b | 893 | |
5899329b | 894 | if (is_linear_pfn_mapping(vma)) { |
895 | /* reserve the whole chunk starting from vm_pgoff */ | |
c1c15b65 | 896 | paddr = (resource_size_t)vma->vm_pgoff << PAGE_SHIFT; |
cdecff68 | 897 | return reserve_pfn_range(paddr, vma_size, prot, 0); |
5899329b | 898 | } |
899 | ||
10876376 VP |
900 | if (!pat_enabled) |
901 | return 0; | |
902 | ||
903 | /* for vm_insert_pfn and friends, we set prot based on lookup */ | |
904 | flags = lookup_memtype(pfn << PAGE_SHIFT); | |
905 | *prot = __pgprot((pgprot_val(vma->vm_page_prot) & (~_PAGE_CACHE_MASK)) | | |
906 | flags); | |
907 | ||
5899329b | 908 | return 0; |
5899329b | 909 | } |
910 | ||
911 | /* | |
912 | * untrack_pfn_vma is called while unmapping a pfnmap for a region. | |
913 | * untrack can be called for a specific region indicated by pfn and size or | |
914 | * can be for the entire vma (in which case size can be zero). | |
915 | */ | |
916 | void untrack_pfn_vma(struct vm_area_struct *vma, unsigned long pfn, | |
917 | unsigned long size) | |
918 | { | |
c1c15b65 | 919 | resource_size_t paddr; |
4b065046 | 920 | unsigned long vma_size = vma->vm_end - vma->vm_start; |
5899329b | 921 | |
5899329b | 922 | if (is_linear_pfn_mapping(vma)) { |
923 | /* free the whole chunk starting from vm_pgoff */ | |
c1c15b65 | 924 | paddr = (resource_size_t)vma->vm_pgoff << PAGE_SHIFT; |
5899329b | 925 | free_pfn_range(paddr, vma_size); |
926 | return; | |
927 | } | |
5899329b | 928 | } |
929 | ||
2520bd31 | 930 | pgprot_t pgprot_writecombine(pgprot_t prot) |
931 | { | |
932 | if (pat_enabled) | |
933 | return __pgprot(pgprot_val(prot) | _PAGE_CACHE_WC); | |
934 | else | |
935 | return pgprot_noncached(prot); | |
936 | } | |
92b9af9e | 937 | EXPORT_SYMBOL_GPL(pgprot_writecombine); |
2520bd31 | 938 | |
012f09e7 | 939 | #if defined(CONFIG_DEBUG_FS) && defined(CONFIG_X86_PAT) |
fec0962e | 940 | |
941 | /* get Nth element of the linked list */ | |
942 | static struct memtype *memtype_get_idx(loff_t pos) | |
943 | { | |
944 | struct memtype *list_node, *print_entry; | |
945 | int i = 1; | |
946 | ||
947 | print_entry = kmalloc(sizeof(struct memtype), GFP_KERNEL); | |
948 | if (!print_entry) | |
949 | return NULL; | |
950 | ||
951 | spin_lock(&memtype_lock); | |
952 | list_for_each_entry(list_node, &memtype_list, nd) { | |
953 | if (pos == i) { | |
954 | *print_entry = *list_node; | |
955 | spin_unlock(&memtype_lock); | |
956 | return print_entry; | |
957 | } | |
958 | ++i; | |
959 | } | |
960 | spin_unlock(&memtype_lock); | |
961 | kfree(print_entry); | |
ad2cde16 | 962 | |
fec0962e | 963 | return NULL; |
964 | } | |
965 | ||
966 | static void *memtype_seq_start(struct seq_file *seq, loff_t *pos) | |
967 | { | |
968 | if (*pos == 0) { | |
969 | ++*pos; | |
970 | seq_printf(seq, "PAT memtype list:\n"); | |
971 | } | |
972 | ||
973 | return memtype_get_idx(*pos); | |
974 | } | |
975 | ||
976 | static void *memtype_seq_next(struct seq_file *seq, void *v, loff_t *pos) | |
977 | { | |
978 | ++*pos; | |
979 | return memtype_get_idx(*pos); | |
980 | } | |
981 | ||
982 | static void memtype_seq_stop(struct seq_file *seq, void *v) | |
983 | { | |
984 | } | |
985 | ||
986 | static int memtype_seq_show(struct seq_file *seq, void *v) | |
987 | { | |
988 | struct memtype *print_entry = (struct memtype *)v; | |
989 | ||
990 | seq_printf(seq, "%s @ 0x%Lx-0x%Lx\n", cattr_name(print_entry->type), | |
991 | print_entry->start, print_entry->end); | |
992 | kfree(print_entry); | |
ad2cde16 | 993 | |
fec0962e | 994 | return 0; |
995 | } | |
996 | ||
d535e431 | 997 | static const struct seq_operations memtype_seq_ops = { |
fec0962e | 998 | .start = memtype_seq_start, |
999 | .next = memtype_seq_next, | |
1000 | .stop = memtype_seq_stop, | |
1001 | .show = memtype_seq_show, | |
1002 | }; | |
1003 | ||
1004 | static int memtype_seq_open(struct inode *inode, struct file *file) | |
1005 | { | |
1006 | return seq_open(file, &memtype_seq_ops); | |
1007 | } | |
1008 | ||
1009 | static const struct file_operations memtype_fops = { | |
1010 | .open = memtype_seq_open, | |
1011 | .read = seq_read, | |
1012 | .llseek = seq_lseek, | |
1013 | .release = seq_release, | |
1014 | }; | |
1015 | ||
1016 | static int __init pat_memtype_list_init(void) | |
1017 | { | |
1018 | debugfs_create_file("pat_memtype_list", S_IRUSR, arch_debugfs_dir, | |
1019 | NULL, &memtype_fops); | |
1020 | return 0; | |
1021 | } | |
1022 | ||
1023 | late_initcall(pat_memtype_list_init); | |
1024 | ||
012f09e7 | 1025 | #endif /* CONFIG_DEBUG_FS && CONFIG_X86_PAT */ |