]>
Commit | Line | Data |
---|---|---|
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 | ||
10 | #include <linux/seq_file.h> | |
11 | #include <linux/bootmem.h> | |
12 | #include <linux/debugfs.h> | |
13 | #include <linux/kernel.h> | |
14 | #include <linux/module.h> | |
15 | #include <linux/slab.h> | |
16 | #include <linux/mm.h> | |
17 | #include <linux/fs.h> | |
18 | #include <linux/rbtree.h> | |
19 | ||
20 | #include <asm/cacheflush.h> | |
21 | #include <asm/processor.h> | |
22 | #include <asm/tlbflush.h> | |
23 | #include <asm/x86_init.h> | |
24 | #include <asm/pgtable.h> | |
25 | #include <asm/fcntl.h> | |
26 | #include <asm/e820.h> | |
27 | #include <asm/mtrr.h> | |
28 | #include <asm/page.h> | |
29 | #include <asm/msr.h> | |
30 | #include <asm/pat.h> | |
31 | #include <asm/io.h> | |
32 | ||
33 | #include "pat_internal.h" | |
34 | #include "mm_internal.h" | |
35 | ||
36 | #ifdef CONFIG_X86_PAT | |
37 | int __read_mostly pat_enabled = 1; | |
38 | ||
39 | static inline void pat_disable(const char *reason) | |
40 | { | |
41 | pat_enabled = 0; | |
42 | printk(KERN_INFO "%s\n", reason); | |
43 | } | |
44 | ||
45 | static int __init nopat(char *str) | |
46 | { | |
47 | pat_disable("PAT support disabled."); | |
48 | return 0; | |
49 | } | |
50 | early_param("nopat", nopat); | |
51 | #else | |
52 | static inline void pat_disable(const char *reason) | |
53 | { | |
54 | (void)reason; | |
55 | } | |
56 | #endif | |
57 | ||
58 | ||
59 | int pat_debug_enable; | |
60 | ||
61 | static int __init pat_debug_setup(char *str) | |
62 | { | |
63 | pat_debug_enable = 1; | |
64 | return 0; | |
65 | } | |
66 | __setup("debugpat", pat_debug_setup); | |
67 | ||
68 | static u64 __read_mostly boot_pat_state; | |
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 | ||
79 | #define CM(c) (_PAGE_CACHE_MODE_ ## c) | |
80 | ||
81 | static enum page_cache_mode pat_get_cache_mode(unsigned pat_val, char *msg) | |
82 | { | |
83 | enum page_cache_mode cache; | |
84 | char *cache_mode; | |
85 | ||
86 | switch (pat_val) { | |
87 | case PAT_UC: cache = CM(UC); cache_mode = "UC "; break; | |
88 | case PAT_WC: cache = CM(WC); cache_mode = "WC "; break; | |
89 | case PAT_WT: cache = CM(WT); cache_mode = "WT "; break; | |
90 | case PAT_WP: cache = CM(WP); cache_mode = "WP "; break; | |
91 | case PAT_WB: cache = CM(WB); cache_mode = "WB "; break; | |
92 | case PAT_UC_MINUS: cache = CM(UC_MINUS); cache_mode = "UC- "; break; | |
93 | default: cache = CM(WB); cache_mode = "WB "; break; | |
94 | } | |
95 | ||
96 | memcpy(msg, cache_mode, 4); | |
97 | ||
98 | return cache; | |
99 | } | |
100 | ||
101 | #undef CM | |
102 | ||
103 | /* | |
104 | * Update the cache mode to pgprot translation tables according to PAT | |
105 | * configuration. | |
106 | * Using lower indices is preferred, so we start with highest index. | |
107 | */ | |
108 | void pat_init_cache_modes(void) | |
109 | { | |
110 | int i; | |
111 | enum page_cache_mode cache; | |
112 | char pat_msg[33]; | |
113 | u64 pat; | |
114 | ||
115 | rdmsrl(MSR_IA32_CR_PAT, pat); | |
116 | pat_msg[32] = 0; | |
117 | for (i = 7; i >= 0; i--) { | |
118 | cache = pat_get_cache_mode((pat >> (i * 8)) & 7, | |
119 | pat_msg + 4 * i); | |
120 | update_cache_mode_entry(i, cache); | |
121 | } | |
122 | pr_info("PAT configuration [0-7]: %s\n", pat_msg); | |
123 | } | |
124 | ||
125 | #define PAT(x, y) ((u64)PAT_ ## y << ((x)*8)) | |
126 | ||
127 | void pat_init(void) | |
128 | { | |
129 | u64 pat; | |
130 | bool boot_cpu = !boot_pat_state; | |
131 | ||
132 | if (!pat_enabled) | |
133 | return; | |
134 | ||
135 | if (!cpu_has_pat) { | |
136 | if (!boot_pat_state) { | |
137 | pat_disable("PAT not supported by CPU."); | |
138 | return; | |
139 | } else { | |
140 | /* | |
141 | * If this happens we are on a secondary CPU, but | |
142 | * switched to PAT on the boot CPU. We have no way to | |
143 | * undo PAT. | |
144 | */ | |
145 | printk(KERN_ERR "PAT enabled, " | |
146 | "but not supported by secondary CPU\n"); | |
147 | BUG(); | |
148 | } | |
149 | } | |
150 | ||
151 | /* Set PWT to Write-Combining. All other bits stay the same */ | |
152 | /* | |
153 | * PTE encoding used in Linux: | |
154 | * PAT | |
155 | * |PCD | |
156 | * ||PWT | |
157 | * ||| | |
158 | * 000 WB _PAGE_CACHE_WB | |
159 | * 001 WC _PAGE_CACHE_WC | |
160 | * 010 UC- _PAGE_CACHE_UC_MINUS | |
161 | * 011 UC _PAGE_CACHE_UC | |
162 | * PAT bit unused | |
163 | */ | |
164 | pat = PAT(0, WB) | PAT(1, WC) | PAT(2, UC_MINUS) | PAT(3, UC) | | |
165 | PAT(4, WB) | PAT(5, WC) | PAT(6, UC_MINUS) | PAT(7, UC); | |
166 | ||
167 | /* Boot CPU check */ | |
168 | if (!boot_pat_state) | |
169 | rdmsrl(MSR_IA32_CR_PAT, boot_pat_state); | |
170 | ||
171 | wrmsrl(MSR_IA32_CR_PAT, pat); | |
172 | ||
173 | if (boot_cpu) | |
174 | pat_init_cache_modes(); | |
175 | } | |
176 | ||
177 | #undef PAT | |
178 | ||
179 | static DEFINE_SPINLOCK(memtype_lock); /* protects memtype accesses */ | |
180 | ||
181 | /* | |
182 | * Does intersection of PAT memory type and MTRR memory type and returns | |
183 | * the resulting memory type as PAT understands it. | |
184 | * (Type in pat and mtrr will not have same value) | |
185 | * The intersection is based on "Effective Memory Type" tables in IA-32 | |
186 | * SDM vol 3a | |
187 | */ | |
188 | static unsigned long pat_x_mtrr_type(u64 start, u64 end, | |
189 | enum page_cache_mode req_type) | |
190 | { | |
191 | /* | |
192 | * Look for MTRR hint to get the effective type in case where PAT | |
193 | * request is for WB. | |
194 | */ | |
195 | if (req_type == _PAGE_CACHE_MODE_WB) { | |
196 | u8 mtrr_type; | |
197 | ||
198 | mtrr_type = mtrr_type_lookup(start, end); | |
199 | if (mtrr_type != MTRR_TYPE_WRBACK) | |
200 | return _PAGE_CACHE_MODE_UC_MINUS; | |
201 | ||
202 | return _PAGE_CACHE_MODE_WB; | |
203 | } | |
204 | ||
205 | return req_type; | |
206 | } | |
207 | ||
208 | struct pagerange_state { | |
209 | unsigned long cur_pfn; | |
210 | int ram; | |
211 | int not_ram; | |
212 | }; | |
213 | ||
214 | static int | |
215 | pagerange_is_ram_callback(unsigned long initial_pfn, unsigned long total_nr_pages, void *arg) | |
216 | { | |
217 | struct pagerange_state *state = arg; | |
218 | ||
219 | state->not_ram |= initial_pfn > state->cur_pfn; | |
220 | state->ram |= total_nr_pages > 0; | |
221 | state->cur_pfn = initial_pfn + total_nr_pages; | |
222 | ||
223 | return state->ram && state->not_ram; | |
224 | } | |
225 | ||
226 | static int pat_pagerange_is_ram(resource_size_t start, resource_size_t end) | |
227 | { | |
228 | int ret = 0; | |
229 | unsigned long start_pfn = start >> PAGE_SHIFT; | |
230 | unsigned long end_pfn = (end + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
231 | struct pagerange_state state = {start_pfn, 0, 0}; | |
232 | ||
233 | /* | |
234 | * For legacy reasons, physical address range in the legacy ISA | |
235 | * region is tracked as non-RAM. This will allow users of | |
236 | * /dev/mem to map portions of legacy ISA region, even when | |
237 | * some of those portions are listed(or not even listed) with | |
238 | * different e820 types(RAM/reserved/..) | |
239 | */ | |
240 | if (start_pfn < ISA_END_ADDRESS >> PAGE_SHIFT) | |
241 | start_pfn = ISA_END_ADDRESS >> PAGE_SHIFT; | |
242 | ||
243 | if (start_pfn < end_pfn) { | |
244 | ret = walk_system_ram_range(start_pfn, end_pfn - start_pfn, | |
245 | &state, pagerange_is_ram_callback); | |
246 | } | |
247 | ||
248 | return (ret > 0) ? -1 : (state.ram ? 1 : 0); | |
249 | } | |
250 | ||
251 | /* | |
252 | * For RAM pages, we use page flags to mark the pages with appropriate type. | |
253 | * Here we do two pass: | |
254 | * - Find the memtype of all the pages in the range, look for any conflicts | |
255 | * - In case of no conflicts, set the new memtype for pages in the range | |
256 | */ | |
257 | static int reserve_ram_pages_type(u64 start, u64 end, | |
258 | enum page_cache_mode req_type, | |
259 | enum page_cache_mode *new_type) | |
260 | { | |
261 | struct page *page; | |
262 | u64 pfn; | |
263 | ||
264 | if (req_type == _PAGE_CACHE_MODE_UC) { | |
265 | /* We do not support strong UC */ | |
266 | WARN_ON_ONCE(1); | |
267 | req_type = _PAGE_CACHE_MODE_UC_MINUS; | |
268 | } | |
269 | ||
270 | for (pfn = (start >> PAGE_SHIFT); pfn < (end >> PAGE_SHIFT); ++pfn) { | |
271 | enum page_cache_mode type; | |
272 | ||
273 | page = pfn_to_page(pfn); | |
274 | type = get_page_memtype(page); | |
275 | if (type != -1) { | |
276 | pr_info("reserve_ram_pages_type failed [mem %#010Lx-%#010Lx], track 0x%x, req 0x%x\n", | |
277 | start, end - 1, type, req_type); | |
278 | if (new_type) | |
279 | *new_type = type; | |
280 | ||
281 | return -EBUSY; | |
282 | } | |
283 | } | |
284 | ||
285 | if (new_type) | |
286 | *new_type = req_type; | |
287 | ||
288 | for (pfn = (start >> PAGE_SHIFT); pfn < (end >> PAGE_SHIFT); ++pfn) { | |
289 | page = pfn_to_page(pfn); | |
290 | set_page_memtype(page, req_type); | |
291 | } | |
292 | return 0; | |
293 | } | |
294 | ||
295 | static int free_ram_pages_type(u64 start, u64 end) | |
296 | { | |
297 | struct page *page; | |
298 | u64 pfn; | |
299 | ||
300 | for (pfn = (start >> PAGE_SHIFT); pfn < (end >> PAGE_SHIFT); ++pfn) { | |
301 | page = pfn_to_page(pfn); | |
302 | set_page_memtype(page, -1); | |
303 | } | |
304 | return 0; | |
305 | } | |
306 | ||
307 | /* | |
308 | * req_type typically has one of the: | |
309 | * - _PAGE_CACHE_MODE_WB | |
310 | * - _PAGE_CACHE_MODE_WC | |
311 | * - _PAGE_CACHE_MODE_UC_MINUS | |
312 | * - _PAGE_CACHE_MODE_UC | |
313 | * | |
314 | * If new_type is NULL, function will return an error if it cannot reserve the | |
315 | * region with req_type. If new_type is non-NULL, function will return | |
316 | * available type in new_type in case of no error. In case of any error | |
317 | * it will return a negative return value. | |
318 | */ | |
319 | int reserve_memtype(u64 start, u64 end, enum page_cache_mode req_type, | |
320 | enum page_cache_mode *new_type) | |
321 | { | |
322 | struct memtype *new; | |
323 | enum page_cache_mode actual_type; | |
324 | int is_range_ram; | |
325 | int err = 0; | |
326 | ||
327 | BUG_ON(start >= end); /* end is exclusive */ | |
328 | ||
329 | if (!pat_enabled) { | |
330 | /* This is identical to page table setting without PAT */ | |
331 | if (new_type) { | |
332 | if (req_type == _PAGE_CACHE_MODE_WC) | |
333 | *new_type = _PAGE_CACHE_MODE_UC_MINUS; | |
334 | else | |
335 | *new_type = req_type; | |
336 | } | |
337 | return 0; | |
338 | } | |
339 | ||
340 | /* Low ISA region is always mapped WB in page table. No need to track */ | |
341 | if (x86_platform.is_untracked_pat_range(start, end)) { | |
342 | if (new_type) | |
343 | *new_type = _PAGE_CACHE_MODE_WB; | |
344 | return 0; | |
345 | } | |
346 | ||
347 | /* | |
348 | * Call mtrr_lookup to get the type hint. This is an | |
349 | * optimization for /dev/mem mmap'ers into WB memory (BIOS | |
350 | * tools and ACPI tools). Use WB request for WB memory and use | |
351 | * UC_MINUS otherwise. | |
352 | */ | |
353 | actual_type = pat_x_mtrr_type(start, end, req_type); | |
354 | ||
355 | if (new_type) | |
356 | *new_type = actual_type; | |
357 | ||
358 | is_range_ram = pat_pagerange_is_ram(start, end); | |
359 | if (is_range_ram == 1) { | |
360 | ||
361 | err = reserve_ram_pages_type(start, end, req_type, new_type); | |
362 | ||
363 | return err; | |
364 | } else if (is_range_ram < 0) { | |
365 | return -EINVAL; | |
366 | } | |
367 | ||
368 | new = kzalloc(sizeof(struct memtype), GFP_KERNEL); | |
369 | if (!new) | |
370 | return -ENOMEM; | |
371 | ||
372 | new->start = start; | |
373 | new->end = end; | |
374 | new->type = actual_type; | |
375 | ||
376 | spin_lock(&memtype_lock); | |
377 | ||
378 | err = rbt_memtype_check_insert(new, new_type); | |
379 | if (err) { | |
380 | printk(KERN_INFO "reserve_memtype failed [mem %#010Lx-%#010Lx], track %s, req %s\n", | |
381 | start, end - 1, | |
382 | cattr_name(new->type), cattr_name(req_type)); | |
383 | kfree(new); | |
384 | spin_unlock(&memtype_lock); | |
385 | ||
386 | return err; | |
387 | } | |
388 | ||
389 | spin_unlock(&memtype_lock); | |
390 | ||
391 | dprintk("reserve_memtype added [mem %#010Lx-%#010Lx], track %s, req %s, ret %s\n", | |
392 | start, end - 1, cattr_name(new->type), cattr_name(req_type), | |
393 | new_type ? cattr_name(*new_type) : "-"); | |
394 | ||
395 | return err; | |
396 | } | |
397 | ||
398 | int free_memtype(u64 start, u64 end) | |
399 | { | |
400 | int err = -EINVAL; | |
401 | int is_range_ram; | |
402 | struct memtype *entry; | |
403 | ||
404 | if (!pat_enabled) | |
405 | return 0; | |
406 | ||
407 | /* Low ISA region is always mapped WB. No need to track */ | |
408 | if (x86_platform.is_untracked_pat_range(start, end)) | |
409 | return 0; | |
410 | ||
411 | is_range_ram = pat_pagerange_is_ram(start, end); | |
412 | if (is_range_ram == 1) { | |
413 | ||
414 | err = free_ram_pages_type(start, end); | |
415 | ||
416 | return err; | |
417 | } else if (is_range_ram < 0) { | |
418 | return -EINVAL; | |
419 | } | |
420 | ||
421 | spin_lock(&memtype_lock); | |
422 | entry = rbt_memtype_erase(start, end); | |
423 | spin_unlock(&memtype_lock); | |
424 | ||
425 | if (!entry) { | |
426 | printk(KERN_INFO "%s:%d freeing invalid memtype [mem %#010Lx-%#010Lx]\n", | |
427 | current->comm, current->pid, start, end - 1); | |
428 | return -EINVAL; | |
429 | } | |
430 | ||
431 | kfree(entry); | |
432 | ||
433 | dprintk("free_memtype request [mem %#010Lx-%#010Lx]\n", start, end - 1); | |
434 | ||
435 | return 0; | |
436 | } | |
437 | ||
438 | ||
439 | /** | |
440 | * lookup_memtype - Looksup the memory type for a physical address | |
441 | * @paddr: physical address of which memory type needs to be looked up | |
442 | * | |
443 | * Only to be called when PAT is enabled | |
444 | * | |
445 | * Returns _PAGE_CACHE_MODE_WB, _PAGE_CACHE_MODE_WC, _PAGE_CACHE_MODE_UC_MINUS | |
446 | * or _PAGE_CACHE_MODE_UC | |
447 | */ | |
448 | static enum page_cache_mode lookup_memtype(u64 paddr) | |
449 | { | |
450 | enum page_cache_mode rettype = _PAGE_CACHE_MODE_WB; | |
451 | struct memtype *entry; | |
452 | ||
453 | if (x86_platform.is_untracked_pat_range(paddr, paddr + PAGE_SIZE)) | |
454 | return rettype; | |
455 | ||
456 | if (pat_pagerange_is_ram(paddr, paddr + PAGE_SIZE)) { | |
457 | struct page *page; | |
458 | page = pfn_to_page(paddr >> PAGE_SHIFT); | |
459 | rettype = get_page_memtype(page); | |
460 | /* | |
461 | * -1 from get_page_memtype() implies RAM page is in its | |
462 | * default state and not reserved, and hence of type WB | |
463 | */ | |
464 | if (rettype == -1) | |
465 | rettype = _PAGE_CACHE_MODE_WB; | |
466 | ||
467 | return rettype; | |
468 | } | |
469 | ||
470 | spin_lock(&memtype_lock); | |
471 | ||
472 | entry = rbt_memtype_lookup(paddr); | |
473 | if (entry != NULL) | |
474 | rettype = entry->type; | |
475 | else | |
476 | rettype = _PAGE_CACHE_MODE_UC_MINUS; | |
477 | ||
478 | spin_unlock(&memtype_lock); | |
479 | return rettype; | |
480 | } | |
481 | ||
482 | /** | |
483 | * io_reserve_memtype - Request a memory type mapping for a region of memory | |
484 | * @start: start (physical address) of the region | |
485 | * @end: end (physical address) of the region | |
486 | * @type: A pointer to memtype, with requested type. On success, requested | |
487 | * or any other compatible type that was available for the region is returned | |
488 | * | |
489 | * On success, returns 0 | |
490 | * On failure, returns non-zero | |
491 | */ | |
492 | int io_reserve_memtype(resource_size_t start, resource_size_t end, | |
493 | enum page_cache_mode *type) | |
494 | { | |
495 | resource_size_t size = end - start; | |
496 | enum page_cache_mode req_type = *type; | |
497 | enum page_cache_mode new_type; | |
498 | int ret; | |
499 | ||
500 | WARN_ON_ONCE(iomem_map_sanity_check(start, size)); | |
501 | ||
502 | ret = reserve_memtype(start, end, req_type, &new_type); | |
503 | if (ret) | |
504 | goto out_err; | |
505 | ||
506 | if (!is_new_memtype_allowed(start, size, req_type, new_type)) | |
507 | goto out_free; | |
508 | ||
509 | if (kernel_map_sync_memtype(start, size, new_type) < 0) | |
510 | goto out_free; | |
511 | ||
512 | *type = new_type; | |
513 | return 0; | |
514 | ||
515 | out_free: | |
516 | free_memtype(start, end); | |
517 | ret = -EBUSY; | |
518 | out_err: | |
519 | return ret; | |
520 | } | |
521 | ||
522 | /** | |
523 | * io_free_memtype - Release a memory type mapping for a region of memory | |
524 | * @start: start (physical address) of the region | |
525 | * @end: end (physical address) of the region | |
526 | */ | |
527 | void io_free_memtype(resource_size_t start, resource_size_t end) | |
528 | { | |
529 | free_memtype(start, end); | |
530 | } | |
531 | ||
532 | pgprot_t phys_mem_access_prot(struct file *file, unsigned long pfn, | |
533 | unsigned long size, pgprot_t vma_prot) | |
534 | { | |
535 | return vma_prot; | |
536 | } | |
537 | ||
538 | #ifdef CONFIG_STRICT_DEVMEM | |
539 | /* This check is done in drivers/char/mem.c in case of STRICT_DEVMEM*/ | |
540 | static inline int range_is_allowed(unsigned long pfn, unsigned long size) | |
541 | { | |
542 | return 1; | |
543 | } | |
544 | #else | |
545 | /* This check is needed to avoid cache aliasing when PAT is enabled */ | |
546 | static inline int range_is_allowed(unsigned long pfn, unsigned long size) | |
547 | { | |
548 | u64 from = ((u64)pfn) << PAGE_SHIFT; | |
549 | u64 to = from + size; | |
550 | u64 cursor = from; | |
551 | ||
552 | if (!pat_enabled) | |
553 | return 1; | |
554 | ||
555 | while (cursor < to) { | |
556 | if (!devmem_is_allowed(pfn)) { | |
557 | printk(KERN_INFO "Program %s tried to access /dev/mem between [mem %#010Lx-%#010Lx]\n", | |
558 | current->comm, from, to - 1); | |
559 | return 0; | |
560 | } | |
561 | cursor += PAGE_SIZE; | |
562 | pfn++; | |
563 | } | |
564 | return 1; | |
565 | } | |
566 | #endif /* CONFIG_STRICT_DEVMEM */ | |
567 | ||
568 | int phys_mem_access_prot_allowed(struct file *file, unsigned long pfn, | |
569 | unsigned long size, pgprot_t *vma_prot) | |
570 | { | |
571 | enum page_cache_mode pcm = _PAGE_CACHE_MODE_WB; | |
572 | ||
573 | if (!range_is_allowed(pfn, size)) | |
574 | return 0; | |
575 | ||
576 | if (file->f_flags & O_DSYNC) | |
577 | pcm = _PAGE_CACHE_MODE_UC_MINUS; | |
578 | ||
579 | #ifdef CONFIG_X86_32 | |
580 | /* | |
581 | * On the PPro and successors, the MTRRs are used to set | |
582 | * memory types for physical addresses outside main memory, | |
583 | * so blindly setting UC or PWT on those pages is wrong. | |
584 | * For Pentiums and earlier, the surround logic should disable | |
585 | * caching for the high addresses through the KEN pin, but | |
586 | * we maintain the tradition of paranoia in this code. | |
587 | */ | |
588 | if (!pat_enabled && | |
589 | !(boot_cpu_has(X86_FEATURE_MTRR) || | |
590 | boot_cpu_has(X86_FEATURE_K6_MTRR) || | |
591 | boot_cpu_has(X86_FEATURE_CYRIX_ARR) || | |
592 | boot_cpu_has(X86_FEATURE_CENTAUR_MCR)) && | |
593 | (pfn << PAGE_SHIFT) >= __pa(high_memory)) { | |
594 | pcm = _PAGE_CACHE_MODE_UC; | |
595 | } | |
596 | #endif | |
597 | ||
598 | *vma_prot = __pgprot((pgprot_val(*vma_prot) & ~_PAGE_CACHE_MASK) | | |
599 | cachemode2protval(pcm)); | |
600 | return 1; | |
601 | } | |
602 | ||
603 | /* | |
604 | * Change the memory type for the physial address range in kernel identity | |
605 | * mapping space if that range is a part of identity map. | |
606 | */ | |
607 | int kernel_map_sync_memtype(u64 base, unsigned long size, | |
608 | enum page_cache_mode pcm) | |
609 | { | |
610 | unsigned long id_sz; | |
611 | ||
612 | if (base > __pa(high_memory-1)) | |
613 | return 0; | |
614 | ||
615 | /* | |
616 | * some areas in the middle of the kernel identity range | |
617 | * are not mapped, like the PCI space. | |
618 | */ | |
619 | if (!page_is_ram(base >> PAGE_SHIFT)) | |
620 | return 0; | |
621 | ||
622 | id_sz = (__pa(high_memory-1) <= base + size) ? | |
623 | __pa(high_memory) - base : | |
624 | size; | |
625 | ||
626 | if (ioremap_change_attr((unsigned long)__va(base), id_sz, pcm) < 0) { | |
627 | printk(KERN_INFO "%s:%d ioremap_change_attr failed %s " | |
628 | "for [mem %#010Lx-%#010Lx]\n", | |
629 | current->comm, current->pid, | |
630 | cattr_name(pcm), | |
631 | base, (unsigned long long)(base + size-1)); | |
632 | return -EINVAL; | |
633 | } | |
634 | return 0; | |
635 | } | |
636 | ||
637 | /* | |
638 | * Internal interface to reserve a range of physical memory with prot. | |
639 | * Reserved non RAM regions only and after successful reserve_memtype, | |
640 | * this func also keeps identity mapping (if any) in sync with this new prot. | |
641 | */ | |
642 | static int reserve_pfn_range(u64 paddr, unsigned long size, pgprot_t *vma_prot, | |
643 | int strict_prot) | |
644 | { | |
645 | int is_ram = 0; | |
646 | int ret; | |
647 | enum page_cache_mode want_pcm = pgprot2cachemode(*vma_prot); | |
648 | enum page_cache_mode pcm = want_pcm; | |
649 | ||
650 | is_ram = pat_pagerange_is_ram(paddr, paddr + size); | |
651 | ||
652 | /* | |
653 | * reserve_pfn_range() for RAM pages. We do not refcount to keep | |
654 | * track of number of mappings of RAM pages. We can assert that | |
655 | * the type requested matches the type of first page in the range. | |
656 | */ | |
657 | if (is_ram) { | |
658 | if (!pat_enabled) | |
659 | return 0; | |
660 | ||
661 | pcm = lookup_memtype(paddr); | |
662 | if (want_pcm != pcm) { | |
663 | printk(KERN_WARNING "%s:%d map pfn RAM range req %s for [mem %#010Lx-%#010Lx], got %s\n", | |
664 | current->comm, current->pid, | |
665 | cattr_name(want_pcm), | |
666 | (unsigned long long)paddr, | |
667 | (unsigned long long)(paddr + size - 1), | |
668 | cattr_name(pcm)); | |
669 | *vma_prot = __pgprot((pgprot_val(*vma_prot) & | |
670 | (~_PAGE_CACHE_MASK)) | | |
671 | cachemode2protval(pcm)); | |
672 | } | |
673 | return 0; | |
674 | } | |
675 | ||
676 | ret = reserve_memtype(paddr, paddr + size, want_pcm, &pcm); | |
677 | if (ret) | |
678 | return ret; | |
679 | ||
680 | if (pcm != want_pcm) { | |
681 | if (strict_prot || | |
682 | !is_new_memtype_allowed(paddr, size, want_pcm, pcm)) { | |
683 | free_memtype(paddr, paddr + size); | |
684 | printk(KERN_ERR "%s:%d map pfn expected mapping type %s" | |
685 | " for [mem %#010Lx-%#010Lx], got %s\n", | |
686 | current->comm, current->pid, | |
687 | cattr_name(want_pcm), | |
688 | (unsigned long long)paddr, | |
689 | (unsigned long long)(paddr + size - 1), | |
690 | cattr_name(pcm)); | |
691 | return -EINVAL; | |
692 | } | |
693 | /* | |
694 | * We allow returning different type than the one requested in | |
695 | * non strict case. | |
696 | */ | |
697 | *vma_prot = __pgprot((pgprot_val(*vma_prot) & | |
698 | (~_PAGE_CACHE_MASK)) | | |
699 | cachemode2protval(pcm)); | |
700 | } | |
701 | ||
702 | if (kernel_map_sync_memtype(paddr, size, pcm) < 0) { | |
703 | free_memtype(paddr, paddr + size); | |
704 | return -EINVAL; | |
705 | } | |
706 | return 0; | |
707 | } | |
708 | ||
709 | /* | |
710 | * Internal interface to free a range of physical memory. | |
711 | * Frees non RAM regions only. | |
712 | */ | |
713 | static void free_pfn_range(u64 paddr, unsigned long size) | |
714 | { | |
715 | int is_ram; | |
716 | ||
717 | is_ram = pat_pagerange_is_ram(paddr, paddr + size); | |
718 | if (is_ram == 0) | |
719 | free_memtype(paddr, paddr + size); | |
720 | } | |
721 | ||
722 | /* | |
723 | * track_pfn_copy is called when vma that is covering the pfnmap gets | |
724 | * copied through copy_page_range(). | |
725 | * | |
726 | * If the vma has a linear pfn mapping for the entire range, we get the prot | |
727 | * from pte and reserve the entire vma range with single reserve_pfn_range call. | |
728 | */ | |
729 | int track_pfn_copy(struct vm_area_struct *vma) | |
730 | { | |
731 | resource_size_t paddr; | |
732 | unsigned long prot; | |
733 | unsigned long vma_size = vma->vm_end - vma->vm_start; | |
734 | pgprot_t pgprot; | |
735 | ||
736 | if (vma->vm_flags & VM_PAT) { | |
737 | /* | |
738 | * reserve the whole chunk covered by vma. We need the | |
739 | * starting address and protection from pte. | |
740 | */ | |
741 | if (follow_phys(vma, vma->vm_start, 0, &prot, &paddr)) { | |
742 | WARN_ON_ONCE(1); | |
743 | return -EINVAL; | |
744 | } | |
745 | pgprot = __pgprot(prot); | |
746 | return reserve_pfn_range(paddr, vma_size, &pgprot, 1); | |
747 | } | |
748 | ||
749 | return 0; | |
750 | } | |
751 | ||
752 | /* | |
753 | * prot is passed in as a parameter for the new mapping. If the vma has a | |
754 | * linear pfn mapping for the entire range reserve the entire vma range with | |
755 | * single reserve_pfn_range call. | |
756 | */ | |
757 | int track_pfn_remap(struct vm_area_struct *vma, pgprot_t *prot, | |
758 | unsigned long pfn, unsigned long addr, unsigned long size) | |
759 | { | |
760 | resource_size_t paddr = (resource_size_t)pfn << PAGE_SHIFT; | |
761 | enum page_cache_mode pcm; | |
762 | ||
763 | /* reserve the whole chunk starting from paddr */ | |
764 | if (addr == vma->vm_start && size == (vma->vm_end - vma->vm_start)) { | |
765 | int ret; | |
766 | ||
767 | ret = reserve_pfn_range(paddr, size, prot, 0); | |
768 | if (!ret) | |
769 | vma->vm_flags |= VM_PAT; | |
770 | return ret; | |
771 | } | |
772 | ||
773 | if (!pat_enabled) | |
774 | return 0; | |
775 | ||
776 | /* | |
777 | * For anything smaller than the vma size we set prot based on the | |
778 | * lookup. | |
779 | */ | |
780 | pcm = lookup_memtype(paddr); | |
781 | ||
782 | /* Check memtype for the remaining pages */ | |
783 | while (size > PAGE_SIZE) { | |
784 | size -= PAGE_SIZE; | |
785 | paddr += PAGE_SIZE; | |
786 | if (pcm != lookup_memtype(paddr)) | |
787 | return -EINVAL; | |
788 | } | |
789 | ||
790 | *prot = __pgprot((pgprot_val(vma->vm_page_prot) & (~_PAGE_CACHE_MASK)) | | |
791 | cachemode2protval(pcm)); | |
792 | ||
793 | return 0; | |
794 | } | |
795 | ||
796 | int track_pfn_insert(struct vm_area_struct *vma, pgprot_t *prot, | |
797 | unsigned long pfn) | |
798 | { | |
799 | enum page_cache_mode pcm; | |
800 | ||
801 | if (!pat_enabled) | |
802 | return 0; | |
803 | ||
804 | /* Set prot based on lookup */ | |
805 | pcm = lookup_memtype((resource_size_t)pfn << PAGE_SHIFT); | |
806 | *prot = __pgprot((pgprot_val(vma->vm_page_prot) & (~_PAGE_CACHE_MASK)) | | |
807 | cachemode2protval(pcm)); | |
808 | ||
809 | return 0; | |
810 | } | |
811 | ||
812 | /* | |
813 | * untrack_pfn is called while unmapping a pfnmap for a region. | |
814 | * untrack can be called for a specific region indicated by pfn and size or | |
815 | * can be for the entire vma (in which case pfn, size are zero). | |
816 | */ | |
817 | void untrack_pfn(struct vm_area_struct *vma, unsigned long pfn, | |
818 | unsigned long size) | |
819 | { | |
820 | resource_size_t paddr; | |
821 | unsigned long prot; | |
822 | ||
823 | if (!(vma->vm_flags & VM_PAT)) | |
824 | return; | |
825 | ||
826 | /* free the chunk starting from pfn or the whole chunk */ | |
827 | paddr = (resource_size_t)pfn << PAGE_SHIFT; | |
828 | if (!paddr && !size) { | |
829 | if (follow_phys(vma, vma->vm_start, 0, &prot, &paddr)) { | |
830 | WARN_ON_ONCE(1); | |
831 | return; | |
832 | } | |
833 | ||
834 | size = vma->vm_end - vma->vm_start; | |
835 | } | |
836 | free_pfn_range(paddr, size); | |
837 | vma->vm_flags &= ~VM_PAT; | |
838 | } | |
839 | ||
840 | pgprot_t pgprot_writecombine(pgprot_t prot) | |
841 | { | |
842 | if (pat_enabled) | |
843 | return __pgprot(pgprot_val(prot) | | |
844 | cachemode2protval(_PAGE_CACHE_MODE_WC)); | |
845 | else | |
846 | return pgprot_noncached(prot); | |
847 | } | |
848 | EXPORT_SYMBOL_GPL(pgprot_writecombine); | |
849 | ||
850 | #if defined(CONFIG_DEBUG_FS) && defined(CONFIG_X86_PAT) | |
851 | ||
852 | static struct memtype *memtype_get_idx(loff_t pos) | |
853 | { | |
854 | struct memtype *print_entry; | |
855 | int ret; | |
856 | ||
857 | print_entry = kzalloc(sizeof(struct memtype), GFP_KERNEL); | |
858 | if (!print_entry) | |
859 | return NULL; | |
860 | ||
861 | spin_lock(&memtype_lock); | |
862 | ret = rbt_memtype_copy_nth_element(print_entry, pos); | |
863 | spin_unlock(&memtype_lock); | |
864 | ||
865 | if (!ret) { | |
866 | return print_entry; | |
867 | } else { | |
868 | kfree(print_entry); | |
869 | return NULL; | |
870 | } | |
871 | } | |
872 | ||
873 | static void *memtype_seq_start(struct seq_file *seq, loff_t *pos) | |
874 | { | |
875 | if (*pos == 0) { | |
876 | ++*pos; | |
877 | seq_printf(seq, "PAT memtype list:\n"); | |
878 | } | |
879 | ||
880 | return memtype_get_idx(*pos); | |
881 | } | |
882 | ||
883 | static void *memtype_seq_next(struct seq_file *seq, void *v, loff_t *pos) | |
884 | { | |
885 | ++*pos; | |
886 | return memtype_get_idx(*pos); | |
887 | } | |
888 | ||
889 | static void memtype_seq_stop(struct seq_file *seq, void *v) | |
890 | { | |
891 | } | |
892 | ||
893 | static int memtype_seq_show(struct seq_file *seq, void *v) | |
894 | { | |
895 | struct memtype *print_entry = (struct memtype *)v; | |
896 | ||
897 | seq_printf(seq, "%s @ 0x%Lx-0x%Lx\n", cattr_name(print_entry->type), | |
898 | print_entry->start, print_entry->end); | |
899 | kfree(print_entry); | |
900 | ||
901 | return 0; | |
902 | } | |
903 | ||
904 | static const struct seq_operations memtype_seq_ops = { | |
905 | .start = memtype_seq_start, | |
906 | .next = memtype_seq_next, | |
907 | .stop = memtype_seq_stop, | |
908 | .show = memtype_seq_show, | |
909 | }; | |
910 | ||
911 | static int memtype_seq_open(struct inode *inode, struct file *file) | |
912 | { | |
913 | return seq_open(file, &memtype_seq_ops); | |
914 | } | |
915 | ||
916 | static const struct file_operations memtype_fops = { | |
917 | .open = memtype_seq_open, | |
918 | .read = seq_read, | |
919 | .llseek = seq_lseek, | |
920 | .release = seq_release, | |
921 | }; | |
922 | ||
923 | static int __init pat_memtype_list_init(void) | |
924 | { | |
925 | if (pat_enabled) { | |
926 | debugfs_create_file("pat_memtype_list", S_IRUSR, | |
927 | arch_debugfs_dir, NULL, &memtype_fops); | |
928 | } | |
929 | return 0; | |
930 | } | |
931 | ||
932 | late_initcall(pat_memtype_list_init); | |
933 | ||
934 | #endif /* CONFIG_DEBUG_FS && CONFIG_X86_PAT */ |