Loading...
Note: File does not exist in v6.2.
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/pfn_t.h>
16#include <linux/slab.h>
17#include <linux/mm.h>
18#include <linux/fs.h>
19#include <linux/rbtree.h>
20
21#include <asm/cacheflush.h>
22#include <asm/processor.h>
23#include <asm/tlbflush.h>
24#include <asm/x86_init.h>
25#include <asm/pgtable.h>
26#include <asm/fcntl.h>
27#include <asm/e820.h>
28#include <asm/mtrr.h>
29#include <asm/page.h>
30#include <asm/msr.h>
31#include <asm/pat.h>
32#include <asm/io.h>
33
34#include "pat_internal.h"
35#include "mm_internal.h"
36
37#undef pr_fmt
38#define pr_fmt(fmt) "" fmt
39
40static bool boot_cpu_done;
41
42static int __read_mostly __pat_enabled = IS_ENABLED(CONFIG_X86_PAT);
43
44static inline void pat_disable(const char *reason)
45{
46 __pat_enabled = 0;
47 pr_info("x86/PAT: %s\n", reason);
48}
49
50static int __init nopat(char *str)
51{
52 pat_disable("PAT support disabled.");
53 return 0;
54}
55early_param("nopat", nopat);
56
57bool pat_enabled(void)
58{
59 return !!__pat_enabled;
60}
61EXPORT_SYMBOL_GPL(pat_enabled);
62
63int pat_debug_enable;
64
65static int __init pat_debug_setup(char *str)
66{
67 pat_debug_enable = 1;
68 return 0;
69}
70__setup("debugpat", pat_debug_setup);
71
72#ifdef CONFIG_X86_PAT
73/*
74 * X86 PAT uses page flags arch_1 and uncached together to keep track of
75 * memory type of pages that have backing page struct.
76 *
77 * X86 PAT supports 4 different memory types:
78 * - _PAGE_CACHE_MODE_WB
79 * - _PAGE_CACHE_MODE_WC
80 * - _PAGE_CACHE_MODE_UC_MINUS
81 * - _PAGE_CACHE_MODE_WT
82 *
83 * _PAGE_CACHE_MODE_WB is the default type.
84 */
85
86#define _PGMT_WB 0
87#define _PGMT_WC (1UL << PG_arch_1)
88#define _PGMT_UC_MINUS (1UL << PG_uncached)
89#define _PGMT_WT (1UL << PG_uncached | 1UL << PG_arch_1)
90#define _PGMT_MASK (1UL << PG_uncached | 1UL << PG_arch_1)
91#define _PGMT_CLEAR_MASK (~_PGMT_MASK)
92
93static inline enum page_cache_mode get_page_memtype(struct page *pg)
94{
95 unsigned long pg_flags = pg->flags & _PGMT_MASK;
96
97 if (pg_flags == _PGMT_WB)
98 return _PAGE_CACHE_MODE_WB;
99 else if (pg_flags == _PGMT_WC)
100 return _PAGE_CACHE_MODE_WC;
101 else if (pg_flags == _PGMT_UC_MINUS)
102 return _PAGE_CACHE_MODE_UC_MINUS;
103 else
104 return _PAGE_CACHE_MODE_WT;
105}
106
107static inline void set_page_memtype(struct page *pg,
108 enum page_cache_mode memtype)
109{
110 unsigned long memtype_flags;
111 unsigned long old_flags;
112 unsigned long new_flags;
113
114 switch (memtype) {
115 case _PAGE_CACHE_MODE_WC:
116 memtype_flags = _PGMT_WC;
117 break;
118 case _PAGE_CACHE_MODE_UC_MINUS:
119 memtype_flags = _PGMT_UC_MINUS;
120 break;
121 case _PAGE_CACHE_MODE_WT:
122 memtype_flags = _PGMT_WT;
123 break;
124 case _PAGE_CACHE_MODE_WB:
125 default:
126 memtype_flags = _PGMT_WB;
127 break;
128 }
129
130 do {
131 old_flags = pg->flags;
132 new_flags = (old_flags & _PGMT_CLEAR_MASK) | memtype_flags;
133 } while (cmpxchg(&pg->flags, old_flags, new_flags) != old_flags);
134}
135#else
136static inline enum page_cache_mode get_page_memtype(struct page *pg)
137{
138 return -1;
139}
140static inline void set_page_memtype(struct page *pg,
141 enum page_cache_mode memtype)
142{
143}
144#endif
145
146enum {
147 PAT_UC = 0, /* uncached */
148 PAT_WC = 1, /* Write combining */
149 PAT_WT = 4, /* Write Through */
150 PAT_WP = 5, /* Write Protected */
151 PAT_WB = 6, /* Write Back (default) */
152 PAT_UC_MINUS = 7, /* UC, but can be overridden by MTRR */
153};
154
155#define CM(c) (_PAGE_CACHE_MODE_ ## c)
156
157static enum page_cache_mode pat_get_cache_mode(unsigned pat_val, char *msg)
158{
159 enum page_cache_mode cache;
160 char *cache_mode;
161
162 switch (pat_val) {
163 case PAT_UC: cache = CM(UC); cache_mode = "UC "; break;
164 case PAT_WC: cache = CM(WC); cache_mode = "WC "; break;
165 case PAT_WT: cache = CM(WT); cache_mode = "WT "; break;
166 case PAT_WP: cache = CM(WP); cache_mode = "WP "; break;
167 case PAT_WB: cache = CM(WB); cache_mode = "WB "; break;
168 case PAT_UC_MINUS: cache = CM(UC_MINUS); cache_mode = "UC- "; break;
169 default: cache = CM(WB); cache_mode = "WB "; break;
170 }
171
172 memcpy(msg, cache_mode, 4);
173
174 return cache;
175}
176
177#undef CM
178
179/*
180 * Update the cache mode to pgprot translation tables according to PAT
181 * configuration.
182 * Using lower indices is preferred, so we start with highest index.
183 */
184void pat_init_cache_modes(u64 pat)
185{
186 enum page_cache_mode cache;
187 char pat_msg[33];
188 int i;
189
190 pat_msg[32] = 0;
191 for (i = 7; i >= 0; i--) {
192 cache = pat_get_cache_mode((pat >> (i * 8)) & 7,
193 pat_msg + 4 * i);
194 update_cache_mode_entry(i, cache);
195 }
196 pr_info("x86/PAT: Configuration [0-7]: %s\n", pat_msg);
197}
198
199#define PAT(x, y) ((u64)PAT_ ## y << ((x)*8))
200
201static void pat_bsp_init(u64 pat)
202{
203 u64 tmp_pat;
204
205 if (!cpu_has_pat) {
206 pat_disable("PAT not supported by CPU.");
207 return;
208 }
209
210 if (!pat_enabled())
211 goto done;
212
213 rdmsrl(MSR_IA32_CR_PAT, tmp_pat);
214 if (!tmp_pat) {
215 pat_disable("PAT MSR is 0, disabled.");
216 return;
217 }
218
219 wrmsrl(MSR_IA32_CR_PAT, pat);
220
221done:
222 pat_init_cache_modes(pat);
223}
224
225static void pat_ap_init(u64 pat)
226{
227 if (!pat_enabled())
228 return;
229
230 if (!cpu_has_pat) {
231 /*
232 * If this happens we are on a secondary CPU, but switched to
233 * PAT on the boot CPU. We have no way to undo PAT.
234 */
235 panic("x86/PAT: PAT enabled, but not supported by secondary CPU\n");
236 }
237
238 wrmsrl(MSR_IA32_CR_PAT, pat);
239}
240
241void pat_init(void)
242{
243 u64 pat;
244 struct cpuinfo_x86 *c = &boot_cpu_data;
245
246 if (!pat_enabled()) {
247 /*
248 * No PAT. Emulate the PAT table that corresponds to the two
249 * cache bits, PWT (Write Through) and PCD (Cache Disable). This
250 * setup is the same as the BIOS default setup when the system
251 * has PAT but the "nopat" boot option has been specified. This
252 * emulated PAT table is used when MSR_IA32_CR_PAT returns 0.
253 *
254 * PTE encoding:
255 *
256 * PCD
257 * |PWT PAT
258 * || slot
259 * 00 0 WB : _PAGE_CACHE_MODE_WB
260 * 01 1 WT : _PAGE_CACHE_MODE_WT
261 * 10 2 UC-: _PAGE_CACHE_MODE_UC_MINUS
262 * 11 3 UC : _PAGE_CACHE_MODE_UC
263 *
264 * NOTE: When WC or WP is used, it is redirected to UC- per
265 * the default setup in __cachemode2pte_tbl[].
266 */
267 pat = PAT(0, WB) | PAT(1, WT) | PAT(2, UC_MINUS) | PAT(3, UC) |
268 PAT(4, WB) | PAT(5, WT) | PAT(6, UC_MINUS) | PAT(7, UC);
269
270 } else if ((c->x86_vendor == X86_VENDOR_INTEL) &&
271 (((c->x86 == 0x6) && (c->x86_model <= 0xd)) ||
272 ((c->x86 == 0xf) && (c->x86_model <= 0x6)))) {
273 /*
274 * PAT support with the lower four entries. Intel Pentium 2,
275 * 3, M, and 4 are affected by PAT errata, which makes the
276 * upper four entries unusable. To be on the safe side, we don't
277 * use those.
278 *
279 * PTE encoding:
280 * PAT
281 * |PCD
282 * ||PWT PAT
283 * ||| slot
284 * 000 0 WB : _PAGE_CACHE_MODE_WB
285 * 001 1 WC : _PAGE_CACHE_MODE_WC
286 * 010 2 UC-: _PAGE_CACHE_MODE_UC_MINUS
287 * 011 3 UC : _PAGE_CACHE_MODE_UC
288 * PAT bit unused
289 *
290 * NOTE: When WT or WP is used, it is redirected to UC- per
291 * the default setup in __cachemode2pte_tbl[].
292 */
293 pat = PAT(0, WB) | PAT(1, WC) | PAT(2, UC_MINUS) | PAT(3, UC) |
294 PAT(4, WB) | PAT(5, WC) | PAT(6, UC_MINUS) | PAT(7, UC);
295 } else {
296 /*
297 * Full PAT support. We put WT in slot 7 to improve
298 * robustness in the presence of errata that might cause
299 * the high PAT bit to be ignored. This way, a buggy slot 7
300 * access will hit slot 3, and slot 3 is UC, so at worst
301 * we lose performance without causing a correctness issue.
302 * Pentium 4 erratum N46 is an example for such an erratum,
303 * although we try not to use PAT at all on affected CPUs.
304 *
305 * PTE encoding:
306 * PAT
307 * |PCD
308 * ||PWT PAT
309 * ||| slot
310 * 000 0 WB : _PAGE_CACHE_MODE_WB
311 * 001 1 WC : _PAGE_CACHE_MODE_WC
312 * 010 2 UC-: _PAGE_CACHE_MODE_UC_MINUS
313 * 011 3 UC : _PAGE_CACHE_MODE_UC
314 * 100 4 WB : Reserved
315 * 101 5 WC : Reserved
316 * 110 6 UC-: Reserved
317 * 111 7 WT : _PAGE_CACHE_MODE_WT
318 *
319 * The reserved slots are unused, but mapped to their
320 * corresponding types in the presence of PAT errata.
321 */
322 pat = PAT(0, WB) | PAT(1, WC) | PAT(2, UC_MINUS) | PAT(3, UC) |
323 PAT(4, WB) | PAT(5, WC) | PAT(6, UC_MINUS) | PAT(7, WT);
324 }
325
326 if (!boot_cpu_done) {
327 pat_bsp_init(pat);
328 boot_cpu_done = true;
329 } else {
330 pat_ap_init(pat);
331 }
332}
333
334#undef PAT
335
336static DEFINE_SPINLOCK(memtype_lock); /* protects memtype accesses */
337
338/*
339 * Does intersection of PAT memory type and MTRR memory type and returns
340 * the resulting memory type as PAT understands it.
341 * (Type in pat and mtrr will not have same value)
342 * The intersection is based on "Effective Memory Type" tables in IA-32
343 * SDM vol 3a
344 */
345static unsigned long pat_x_mtrr_type(u64 start, u64 end,
346 enum page_cache_mode req_type)
347{
348 /*
349 * Look for MTRR hint to get the effective type in case where PAT
350 * request is for WB.
351 */
352 if (req_type == _PAGE_CACHE_MODE_WB) {
353 u8 mtrr_type, uniform;
354
355 mtrr_type = mtrr_type_lookup(start, end, &uniform);
356 if (mtrr_type != MTRR_TYPE_WRBACK)
357 return _PAGE_CACHE_MODE_UC_MINUS;
358
359 return _PAGE_CACHE_MODE_WB;
360 }
361
362 return req_type;
363}
364
365struct pagerange_state {
366 unsigned long cur_pfn;
367 int ram;
368 int not_ram;
369};
370
371static int
372pagerange_is_ram_callback(unsigned long initial_pfn, unsigned long total_nr_pages, void *arg)
373{
374 struct pagerange_state *state = arg;
375
376 state->not_ram |= initial_pfn > state->cur_pfn;
377 state->ram |= total_nr_pages > 0;
378 state->cur_pfn = initial_pfn + total_nr_pages;
379
380 return state->ram && state->not_ram;
381}
382
383static int pat_pagerange_is_ram(resource_size_t start, resource_size_t end)
384{
385 int ret = 0;
386 unsigned long start_pfn = start >> PAGE_SHIFT;
387 unsigned long end_pfn = (end + PAGE_SIZE - 1) >> PAGE_SHIFT;
388 struct pagerange_state state = {start_pfn, 0, 0};
389
390 /*
391 * For legacy reasons, physical address range in the legacy ISA
392 * region is tracked as non-RAM. This will allow users of
393 * /dev/mem to map portions of legacy ISA region, even when
394 * some of those portions are listed(or not even listed) with
395 * different e820 types(RAM/reserved/..)
396 */
397 if (start_pfn < ISA_END_ADDRESS >> PAGE_SHIFT)
398 start_pfn = ISA_END_ADDRESS >> PAGE_SHIFT;
399
400 if (start_pfn < end_pfn) {
401 ret = walk_system_ram_range(start_pfn, end_pfn - start_pfn,
402 &state, pagerange_is_ram_callback);
403 }
404
405 return (ret > 0) ? -1 : (state.ram ? 1 : 0);
406}
407
408/*
409 * For RAM pages, we use page flags to mark the pages with appropriate type.
410 * The page flags are limited to four types, WB (default), WC, WT and UC-.
411 * WP request fails with -EINVAL, and UC gets redirected to UC-. Setting
412 * a new memory type is only allowed for a page mapped with the default WB
413 * type.
414 *
415 * Here we do two passes:
416 * - Find the memtype of all the pages in the range, look for any conflicts.
417 * - In case of no conflicts, set the new memtype for pages in the range.
418 */
419static int reserve_ram_pages_type(u64 start, u64 end,
420 enum page_cache_mode req_type,
421 enum page_cache_mode *new_type)
422{
423 struct page *page;
424 u64 pfn;
425
426 if (req_type == _PAGE_CACHE_MODE_WP) {
427 if (new_type)
428 *new_type = _PAGE_CACHE_MODE_UC_MINUS;
429 return -EINVAL;
430 }
431
432 if (req_type == _PAGE_CACHE_MODE_UC) {
433 /* We do not support strong UC */
434 WARN_ON_ONCE(1);
435 req_type = _PAGE_CACHE_MODE_UC_MINUS;
436 }
437
438 for (pfn = (start >> PAGE_SHIFT); pfn < (end >> PAGE_SHIFT); ++pfn) {
439 enum page_cache_mode type;
440
441 page = pfn_to_page(pfn);
442 type = get_page_memtype(page);
443 if (type != _PAGE_CACHE_MODE_WB) {
444 pr_info("x86/PAT: reserve_ram_pages_type failed [mem %#010Lx-%#010Lx], track 0x%x, req 0x%x\n",
445 start, end - 1, type, req_type);
446 if (new_type)
447 *new_type = type;
448
449 return -EBUSY;
450 }
451 }
452
453 if (new_type)
454 *new_type = req_type;
455
456 for (pfn = (start >> PAGE_SHIFT); pfn < (end >> PAGE_SHIFT); ++pfn) {
457 page = pfn_to_page(pfn);
458 set_page_memtype(page, req_type);
459 }
460 return 0;
461}
462
463static int free_ram_pages_type(u64 start, u64 end)
464{
465 struct page *page;
466 u64 pfn;
467
468 for (pfn = (start >> PAGE_SHIFT); pfn < (end >> PAGE_SHIFT); ++pfn) {
469 page = pfn_to_page(pfn);
470 set_page_memtype(page, _PAGE_CACHE_MODE_WB);
471 }
472 return 0;
473}
474
475/*
476 * req_type typically has one of the:
477 * - _PAGE_CACHE_MODE_WB
478 * - _PAGE_CACHE_MODE_WC
479 * - _PAGE_CACHE_MODE_UC_MINUS
480 * - _PAGE_CACHE_MODE_UC
481 * - _PAGE_CACHE_MODE_WT
482 *
483 * If new_type is NULL, function will return an error if it cannot reserve the
484 * region with req_type. If new_type is non-NULL, function will return
485 * available type in new_type in case of no error. In case of any error
486 * it will return a negative return value.
487 */
488int reserve_memtype(u64 start, u64 end, enum page_cache_mode req_type,
489 enum page_cache_mode *new_type)
490{
491 struct memtype *new;
492 enum page_cache_mode actual_type;
493 int is_range_ram;
494 int err = 0;
495
496 BUG_ON(start >= end); /* end is exclusive */
497
498 if (!pat_enabled()) {
499 /* This is identical to page table setting without PAT */
500 if (new_type)
501 *new_type = req_type;
502 return 0;
503 }
504
505 /* Low ISA region is always mapped WB in page table. No need to track */
506 if (x86_platform.is_untracked_pat_range(start, end)) {
507 if (new_type)
508 *new_type = _PAGE_CACHE_MODE_WB;
509 return 0;
510 }
511
512 /*
513 * Call mtrr_lookup to get the type hint. This is an
514 * optimization for /dev/mem mmap'ers into WB memory (BIOS
515 * tools and ACPI tools). Use WB request for WB memory and use
516 * UC_MINUS otherwise.
517 */
518 actual_type = pat_x_mtrr_type(start, end, req_type);
519
520 if (new_type)
521 *new_type = actual_type;
522
523 is_range_ram = pat_pagerange_is_ram(start, end);
524 if (is_range_ram == 1) {
525
526 err = reserve_ram_pages_type(start, end, req_type, new_type);
527
528 return err;
529 } else if (is_range_ram < 0) {
530 return -EINVAL;
531 }
532
533 new = kzalloc(sizeof(struct memtype), GFP_KERNEL);
534 if (!new)
535 return -ENOMEM;
536
537 new->start = start;
538 new->end = end;
539 new->type = actual_type;
540
541 spin_lock(&memtype_lock);
542
543 err = rbt_memtype_check_insert(new, new_type);
544 if (err) {
545 pr_info("x86/PAT: reserve_memtype failed [mem %#010Lx-%#010Lx], track %s, req %s\n",
546 start, end - 1,
547 cattr_name(new->type), cattr_name(req_type));
548 kfree(new);
549 spin_unlock(&memtype_lock);
550
551 return err;
552 }
553
554 spin_unlock(&memtype_lock);
555
556 dprintk("reserve_memtype added [mem %#010Lx-%#010Lx], track %s, req %s, ret %s\n",
557 start, end - 1, cattr_name(new->type), cattr_name(req_type),
558 new_type ? cattr_name(*new_type) : "-");
559
560 return err;
561}
562
563int free_memtype(u64 start, u64 end)
564{
565 int err = -EINVAL;
566 int is_range_ram;
567 struct memtype *entry;
568
569 if (!pat_enabled())
570 return 0;
571
572 /* Low ISA region is always mapped WB. No need to track */
573 if (x86_platform.is_untracked_pat_range(start, end))
574 return 0;
575
576 is_range_ram = pat_pagerange_is_ram(start, end);
577 if (is_range_ram == 1) {
578
579 err = free_ram_pages_type(start, end);
580
581 return err;
582 } else if (is_range_ram < 0) {
583 return -EINVAL;
584 }
585
586 spin_lock(&memtype_lock);
587 entry = rbt_memtype_erase(start, end);
588 spin_unlock(&memtype_lock);
589
590 if (IS_ERR(entry)) {
591 pr_info("x86/PAT: %s:%d freeing invalid memtype [mem %#010Lx-%#010Lx]\n",
592 current->comm, current->pid, start, end - 1);
593 return -EINVAL;
594 }
595
596 kfree(entry);
597
598 dprintk("free_memtype request [mem %#010Lx-%#010Lx]\n", start, end - 1);
599
600 return 0;
601}
602
603
604/**
605 * lookup_memtype - Looksup the memory type for a physical address
606 * @paddr: physical address of which memory type needs to be looked up
607 *
608 * Only to be called when PAT is enabled
609 *
610 * Returns _PAGE_CACHE_MODE_WB, _PAGE_CACHE_MODE_WC, _PAGE_CACHE_MODE_UC_MINUS
611 * or _PAGE_CACHE_MODE_WT.
612 */
613static enum page_cache_mode lookup_memtype(u64 paddr)
614{
615 enum page_cache_mode rettype = _PAGE_CACHE_MODE_WB;
616 struct memtype *entry;
617
618 if (x86_platform.is_untracked_pat_range(paddr, paddr + PAGE_SIZE))
619 return rettype;
620
621 if (pat_pagerange_is_ram(paddr, paddr + PAGE_SIZE)) {
622 struct page *page;
623
624 page = pfn_to_page(paddr >> PAGE_SHIFT);
625 return get_page_memtype(page);
626 }
627
628 spin_lock(&memtype_lock);
629
630 entry = rbt_memtype_lookup(paddr);
631 if (entry != NULL)
632 rettype = entry->type;
633 else
634 rettype = _PAGE_CACHE_MODE_UC_MINUS;
635
636 spin_unlock(&memtype_lock);
637 return rettype;
638}
639
640/**
641 * io_reserve_memtype - Request a memory type mapping for a region of memory
642 * @start: start (physical address) of the region
643 * @end: end (physical address) of the region
644 * @type: A pointer to memtype, with requested type. On success, requested
645 * or any other compatible type that was available for the region is returned
646 *
647 * On success, returns 0
648 * On failure, returns non-zero
649 */
650int io_reserve_memtype(resource_size_t start, resource_size_t end,
651 enum page_cache_mode *type)
652{
653 resource_size_t size = end - start;
654 enum page_cache_mode req_type = *type;
655 enum page_cache_mode new_type;
656 int ret;
657
658 WARN_ON_ONCE(iomem_map_sanity_check(start, size));
659
660 ret = reserve_memtype(start, end, req_type, &new_type);
661 if (ret)
662 goto out_err;
663
664 if (!is_new_memtype_allowed(start, size, req_type, new_type))
665 goto out_free;
666
667 if (kernel_map_sync_memtype(start, size, new_type) < 0)
668 goto out_free;
669
670 *type = new_type;
671 return 0;
672
673out_free:
674 free_memtype(start, end);
675 ret = -EBUSY;
676out_err:
677 return ret;
678}
679
680/**
681 * io_free_memtype - Release a memory type mapping for a region of memory
682 * @start: start (physical address) of the region
683 * @end: end (physical address) of the region
684 */
685void io_free_memtype(resource_size_t start, resource_size_t end)
686{
687 free_memtype(start, end);
688}
689
690pgprot_t phys_mem_access_prot(struct file *file, unsigned long pfn,
691 unsigned long size, pgprot_t vma_prot)
692{
693 return vma_prot;
694}
695
696#ifdef CONFIG_STRICT_DEVMEM
697/* This check is done in drivers/char/mem.c in case of STRICT_DEVMEM */
698static inline int range_is_allowed(unsigned long pfn, unsigned long size)
699{
700 return 1;
701}
702#else
703/* This check is needed to avoid cache aliasing when PAT is enabled */
704static inline int range_is_allowed(unsigned long pfn, unsigned long size)
705{
706 u64 from = ((u64)pfn) << PAGE_SHIFT;
707 u64 to = from + size;
708 u64 cursor = from;
709
710 if (!pat_enabled())
711 return 1;
712
713 while (cursor < to) {
714 if (!devmem_is_allowed(pfn)) {
715 pr_info("x86/PAT: Program %s tried to access /dev/mem between [mem %#010Lx-%#010Lx], PAT prevents it\n",
716 current->comm, from, to - 1);
717 return 0;
718 }
719 cursor += PAGE_SIZE;
720 pfn++;
721 }
722 return 1;
723}
724#endif /* CONFIG_STRICT_DEVMEM */
725
726int phys_mem_access_prot_allowed(struct file *file, unsigned long pfn,
727 unsigned long size, pgprot_t *vma_prot)
728{
729 enum page_cache_mode pcm = _PAGE_CACHE_MODE_WB;
730
731 if (!range_is_allowed(pfn, size))
732 return 0;
733
734 if (file->f_flags & O_DSYNC)
735 pcm = _PAGE_CACHE_MODE_UC_MINUS;
736
737#ifdef CONFIG_X86_32
738 /*
739 * On the PPro and successors, the MTRRs are used to set
740 * memory types for physical addresses outside main memory,
741 * so blindly setting UC or PWT on those pages is wrong.
742 * For Pentiums and earlier, the surround logic should disable
743 * caching for the high addresses through the KEN pin, but
744 * we maintain the tradition of paranoia in this code.
745 */
746 if (!pat_enabled() &&
747 !(boot_cpu_has(X86_FEATURE_MTRR) ||
748 boot_cpu_has(X86_FEATURE_K6_MTRR) ||
749 boot_cpu_has(X86_FEATURE_CYRIX_ARR) ||
750 boot_cpu_has(X86_FEATURE_CENTAUR_MCR)) &&
751 (pfn << PAGE_SHIFT) >= __pa(high_memory)) {
752 pcm = _PAGE_CACHE_MODE_UC;
753 }
754#endif
755
756 *vma_prot = __pgprot((pgprot_val(*vma_prot) & ~_PAGE_CACHE_MASK) |
757 cachemode2protval(pcm));
758 return 1;
759}
760
761/*
762 * Change the memory type for the physial address range in kernel identity
763 * mapping space if that range is a part of identity map.
764 */
765int kernel_map_sync_memtype(u64 base, unsigned long size,
766 enum page_cache_mode pcm)
767{
768 unsigned long id_sz;
769
770 if (base > __pa(high_memory-1))
771 return 0;
772
773 /*
774 * some areas in the middle of the kernel identity range
775 * are not mapped, like the PCI space.
776 */
777 if (!page_is_ram(base >> PAGE_SHIFT))
778 return 0;
779
780 id_sz = (__pa(high_memory-1) <= base + size) ?
781 __pa(high_memory) - base :
782 size;
783
784 if (ioremap_change_attr((unsigned long)__va(base), id_sz, pcm) < 0) {
785 pr_info("x86/PAT: %s:%d ioremap_change_attr failed %s for [mem %#010Lx-%#010Lx]\n",
786 current->comm, current->pid,
787 cattr_name(pcm),
788 base, (unsigned long long)(base + size-1));
789 return -EINVAL;
790 }
791 return 0;
792}
793
794/*
795 * Internal interface to reserve a range of physical memory with prot.
796 * Reserved non RAM regions only and after successful reserve_memtype,
797 * this func also keeps identity mapping (if any) in sync with this new prot.
798 */
799static int reserve_pfn_range(u64 paddr, unsigned long size, pgprot_t *vma_prot,
800 int strict_prot)
801{
802 int is_ram = 0;
803 int ret;
804 enum page_cache_mode want_pcm = pgprot2cachemode(*vma_prot);
805 enum page_cache_mode pcm = want_pcm;
806
807 is_ram = pat_pagerange_is_ram(paddr, paddr + size);
808
809 /*
810 * reserve_pfn_range() for RAM pages. We do not refcount to keep
811 * track of number of mappings of RAM pages. We can assert that
812 * the type requested matches the type of first page in the range.
813 */
814 if (is_ram) {
815 if (!pat_enabled())
816 return 0;
817
818 pcm = lookup_memtype(paddr);
819 if (want_pcm != pcm) {
820 pr_warn("x86/PAT: %s:%d map pfn RAM range req %s for [mem %#010Lx-%#010Lx], got %s\n",
821 current->comm, current->pid,
822 cattr_name(want_pcm),
823 (unsigned long long)paddr,
824 (unsigned long long)(paddr + size - 1),
825 cattr_name(pcm));
826 *vma_prot = __pgprot((pgprot_val(*vma_prot) &
827 (~_PAGE_CACHE_MASK)) |
828 cachemode2protval(pcm));
829 }
830 return 0;
831 }
832
833 ret = reserve_memtype(paddr, paddr + size, want_pcm, &pcm);
834 if (ret)
835 return ret;
836
837 if (pcm != want_pcm) {
838 if (strict_prot ||
839 !is_new_memtype_allowed(paddr, size, want_pcm, pcm)) {
840 free_memtype(paddr, paddr + size);
841 pr_err("x86/PAT: %s:%d map pfn expected mapping type %s for [mem %#010Lx-%#010Lx], got %s\n",
842 current->comm, current->pid,
843 cattr_name(want_pcm),
844 (unsigned long long)paddr,
845 (unsigned long long)(paddr + size - 1),
846 cattr_name(pcm));
847 return -EINVAL;
848 }
849 /*
850 * We allow returning different type than the one requested in
851 * non strict case.
852 */
853 *vma_prot = __pgprot((pgprot_val(*vma_prot) &
854 (~_PAGE_CACHE_MASK)) |
855 cachemode2protval(pcm));
856 }
857
858 if (kernel_map_sync_memtype(paddr, size, pcm) < 0) {
859 free_memtype(paddr, paddr + size);
860 return -EINVAL;
861 }
862 return 0;
863}
864
865/*
866 * Internal interface to free a range of physical memory.
867 * Frees non RAM regions only.
868 */
869static void free_pfn_range(u64 paddr, unsigned long size)
870{
871 int is_ram;
872
873 is_ram = pat_pagerange_is_ram(paddr, paddr + size);
874 if (is_ram == 0)
875 free_memtype(paddr, paddr + size);
876}
877
878/*
879 * track_pfn_copy is called when vma that is covering the pfnmap gets
880 * copied through copy_page_range().
881 *
882 * If the vma has a linear pfn mapping for the entire range, we get the prot
883 * from pte and reserve the entire vma range with single reserve_pfn_range call.
884 */
885int track_pfn_copy(struct vm_area_struct *vma)
886{
887 resource_size_t paddr;
888 unsigned long prot;
889 unsigned long vma_size = vma->vm_end - vma->vm_start;
890 pgprot_t pgprot;
891
892 if (vma->vm_flags & VM_PAT) {
893 /*
894 * reserve the whole chunk covered by vma. We need the
895 * starting address and protection from pte.
896 */
897 if (follow_phys(vma, vma->vm_start, 0, &prot, &paddr)) {
898 WARN_ON_ONCE(1);
899 return -EINVAL;
900 }
901 pgprot = __pgprot(prot);
902 return reserve_pfn_range(paddr, vma_size, &pgprot, 1);
903 }
904
905 return 0;
906}
907
908/*
909 * prot is passed in as a parameter for the new mapping. If the vma has a
910 * linear pfn mapping for the entire range reserve the entire vma range with
911 * single reserve_pfn_range call.
912 */
913int track_pfn_remap(struct vm_area_struct *vma, pgprot_t *prot,
914 unsigned long pfn, unsigned long addr, unsigned long size)
915{
916 resource_size_t paddr = (resource_size_t)pfn << PAGE_SHIFT;
917 enum page_cache_mode pcm;
918
919 /* reserve the whole chunk starting from paddr */
920 if (addr == vma->vm_start && size == (vma->vm_end - vma->vm_start)) {
921 int ret;
922
923 ret = reserve_pfn_range(paddr, size, prot, 0);
924 if (!ret)
925 vma->vm_flags |= VM_PAT;
926 return ret;
927 }
928
929 if (!pat_enabled())
930 return 0;
931
932 /*
933 * For anything smaller than the vma size we set prot based on the
934 * lookup.
935 */
936 pcm = lookup_memtype(paddr);
937
938 /* Check memtype for the remaining pages */
939 while (size > PAGE_SIZE) {
940 size -= PAGE_SIZE;
941 paddr += PAGE_SIZE;
942 if (pcm != lookup_memtype(paddr))
943 return -EINVAL;
944 }
945
946 *prot = __pgprot((pgprot_val(*prot) & (~_PAGE_CACHE_MASK)) |
947 cachemode2protval(pcm));
948
949 return 0;
950}
951
952int track_pfn_insert(struct vm_area_struct *vma, pgprot_t *prot,
953 pfn_t pfn)
954{
955 enum page_cache_mode pcm;
956
957 if (!pat_enabled())
958 return 0;
959
960 /* Set prot based on lookup */
961 pcm = lookup_memtype(pfn_t_to_phys(pfn));
962 *prot = __pgprot((pgprot_val(*prot) & (~_PAGE_CACHE_MASK)) |
963 cachemode2protval(pcm));
964
965 return 0;
966}
967
968/*
969 * untrack_pfn is called while unmapping a pfnmap for a region.
970 * untrack can be called for a specific region indicated by pfn and size or
971 * can be for the entire vma (in which case pfn, size are zero).
972 */
973void untrack_pfn(struct vm_area_struct *vma, unsigned long pfn,
974 unsigned long size)
975{
976 resource_size_t paddr;
977 unsigned long prot;
978
979 if (!(vma->vm_flags & VM_PAT))
980 return;
981
982 /* free the chunk starting from pfn or the whole chunk */
983 paddr = (resource_size_t)pfn << PAGE_SHIFT;
984 if (!paddr && !size) {
985 if (follow_phys(vma, vma->vm_start, 0, &prot, &paddr)) {
986 WARN_ON_ONCE(1);
987 return;
988 }
989
990 size = vma->vm_end - vma->vm_start;
991 }
992 free_pfn_range(paddr, size);
993 vma->vm_flags &= ~VM_PAT;
994}
995
996/*
997 * untrack_pfn_moved is called, while mremapping a pfnmap for a new region,
998 * with the old vma after its pfnmap page table has been removed. The new
999 * vma has a new pfnmap to the same pfn & cache type with VM_PAT set.
1000 */
1001void untrack_pfn_moved(struct vm_area_struct *vma)
1002{
1003 vma->vm_flags &= ~VM_PAT;
1004}
1005
1006pgprot_t pgprot_writecombine(pgprot_t prot)
1007{
1008 return __pgprot(pgprot_val(prot) |
1009 cachemode2protval(_PAGE_CACHE_MODE_WC));
1010}
1011EXPORT_SYMBOL_GPL(pgprot_writecombine);
1012
1013pgprot_t pgprot_writethrough(pgprot_t prot)
1014{
1015 return __pgprot(pgprot_val(prot) |
1016 cachemode2protval(_PAGE_CACHE_MODE_WT));
1017}
1018EXPORT_SYMBOL_GPL(pgprot_writethrough);
1019
1020#if defined(CONFIG_DEBUG_FS) && defined(CONFIG_X86_PAT)
1021
1022static struct memtype *memtype_get_idx(loff_t pos)
1023{
1024 struct memtype *print_entry;
1025 int ret;
1026
1027 print_entry = kzalloc(sizeof(struct memtype), GFP_KERNEL);
1028 if (!print_entry)
1029 return NULL;
1030
1031 spin_lock(&memtype_lock);
1032 ret = rbt_memtype_copy_nth_element(print_entry, pos);
1033 spin_unlock(&memtype_lock);
1034
1035 if (!ret) {
1036 return print_entry;
1037 } else {
1038 kfree(print_entry);
1039 return NULL;
1040 }
1041}
1042
1043static void *memtype_seq_start(struct seq_file *seq, loff_t *pos)
1044{
1045 if (*pos == 0) {
1046 ++*pos;
1047 seq_puts(seq, "PAT memtype list:\n");
1048 }
1049
1050 return memtype_get_idx(*pos);
1051}
1052
1053static void *memtype_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1054{
1055 ++*pos;
1056 return memtype_get_idx(*pos);
1057}
1058
1059static void memtype_seq_stop(struct seq_file *seq, void *v)
1060{
1061}
1062
1063static int memtype_seq_show(struct seq_file *seq, void *v)
1064{
1065 struct memtype *print_entry = (struct memtype *)v;
1066
1067 seq_printf(seq, "%s @ 0x%Lx-0x%Lx\n", cattr_name(print_entry->type),
1068 print_entry->start, print_entry->end);
1069 kfree(print_entry);
1070
1071 return 0;
1072}
1073
1074static const struct seq_operations memtype_seq_ops = {
1075 .start = memtype_seq_start,
1076 .next = memtype_seq_next,
1077 .stop = memtype_seq_stop,
1078 .show = memtype_seq_show,
1079};
1080
1081static int memtype_seq_open(struct inode *inode, struct file *file)
1082{
1083 return seq_open(file, &memtype_seq_ops);
1084}
1085
1086static const struct file_operations memtype_fops = {
1087 .open = memtype_seq_open,
1088 .read = seq_read,
1089 .llseek = seq_lseek,
1090 .release = seq_release,
1091};
1092
1093static int __init pat_memtype_list_init(void)
1094{
1095 if (pat_enabled()) {
1096 debugfs_create_file("pat_memtype_list", S_IRUSR,
1097 arch_debugfs_dir, NULL, &memtype_fops);
1098 }
1099 return 0;
1100}
1101
1102late_initcall(pat_memtype_list_init);
1103
1104#endif /* CONFIG_DEBUG_FS && CONFIG_X86_PAT */