mm: remove __GFP_NO_KSWAPD
[linux-2.6-block.git] / arch / x86 / mm / pat.c
CommitLineData
2e5d9c85 1/*
2 * Handle caching attributes in page tables (PAT)
3 *
4 * Authors: Venkatesh Pallipadi <venkatesh.pallipadi@intel.com>
5 * Suresh B Siddha <suresh.b.siddha@intel.com>
6 *
7 * Loosely based on earlier PAT patchset from Eric Biederman and Andi Kleen.
8 */
9
ad2cde16
IM
10#include <linux/seq_file.h>
11#include <linux/bootmem.h>
12#include <linux/debugfs.h>
2e5d9c85 13#include <linux/kernel.h>
92b9af9e 14#include <linux/module.h>
5a0e3ad6 15#include <linux/slab.h>
ad2cde16 16#include <linux/mm.h>
2e5d9c85 17#include <linux/fs.h>
335ef896 18#include <linux/rbtree.h>
2e5d9c85 19
ad2cde16 20#include <asm/cacheflush.h>
2e5d9c85 21#include <asm/processor.h>
ad2cde16 22#include <asm/tlbflush.h>
fd12a0d6 23#include <asm/x86_init.h>
2e5d9c85 24#include <asm/pgtable.h>
2e5d9c85 25#include <asm/fcntl.h>
ad2cde16 26#include <asm/e820.h>
2e5d9c85 27#include <asm/mtrr.h>
ad2cde16
IM
28#include <asm/page.h>
29#include <asm/msr.h>
30#include <asm/pat.h>
e7f260a2 31#include <asm/io.h>
2e5d9c85 32
be5a0c12 33#include "pat_internal.h"
34
8d4a4300 35#ifdef CONFIG_X86_PAT
499f8f84 36int __read_mostly pat_enabled = 1;
2e5d9c85 37
1ee4bd92 38static inline void pat_disable(const char *reason)
2e5d9c85 39{
499f8f84 40 pat_enabled = 0;
8d4a4300 41 printk(KERN_INFO "%s\n", reason);
2e5d9c85 42}
2e5d9c85 43
be524fb9 44static int __init nopat(char *str)
2e5d9c85 45{
8d4a4300 46 pat_disable("PAT support disabled.");
2e5d9c85 47 return 0;
48}
8d4a4300 49early_param("nopat", nopat);
75a04811
PA
50#else
51static inline void pat_disable(const char *reason)
52{
53 (void)reason;
54}
8d4a4300
TG
55#endif
56
77b52b4c 57
be5a0c12 58int pat_debug_enable;
ad2cde16 59
77b52b4c
VP
60static int __init pat_debug_setup(char *str)
61{
be5a0c12 62 pat_debug_enable = 1;
77b52b4c
VP
63 return 0;
64}
65__setup("debugpat", pat_debug_setup);
66
8d4a4300 67static u64 __read_mostly boot_pat_state;
2e5d9c85 68
69enum {
70 PAT_UC = 0, /* uncached */
71 PAT_WC = 1, /* Write combining */
72 PAT_WT = 4, /* Write Through */
73 PAT_WP = 5, /* Write Protected */
74 PAT_WB = 6, /* Write Back (default) */
75 PAT_UC_MINUS = 7, /* UC, but can be overriden by MTRR */
76};
77
cd7a4e93 78#define PAT(x, y) ((u64)PAT_ ## y << ((x)*8))
2e5d9c85 79
80void pat_init(void)
81{
82 u64 pat;
e23a8b6a 83 bool boot_cpu = !boot_pat_state;
2e5d9c85 84
499f8f84 85 if (!pat_enabled)
2e5d9c85 86 return;
87
75a04811
PA
88 if (!cpu_has_pat) {
89 if (!boot_pat_state) {
90 pat_disable("PAT not supported by CPU.");
91 return;
92 } else {
93 /*
94 * If this happens we are on a secondary CPU, but
95 * switched to PAT on the boot CPU. We have no way to
96 * undo PAT.
97 */
98 printk(KERN_ERR "PAT enabled, "
99 "but not supported by secondary CPU\n");
100 BUG();
101 }
8d4a4300 102 }
2e5d9c85 103
104 /* Set PWT to Write-Combining. All other bits stay the same */
105 /*
106 * PTE encoding used in Linux:
107 * PAT
108 * |PCD
109 * ||PWT
110 * |||
111 * 000 WB _PAGE_CACHE_WB
112 * 001 WC _PAGE_CACHE_WC
113 * 010 UC- _PAGE_CACHE_UC_MINUS
114 * 011 UC _PAGE_CACHE_UC
115 * PAT bit unused
116 */
cd7a4e93
AH
117 pat = PAT(0, WB) | PAT(1, WC) | PAT(2, UC_MINUS) | PAT(3, UC) |
118 PAT(4, WB) | PAT(5, WC) | PAT(6, UC_MINUS) | PAT(7, UC);
2e5d9c85 119
120 /* Boot CPU check */
8d4a4300 121 if (!boot_pat_state)
2e5d9c85 122 rdmsrl(MSR_IA32_CR_PAT, boot_pat_state);
2e5d9c85 123
124 wrmsrl(MSR_IA32_CR_PAT, pat);
e23a8b6a
RD
125
126 if (boot_cpu)
127 printk(KERN_INFO "x86 PAT enabled: cpu %d, old 0x%Lx, new 0x%Lx\n",
128 smp_processor_id(), boot_pat_state, pat);
2e5d9c85 129}
130
131#undef PAT
132
9e41a49a 133static DEFINE_SPINLOCK(memtype_lock); /* protects memtype accesses */
335ef896 134
2e5d9c85 135/*
136 * Does intersection of PAT memory type and MTRR memory type and returns
137 * the resulting memory type as PAT understands it.
138 * (Type in pat and mtrr will not have same value)
139 * The intersection is based on "Effective Memory Type" tables in IA-32
140 * SDM vol 3a
141 */
6cf514fc 142static unsigned long pat_x_mtrr_type(u64 start, u64 end, unsigned long req_type)
2e5d9c85 143{
c26421d0
VP
144 /*
145 * Look for MTRR hint to get the effective type in case where PAT
146 * request is for WB.
147 */
dd0c7c49
AH
148 if (req_type == _PAGE_CACHE_WB) {
149 u8 mtrr_type;
150
151 mtrr_type = mtrr_type_lookup(start, end);
b6ff32d9
SS
152 if (mtrr_type != MTRR_TYPE_WRBACK)
153 return _PAGE_CACHE_UC_MINUS;
154
155 return _PAGE_CACHE_WB;
dd0c7c49
AH
156 }
157
158 return req_type;
2e5d9c85 159}
160
fa83523f
JD
161struct pagerange_state {
162 unsigned long cur_pfn;
163 int ram;
164 int not_ram;
165};
166
167static int
168pagerange_is_ram_callback(unsigned long initial_pfn, unsigned long total_nr_pages, void *arg)
169{
170 struct pagerange_state *state = arg;
171
172 state->not_ram |= initial_pfn > state->cur_pfn;
173 state->ram |= total_nr_pages > 0;
174 state->cur_pfn = initial_pfn + total_nr_pages;
175
176 return state->ram && state->not_ram;
177}
178
3709c857 179static int pat_pagerange_is_ram(resource_size_t start, resource_size_t end)
be03d9e8 180{
fa83523f
JD
181 int ret = 0;
182 unsigned long start_pfn = start >> PAGE_SHIFT;
183 unsigned long end_pfn = (end + PAGE_SIZE - 1) >> PAGE_SHIFT;
184 struct pagerange_state state = {start_pfn, 0, 0};
185
186 /*
187 * For legacy reasons, physical address range in the legacy ISA
188 * region is tracked as non-RAM. This will allow users of
189 * /dev/mem to map portions of legacy ISA region, even when
190 * some of those portions are listed(or not even listed) with
191 * different e820 types(RAM/reserved/..)
192 */
193 if (start_pfn < ISA_END_ADDRESS >> PAGE_SHIFT)
194 start_pfn = ISA_END_ADDRESS >> PAGE_SHIFT;
195
196 if (start_pfn < end_pfn) {
197 ret = walk_system_ram_range(start_pfn, end_pfn - start_pfn,
198 &state, pagerange_is_ram_callback);
be03d9e8
SS
199 }
200
fa83523f 201 return (ret > 0) ? -1 : (state.ram ? 1 : 0);
be03d9e8
SS
202}
203
9542ada8 204/*
f5841740
VP
205 * For RAM pages, we use page flags to mark the pages with appropriate type.
206 * Here we do two pass:
207 * - Find the memtype of all the pages in the range, look for any conflicts
208 * - In case of no conflicts, set the new memtype for pages in the range
9542ada8
SS
209 */
210static int reserve_ram_pages_type(u64 start, u64 end, unsigned long req_type,
ad2cde16 211 unsigned long *new_type)
9542ada8
SS
212{
213 struct page *page;
f5841740
VP
214 u64 pfn;
215
216 if (req_type == _PAGE_CACHE_UC) {
217 /* We do not support strong UC */
218 WARN_ON_ONCE(1);
219 req_type = _PAGE_CACHE_UC_MINUS;
220 }
9542ada8
SS
221
222 for (pfn = (start >> PAGE_SHIFT); pfn < (end >> PAGE_SHIFT); ++pfn) {
f5841740 223 unsigned long type;
9542ada8 224
f5841740
VP
225 page = pfn_to_page(pfn);
226 type = get_page_memtype(page);
227 if (type != -1) {
365811d6
BH
228 printk(KERN_INFO "reserve_ram_pages_type failed [mem %#010Lx-%#010Lx], track 0x%lx, req 0x%lx\n",
229 start, end - 1, type, req_type);
f5841740
VP
230 if (new_type)
231 *new_type = type;
232
233 return -EBUSY;
234 }
9542ada8 235 }
9542ada8 236
f5841740
VP
237 if (new_type)
238 *new_type = req_type;
239
240 for (pfn = (start >> PAGE_SHIFT); pfn < (end >> PAGE_SHIFT); ++pfn) {
9542ada8 241 page = pfn_to_page(pfn);
f5841740 242 set_page_memtype(page, req_type);
9542ada8 243 }
f5841740 244 return 0;
9542ada8
SS
245}
246
247static int free_ram_pages_type(u64 start, u64 end)
248{
249 struct page *page;
f5841740 250 u64 pfn;
9542ada8
SS
251
252 for (pfn = (start >> PAGE_SHIFT); pfn < (end >> PAGE_SHIFT); ++pfn) {
253 page = pfn_to_page(pfn);
f5841740 254 set_page_memtype(page, -1);
9542ada8
SS
255 }
256 return 0;
9542ada8
SS
257}
258
e7f260a2 259/*
260 * req_type typically has one of the:
261 * - _PAGE_CACHE_WB
262 * - _PAGE_CACHE_WC
263 * - _PAGE_CACHE_UC_MINUS
264 * - _PAGE_CACHE_UC
265 *
ac97991e
AH
266 * If new_type is NULL, function will return an error if it cannot reserve the
267 * region with req_type. If new_type is non-NULL, function will return
268 * available type in new_type in case of no error. In case of any error
e7f260a2 269 * it will return a negative return value.
270 */
2e5d9c85 271int reserve_memtype(u64 start, u64 end, unsigned long req_type,
ad2cde16 272 unsigned long *new_type)
2e5d9c85 273{
be5a0c12 274 struct memtype *new;
2e5d9c85 275 unsigned long actual_type;
9542ada8 276 int is_range_ram;
ad2cde16 277 int err = 0;
2e5d9c85 278
ad2cde16 279 BUG_ON(start >= end); /* end is exclusive */
69e26be9 280
499f8f84 281 if (!pat_enabled) {
e7f260a2 282 /* This is identical to page table setting without PAT */
ac97991e 283 if (new_type) {
83ea05ea 284 if (req_type == _PAGE_CACHE_WC)
5fc51746 285 *new_type = _PAGE_CACHE_UC_MINUS;
ac97991e
AH
286 else
287 *new_type = req_type & _PAGE_CACHE_MASK;
e7f260a2 288 }
2e5d9c85 289 return 0;
290 }
291
292 /* Low ISA region is always mapped WB in page table. No need to track */
8a271389 293 if (x86_platform.is_untracked_pat_range(start, end)) {
ac97991e
AH
294 if (new_type)
295 *new_type = _PAGE_CACHE_WB;
2e5d9c85 296 return 0;
297 }
298
b6ff32d9
SS
299 /*
300 * Call mtrr_lookup to get the type hint. This is an
301 * optimization for /dev/mem mmap'ers into WB memory (BIOS
302 * tools and ACPI tools). Use WB request for WB memory and use
303 * UC_MINUS otherwise.
304 */
305 actual_type = pat_x_mtrr_type(start, end, req_type & _PAGE_CACHE_MASK);
2e5d9c85 306
95971342
SS
307 if (new_type)
308 *new_type = actual_type;
309
be03d9e8 310 is_range_ram = pat_pagerange_is_ram(start, end);
f5841740
VP
311 if (is_range_ram == 1) {
312
f5841740 313 err = reserve_ram_pages_type(start, end, req_type, new_type);
f5841740
VP
314
315 return err;
316 } else if (is_range_ram < 0) {
9542ada8 317 return -EINVAL;
f5841740 318 }
9542ada8 319
6a4f3b52 320 new = kzalloc(sizeof(struct memtype), GFP_KERNEL);
ac97991e 321 if (!new)
2e5d9c85 322 return -ENOMEM;
323
ad2cde16
IM
324 new->start = start;
325 new->end = end;
326 new->type = actual_type;
2e5d9c85 327
2e5d9c85 328 spin_lock(&memtype_lock);
329
9e41a49a 330 err = rbt_memtype_check_insert(new, new_type);
2e5d9c85 331 if (err) {
365811d6
BH
332 printk(KERN_INFO "reserve_memtype failed [mem %#010Lx-%#010Lx], track %s, req %s\n",
333 start, end - 1,
334 cattr_name(new->type), cattr_name(req_type));
ac97991e 335 kfree(new);
2e5d9c85 336 spin_unlock(&memtype_lock);
ad2cde16 337
2e5d9c85 338 return err;
339 }
340
2e5d9c85 341 spin_unlock(&memtype_lock);
3e9c83b3 342
365811d6
BH
343 dprintk("reserve_memtype added [mem %#010Lx-%#010Lx], track %s, req %s, ret %s\n",
344 start, end - 1, cattr_name(new->type), cattr_name(req_type),
3e9c83b3
AH
345 new_type ? cattr_name(*new_type) : "-");
346
2e5d9c85 347 return err;
348}
349
350int free_memtype(u64 start, u64 end)
351{
2e5d9c85 352 int err = -EINVAL;
9542ada8 353 int is_range_ram;
20413f27 354 struct memtype *entry;
2e5d9c85 355
69e26be9 356 if (!pat_enabled)
2e5d9c85 357 return 0;
2e5d9c85 358
359 /* Low ISA region is always mapped WB. No need to track */
8a271389 360 if (x86_platform.is_untracked_pat_range(start, end))
2e5d9c85 361 return 0;
2e5d9c85 362
be03d9e8 363 is_range_ram = pat_pagerange_is_ram(start, end);
f5841740
VP
364 if (is_range_ram == 1) {
365
f5841740 366 err = free_ram_pages_type(start, end);
f5841740
VP
367
368 return err;
369 } else if (is_range_ram < 0) {
9542ada8 370 return -EINVAL;
f5841740 371 }
9542ada8 372
2e5d9c85 373 spin_lock(&memtype_lock);
20413f27 374 entry = rbt_memtype_erase(start, end);
2e5d9c85 375 spin_unlock(&memtype_lock);
376
20413f27 377 if (!entry) {
365811d6
BH
378 printk(KERN_INFO "%s:%d freeing invalid memtype [mem %#010Lx-%#010Lx]\n",
379 current->comm, current->pid, start, end - 1);
20413f27 380 return -EINVAL;
2e5d9c85 381 }
6997ab49 382
20413f27
XF
383 kfree(entry);
384
365811d6 385 dprintk("free_memtype request [mem %#010Lx-%#010Lx]\n", start, end - 1);
ad2cde16 386
20413f27 387 return 0;
2e5d9c85 388}
389
f0970c13 390
637b86e7
VP
391/**
392 * lookup_memtype - Looksup the memory type for a physical address
393 * @paddr: physical address of which memory type needs to be looked up
394 *
395 * Only to be called when PAT is enabled
396 *
397 * Returns _PAGE_CACHE_WB, _PAGE_CACHE_WC, _PAGE_CACHE_UC_MINUS or
398 * _PAGE_CACHE_UC
399 */
400static unsigned long lookup_memtype(u64 paddr)
401{
402 int rettype = _PAGE_CACHE_WB;
403 struct memtype *entry;
404
8a271389 405 if (x86_platform.is_untracked_pat_range(paddr, paddr + PAGE_SIZE))
637b86e7
VP
406 return rettype;
407
408 if (pat_pagerange_is_ram(paddr, paddr + PAGE_SIZE)) {
409 struct page *page;
637b86e7
VP
410 page = pfn_to_page(paddr >> PAGE_SHIFT);
411 rettype = get_page_memtype(page);
637b86e7
VP
412 /*
413 * -1 from get_page_memtype() implies RAM page is in its
414 * default state and not reserved, and hence of type WB
415 */
416 if (rettype == -1)
417 rettype = _PAGE_CACHE_WB;
418
419 return rettype;
420 }
421
422 spin_lock(&memtype_lock);
423
9e41a49a 424 entry = rbt_memtype_lookup(paddr);
637b86e7
VP
425 if (entry != NULL)
426 rettype = entry->type;
427 else
428 rettype = _PAGE_CACHE_UC_MINUS;
429
430 spin_unlock(&memtype_lock);
431 return rettype;
432}
433
9fd126bc
VP
434/**
435 * io_reserve_memtype - Request a memory type mapping for a region of memory
436 * @start: start (physical address) of the region
437 * @end: end (physical address) of the region
438 * @type: A pointer to memtype, with requested type. On success, requested
439 * or any other compatible type that was available for the region is returned
440 *
441 * On success, returns 0
442 * On failure, returns non-zero
443 */
444int io_reserve_memtype(resource_size_t start, resource_size_t end,
445 unsigned long *type)
446{
b855192c 447 resource_size_t size = end - start;
9fd126bc
VP
448 unsigned long req_type = *type;
449 unsigned long new_type;
450 int ret;
451
b855192c 452 WARN_ON_ONCE(iomem_map_sanity_check(start, size));
9fd126bc
VP
453
454 ret = reserve_memtype(start, end, req_type, &new_type);
455 if (ret)
456 goto out_err;
457
b855192c 458 if (!is_new_memtype_allowed(start, size, req_type, new_type))
9fd126bc
VP
459 goto out_free;
460
b855192c 461 if (kernel_map_sync_memtype(start, size, new_type) < 0)
9fd126bc
VP
462 goto out_free;
463
464 *type = new_type;
465 return 0;
466
467out_free:
468 free_memtype(start, end);
469 ret = -EBUSY;
470out_err:
471 return ret;
472}
473
474/**
475 * io_free_memtype - Release a memory type mapping for a region of memory
476 * @start: start (physical address) of the region
477 * @end: end (physical address) of the region
478 */
479void io_free_memtype(resource_size_t start, resource_size_t end)
480{
481 free_memtype(start, end);
482}
483
f0970c13 484pgprot_t phys_mem_access_prot(struct file *file, unsigned long pfn,
485 unsigned long size, pgprot_t vma_prot)
486{
487 return vma_prot;
488}
489
d092633b
IM
490#ifdef CONFIG_STRICT_DEVMEM
491/* This check is done in drivers/char/mem.c in case of STRICT_DEVMEM*/
0124cecf
VP
492static inline int range_is_allowed(unsigned long pfn, unsigned long size)
493{
494 return 1;
495}
496#else
9e41bff2 497/* This check is needed to avoid cache aliasing when PAT is enabled */
0124cecf
VP
498static inline int range_is_allowed(unsigned long pfn, unsigned long size)
499{
500 u64 from = ((u64)pfn) << PAGE_SHIFT;
501 u64 to = from + size;
502 u64 cursor = from;
503
9e41bff2
RT
504 if (!pat_enabled)
505 return 1;
506
0124cecf
VP
507 while (cursor < to) {
508 if (!devmem_is_allowed(pfn)) {
365811d6
BH
509 printk(KERN_INFO "Program %s tried to access /dev/mem between [mem %#010Lx-%#010Lx]\n",
510 current->comm, from, to - 1);
0124cecf
VP
511 return 0;
512 }
513 cursor += PAGE_SIZE;
514 pfn++;
515 }
516 return 1;
517}
d092633b 518#endif /* CONFIG_STRICT_DEVMEM */
0124cecf 519
f0970c13 520int phys_mem_access_prot_allowed(struct file *file, unsigned long pfn,
521 unsigned long size, pgprot_t *vma_prot)
522{
0c3c8a18 523 unsigned long flags = _PAGE_CACHE_WB;
f0970c13 524
0124cecf
VP
525 if (!range_is_allowed(pfn, size))
526 return 0;
527
6b2f3d1f 528 if (file->f_flags & O_DSYNC)
28df82eb 529 flags = _PAGE_CACHE_UC_MINUS;
f0970c13 530
531#ifdef CONFIG_X86_32
532 /*
533 * On the PPro and successors, the MTRRs are used to set
534 * memory types for physical addresses outside main memory,
535 * so blindly setting UC or PWT on those pages is wrong.
536 * For Pentiums and earlier, the surround logic should disable
537 * caching for the high addresses through the KEN pin, but
538 * we maintain the tradition of paranoia in this code.
539 */
499f8f84 540 if (!pat_enabled &&
cd7a4e93
AH
541 !(boot_cpu_has(X86_FEATURE_MTRR) ||
542 boot_cpu_has(X86_FEATURE_K6_MTRR) ||
543 boot_cpu_has(X86_FEATURE_CYRIX_ARR) ||
544 boot_cpu_has(X86_FEATURE_CENTAUR_MCR)) &&
545 (pfn << PAGE_SHIFT) >= __pa(high_memory)) {
e7f260a2 546 flags = _PAGE_CACHE_UC;
f0970c13 547 }
548#endif
549
e7f260a2 550 *vma_prot = __pgprot((pgprot_val(*vma_prot) & ~_PAGE_CACHE_MASK) |
551 flags);
f0970c13 552 return 1;
553}
e7f260a2 554
7880f746
VP
555/*
556 * Change the memory type for the physial address range in kernel identity
557 * mapping space if that range is a part of identity map.
558 */
559int kernel_map_sync_memtype(u64 base, unsigned long size, unsigned long flags)
560{
561 unsigned long id_sz;
562
5fc51746 563 if (base >= __pa(high_memory))
7880f746
VP
564 return 0;
565
566 id_sz = (__pa(high_memory) < base + size) ?
567 __pa(high_memory) - base :
568 size;
569
570 if (ioremap_change_attr((unsigned long)__va(base), id_sz, flags) < 0) {
365811d6
BH
571 printk(KERN_INFO "%s:%d ioremap_change_attr failed %s "
572 "for [mem %#010Lx-%#010Lx]\n",
7880f746
VP
573 current->comm, current->pid,
574 cattr_name(flags),
365811d6 575 base, (unsigned long long)(base + size-1));
7880f746
VP
576 return -EINVAL;
577 }
578 return 0;
579}
580
5899329b 581/*
582 * Internal interface to reserve a range of physical memory with prot.
583 * Reserved non RAM regions only and after successful reserve_memtype,
584 * this func also keeps identity mapping (if any) in sync with this new prot.
585 */
cdecff68 586static int reserve_pfn_range(u64 paddr, unsigned long size, pgprot_t *vma_prot,
587 int strict_prot)
5899329b 588{
589 int is_ram = 0;
7880f746 590 int ret;
cdecff68 591 unsigned long want_flags = (pgprot_val(*vma_prot) & _PAGE_CACHE_MASK);
0c3c8a18 592 unsigned long flags = want_flags;
5899329b 593
be03d9e8 594 is_ram = pat_pagerange_is_ram(paddr, paddr + size);
5899329b 595
be03d9e8 596 /*
d886c73c
VP
597 * reserve_pfn_range() for RAM pages. We do not refcount to keep
598 * track of number of mappings of RAM pages. We can assert that
599 * the type requested matches the type of first page in the range.
be03d9e8 600 */
d886c73c
VP
601 if (is_ram) {
602 if (!pat_enabled)
603 return 0;
604
605 flags = lookup_memtype(paddr);
606 if (want_flags != flags) {
365811d6 607 printk(KERN_WARNING "%s:%d map pfn RAM range req %s for [mem %#010Lx-%#010Lx], got %s\n",
d886c73c
VP
608 current->comm, current->pid,
609 cattr_name(want_flags),
610 (unsigned long long)paddr,
365811d6 611 (unsigned long long)(paddr + size - 1),
d886c73c
VP
612 cattr_name(flags));
613 *vma_prot = __pgprot((pgprot_val(*vma_prot) &
614 (~_PAGE_CACHE_MASK)) |
615 flags);
616 }
4bb9c5c0 617 return 0;
d886c73c 618 }
5899329b 619
620 ret = reserve_memtype(paddr, paddr + size, want_flags, &flags);
621 if (ret)
622 return ret;
623
624 if (flags != want_flags) {
1adcaafe
SS
625 if (strict_prot ||
626 !is_new_memtype_allowed(paddr, size, want_flags, flags)) {
cdecff68 627 free_memtype(paddr, paddr + size);
628 printk(KERN_ERR "%s:%d map pfn expected mapping type %s"
365811d6 629 " for [mem %#010Lx-%#010Lx], got %s\n",
cdecff68 630 current->comm, current->pid,
631 cattr_name(want_flags),
632 (unsigned long long)paddr,
365811d6 633 (unsigned long long)(paddr + size - 1),
cdecff68 634 cattr_name(flags));
635 return -EINVAL;
636 }
637 /*
638 * We allow returning different type than the one requested in
639 * non strict case.
640 */
641 *vma_prot = __pgprot((pgprot_val(*vma_prot) &
642 (~_PAGE_CACHE_MASK)) |
643 flags);
5899329b 644 }
645
7880f746 646 if (kernel_map_sync_memtype(paddr, size, flags) < 0) {
5899329b 647 free_memtype(paddr, paddr + size);
5899329b 648 return -EINVAL;
649 }
650 return 0;
651}
652
653/*
654 * Internal interface to free a range of physical memory.
655 * Frees non RAM regions only.
656 */
657static void free_pfn_range(u64 paddr, unsigned long size)
658{
659 int is_ram;
660
be03d9e8 661 is_ram = pat_pagerange_is_ram(paddr, paddr + size);
5899329b 662 if (is_ram == 0)
663 free_memtype(paddr, paddr + size);
664}
665
666/*
667 * track_pfn_vma_copy is called when vma that is covering the pfnmap gets
668 * copied through copy_page_range().
669 *
670 * If the vma has a linear pfn mapping for the entire range, we get the prot
671 * from pte and reserve the entire vma range with single reserve_pfn_range call.
5899329b 672 */
673int track_pfn_vma_copy(struct vm_area_struct *vma)
674{
c1c15b65 675 resource_size_t paddr;
982d789a 676 unsigned long prot;
4b065046 677 unsigned long vma_size = vma->vm_end - vma->vm_start;
cdecff68 678 pgprot_t pgprot;
5899329b 679
5899329b 680 if (is_linear_pfn_mapping(vma)) {
681 /*
982d789a 682 * reserve the whole chunk covered by vma. We need the
683 * starting address and protection from pte.
5899329b 684 */
4b065046 685 if (follow_phys(vma, vma->vm_start, 0, &prot, &paddr)) {
5899329b 686 WARN_ON_ONCE(1);
982d789a 687 return -EINVAL;
5899329b 688 }
cdecff68 689 pgprot = __pgprot(prot);
690 return reserve_pfn_range(paddr, vma_size, &pgprot, 1);
5899329b 691 }
692
5899329b 693 return 0;
5899329b 694}
695
696/*
697 * track_pfn_vma_new is called when a _new_ pfn mapping is being established
698 * for physical range indicated by pfn and size.
699 *
700 * prot is passed in as a parameter for the new mapping. If the vma has a
701 * linear pfn mapping for the entire range reserve the entire vma range with
702 * single reserve_pfn_range call.
5899329b 703 */
e4b866ed 704int track_pfn_vma_new(struct vm_area_struct *vma, pgprot_t *prot,
5899329b 705 unsigned long pfn, unsigned long size)
706{
10876376 707 unsigned long flags;
c1c15b65 708 resource_size_t paddr;
4b065046 709 unsigned long vma_size = vma->vm_end - vma->vm_start;
5899329b 710
5899329b 711 if (is_linear_pfn_mapping(vma)) {
712 /* reserve the whole chunk starting from vm_pgoff */
c1c15b65 713 paddr = (resource_size_t)vma->vm_pgoff << PAGE_SHIFT;
cdecff68 714 return reserve_pfn_range(paddr, vma_size, prot, 0);
5899329b 715 }
716
10876376
VP
717 if (!pat_enabled)
718 return 0;
719
720 /* for vm_insert_pfn and friends, we set prot based on lookup */
721 flags = lookup_memtype(pfn << PAGE_SHIFT);
722 *prot = __pgprot((pgprot_val(vma->vm_page_prot) & (~_PAGE_CACHE_MASK)) |
723 flags);
724
5899329b 725 return 0;
5899329b 726}
727
728/*
729 * untrack_pfn_vma is called while unmapping a pfnmap for a region.
730 * untrack can be called for a specific region indicated by pfn and size or
731 * can be for the entire vma (in which case size can be zero).
732 */
733void untrack_pfn_vma(struct vm_area_struct *vma, unsigned long pfn,
734 unsigned long size)
735{
c1c15b65 736 resource_size_t paddr;
4b065046 737 unsigned long vma_size = vma->vm_end - vma->vm_start;
5899329b 738
5899329b 739 if (is_linear_pfn_mapping(vma)) {
740 /* free the whole chunk starting from vm_pgoff */
c1c15b65 741 paddr = (resource_size_t)vma->vm_pgoff << PAGE_SHIFT;
5899329b 742 free_pfn_range(paddr, vma_size);
743 return;
744 }
5899329b 745}
746
2520bd31 747pgprot_t pgprot_writecombine(pgprot_t prot)
748{
749 if (pat_enabled)
750 return __pgprot(pgprot_val(prot) | _PAGE_CACHE_WC);
751 else
752 return pgprot_noncached(prot);
753}
92b9af9e 754EXPORT_SYMBOL_GPL(pgprot_writecombine);
2520bd31 755
012f09e7 756#if defined(CONFIG_DEBUG_FS) && defined(CONFIG_X86_PAT)
fec0962e 757
fec0962e 758static struct memtype *memtype_get_idx(loff_t pos)
759{
be5a0c12 760 struct memtype *print_entry;
761 int ret;
fec0962e 762
be5a0c12 763 print_entry = kzalloc(sizeof(struct memtype), GFP_KERNEL);
fec0962e 764 if (!print_entry)
765 return NULL;
766
767 spin_lock(&memtype_lock);
9e41a49a 768 ret = rbt_memtype_copy_nth_element(print_entry, pos);
fec0962e 769 spin_unlock(&memtype_lock);
ad2cde16 770
be5a0c12 771 if (!ret) {
772 return print_entry;
773 } else {
774 kfree(print_entry);
775 return NULL;
776 }
fec0962e 777}
778
779static void *memtype_seq_start(struct seq_file *seq, loff_t *pos)
780{
781 if (*pos == 0) {
782 ++*pos;
783 seq_printf(seq, "PAT memtype list:\n");
784 }
785
786 return memtype_get_idx(*pos);
787}
788
789static void *memtype_seq_next(struct seq_file *seq, void *v, loff_t *pos)
790{
791 ++*pos;
792 return memtype_get_idx(*pos);
793}
794
795static void memtype_seq_stop(struct seq_file *seq, void *v)
796{
797}
798
799static int memtype_seq_show(struct seq_file *seq, void *v)
800{
801 struct memtype *print_entry = (struct memtype *)v;
802
803 seq_printf(seq, "%s @ 0x%Lx-0x%Lx\n", cattr_name(print_entry->type),
804 print_entry->start, print_entry->end);
805 kfree(print_entry);
ad2cde16 806
fec0962e 807 return 0;
808}
809
d535e431 810static const struct seq_operations memtype_seq_ops = {
fec0962e 811 .start = memtype_seq_start,
812 .next = memtype_seq_next,
813 .stop = memtype_seq_stop,
814 .show = memtype_seq_show,
815};
816
817static int memtype_seq_open(struct inode *inode, struct file *file)
818{
819 return seq_open(file, &memtype_seq_ops);
820}
821
822static const struct file_operations memtype_fops = {
823 .open = memtype_seq_open,
824 .read = seq_read,
825 .llseek = seq_lseek,
826 .release = seq_release,
827};
828
829static int __init pat_memtype_list_init(void)
830{
dd4377b0
XF
831 if (pat_enabled) {
832 debugfs_create_file("pat_memtype_list", S_IRUSR,
833 arch_debugfs_dir, NULL, &memtype_fops);
834 }
fec0962e 835 return 0;
836}
837
838late_initcall(pat_memtype_list_init);
839
012f09e7 840#endif /* CONFIG_DEBUG_FS && CONFIG_X86_PAT */