~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

TOMOYO Linux Cross Reference
Linux/arch/x86/mm/pat.c

Version: ~ [ linux-5.4-rc3 ] ~ [ linux-5.3.6 ] ~ [ linux-5.2.21 ] ~ [ linux-5.1.21 ] ~ [ linux-5.0.21 ] ~ [ linux-4.20.17 ] ~ [ linux-4.19.79 ] ~ [ linux-4.18.20 ] ~ [ linux-4.17.19 ] ~ [ linux-4.16.18 ] ~ [ linux-4.15.18 ] ~ [ linux-4.14.149 ] ~ [ linux-4.13.16 ] ~ [ linux-4.12.14 ] ~ [ linux-4.11.12 ] ~ [ linux-4.10.17 ] ~ [ linux-4.9.196 ] ~ [ linux-4.8.17 ] ~ [ linux-4.7.10 ] ~ [ linux-4.6.7 ] ~ [ linux-4.5.7 ] ~ [ linux-4.4.196 ] ~ [ linux-4.3.6 ] ~ [ linux-4.2.8 ] ~ [ linux-4.1.52 ] ~ [ linux-4.0.9 ] ~ [ linux-3.19.8 ] ~ [ linux-3.18.140 ] ~ [ linux-3.17.8 ] ~ [ linux-3.16.75 ] ~ [ linux-3.15.10 ] ~ [ linux-3.14.79 ] ~ [ linux-3.13.11 ] ~ [ linux-3.12.74 ] ~ [ linux-3.11.10 ] ~ [ linux-3.10.108 ] ~ [ linux-3.9.11 ] ~ [ linux-3.8.13 ] ~ [ linux-3.7.10 ] ~ [ linux-3.6.11 ] ~ [ linux-3.5.7 ] ~ [ linux-3.4.113 ] ~ [ linux-3.3.8 ] ~ [ linux-3.2.102 ] ~ [ linux-3.1.10 ] ~ [ linux-3.0.101 ] ~ [ linux-2.6.32.71 ] ~ [ linux-2.6.0 ] ~ [ linux-2.4.37.11 ] ~ [ unix-v6-master ] ~ [ ccs-tools-1.8.5 ] ~ [ policy-sample ] ~
Architecture: ~ [ i386 ] ~ [ alpha ] ~ [ m68k ] ~ [ mips ] ~ [ ppc ] ~ [ sparc ] ~ [ sparc64 ] ~

  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 
 35 #ifdef CONFIG_X86_PAT
 36 int __read_mostly pat_enabled = 1;
 37 
 38 static inline void pat_disable(const char *reason)
 39 {
 40         pat_enabled = 0;
 41         printk(KERN_INFO "%s\n", reason);
 42 }
 43 
 44 static int __init nopat(char *str)
 45 {
 46         pat_disable("PAT support disabled.");
 47         return 0;
 48 }
 49 early_param("nopat", nopat);
 50 #else
 51 static inline void pat_disable(const char *reason)
 52 {
 53         (void)reason;
 54 }
 55 #endif
 56 
 57 
 58 int pat_debug_enable;
 59 
 60 static int __init pat_debug_setup(char *str)
 61 {
 62         pat_debug_enable = 1;
 63         return 0;
 64 }
 65 __setup("debugpat", pat_debug_setup);
 66 
 67 static u64 __read_mostly boot_pat_state;
 68 
 69 enum {
 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 
 78 #define PAT(x, y)       ((u64)PAT_ ## y << ((x)*8))
 79 
 80 void pat_init(void)
 81 {
 82         u64 pat;
 83         bool boot_cpu = !boot_pat_state;
 84 
 85         if (!pat_enabled)
 86                 return;
 87 
 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                 }
102         }
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          */
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);
119 
120         /* Boot CPU check */
121         if (!boot_pat_state)
122                 rdmsrl(MSR_IA32_CR_PAT, boot_pat_state);
123 
124         wrmsrl(MSR_IA32_CR_PAT, pat);
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);
129 }
130 
131 #undef PAT
132 
133 static DEFINE_SPINLOCK(memtype_lock);   /* protects memtype accesses */
134 
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  */
142 static unsigned long pat_x_mtrr_type(u64 start, u64 end, unsigned long req_type)
143 {
144         /*
145          * Look for MTRR hint to get the effective type in case where PAT
146          * request is for WB.
147          */
148         if (req_type == _PAGE_CACHE_WB) {
149                 u8 mtrr_type;
150 
151                 mtrr_type = mtrr_type_lookup(start, end);
152                 if (mtrr_type != MTRR_TYPE_WRBACK)
153                         return _PAGE_CACHE_UC_MINUS;
154 
155                 return _PAGE_CACHE_WB;
156         }
157 
158         return req_type;
159 }
160 
161 struct pagerange_state {
162         unsigned long           cur_pfn;
163         int                     ram;
164         int                     not_ram;
165 };
166 
167 static int
168 pagerange_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 
179 static int pat_pagerange_is_ram(resource_size_t start, resource_size_t end)
180 {
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);
199         }
200 
201         return (ret > 0) ? -1 : (state.ram ? 1 : 0);
202 }
203 
204 /*
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
209  */
210 static int reserve_ram_pages_type(u64 start, u64 end, unsigned long req_type,
211                                   unsigned long *new_type)
212 {
213         struct page *page;
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         }
221 
222         for (pfn = (start >> PAGE_SHIFT); pfn < (end >> PAGE_SHIFT); ++pfn) {
223                 unsigned long type;
224 
225                 page = pfn_to_page(pfn);
226                 type = get_page_memtype(page);
227                 if (type != -1) {
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);
230                         if (new_type)
231                                 *new_type = type;
232 
233                         return -EBUSY;
234                 }
235         }
236 
237         if (new_type)
238                 *new_type = req_type;
239 
240         for (pfn = (start >> PAGE_SHIFT); pfn < (end >> PAGE_SHIFT); ++pfn) {
241                 page = pfn_to_page(pfn);
242                 set_page_memtype(page, req_type);
243         }
244         return 0;
245 }
246 
247 static int free_ram_pages_type(u64 start, u64 end)
248 {
249         struct page *page;
250         u64 pfn;
251 
252         for (pfn = (start >> PAGE_SHIFT); pfn < (end >> PAGE_SHIFT); ++pfn) {
253                 page = pfn_to_page(pfn);
254                 set_page_memtype(page, -1);
255         }
256         return 0;
257 }
258 
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  *
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
269  * it will return a negative return value.
270  */
271 int reserve_memtype(u64 start, u64 end, unsigned long req_type,
272                     unsigned long *new_type)
273 {
274         struct memtype *new;
275         unsigned long actual_type;
276         int is_range_ram;
277         int err = 0;
278 
279         BUG_ON(start >= end); /* end is exclusive */
280 
281         if (!pat_enabled) {
282                 /* This is identical to page table setting without PAT */
283                 if (new_type) {
284                         if (req_type == _PAGE_CACHE_WC)
285                                 *new_type = _PAGE_CACHE_UC_MINUS;
286                         else
287                                 *new_type = req_type & _PAGE_CACHE_MASK;
288                 }
289                 return 0;
290         }
291 
292         /* Low ISA region is always mapped WB in page table. No need to track */
293         if (x86_platform.is_untracked_pat_range(start, end)) {
294                 if (new_type)
295                         *new_type = _PAGE_CACHE_WB;
296                 return 0;
297         }
298 
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);
306 
307         if (new_type)
308                 *new_type = actual_type;
309 
310         is_range_ram = pat_pagerange_is_ram(start, end);
311         if (is_range_ram == 1) {
312 
313                 err = reserve_ram_pages_type(start, end, req_type, new_type);
314 
315                 return err;
316         } else if (is_range_ram < 0) {
317                 return -EINVAL;
318         }
319 
320         new  = kzalloc(sizeof(struct memtype), GFP_KERNEL);
321         if (!new)
322                 return -ENOMEM;
323 
324         new->start      = start;
325         new->end        = end;
326         new->type       = actual_type;
327 
328         spin_lock(&memtype_lock);
329 
330         err = rbt_memtype_check_insert(new, new_type);
331         if (err) {
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));
335                 kfree(new);
336                 spin_unlock(&memtype_lock);
337 
338                 return err;
339         }
340 
341         spin_unlock(&memtype_lock);
342 
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),
345                 new_type ? cattr_name(*new_type) : "-");
346 
347         return err;
348 }
349 
350 int free_memtype(u64 start, u64 end)
351 {
352         int err = -EINVAL;
353         int is_range_ram;
354         struct memtype *entry;
355 
356         if (!pat_enabled)
357                 return 0;
358 
359         /* Low ISA region is always mapped WB. No need to track */
360         if (x86_platform.is_untracked_pat_range(start, end))
361                 return 0;
362 
363         is_range_ram = pat_pagerange_is_ram(start, end);
364         if (is_range_ram == 1) {
365 
366                 err = free_ram_pages_type(start, end);
367 
368                 return err;
369         } else if (is_range_ram < 0) {
370                 return -EINVAL;
371         }
372 
373         spin_lock(&memtype_lock);
374         entry = rbt_memtype_erase(start, end);
375         spin_unlock(&memtype_lock);
376 
377         if (!entry) {
378                 printk(KERN_INFO "%s:%d freeing invalid memtype [mem %#010Lx-%#010Lx]\n",
379                        current->comm, current->pid, start, end - 1);
380                 return -EINVAL;
381         }
382 
383         kfree(entry);
384 
385         dprintk("free_memtype request [mem %#010Lx-%#010Lx]\n", start, end - 1);
386 
387         return 0;
388 }
389 
390 
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  */
400 static unsigned long lookup_memtype(u64 paddr)
401 {
402         int rettype = _PAGE_CACHE_WB;
403         struct memtype *entry;
404 
405         if (x86_platform.is_untracked_pat_range(paddr, paddr + PAGE_SIZE))
406                 return rettype;
407 
408         if (pat_pagerange_is_ram(paddr, paddr + PAGE_SIZE)) {
409                 struct page *page;
410                 page = pfn_to_page(paddr >> PAGE_SHIFT);
411                 rettype = get_page_memtype(page);
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 
424         entry = rbt_memtype_lookup(paddr);
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 
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  */
444 int io_reserve_memtype(resource_size_t start, resource_size_t end,
445                         unsigned long *type)
446 {
447         resource_size_t size = end - start;
448         unsigned long req_type = *type;
449         unsigned long new_type;
450         int ret;
451 
452         WARN_ON_ONCE(iomem_map_sanity_check(start, size));
453 
454         ret = reserve_memtype(start, end, req_type, &new_type);
455         if (ret)
456                 goto out_err;
457 
458         if (!is_new_memtype_allowed(start, size, req_type, new_type))
459                 goto out_free;
460 
461         if (kernel_map_sync_memtype(start, size, new_type) < 0)
462                 goto out_free;
463 
464         *type = new_type;
465         return 0;
466 
467 out_free:
468         free_memtype(start, end);
469         ret = -EBUSY;
470 out_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  */
479 void io_free_memtype(resource_size_t start, resource_size_t end)
480 {
481         free_memtype(start, end);
482 }
483 
484 pgprot_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 
490 #ifdef CONFIG_STRICT_DEVMEM
491 /* This check is done in drivers/char/mem.c in case of STRICT_DEVMEM*/
492 static inline int range_is_allowed(unsigned long pfn, unsigned long size)
493 {
494         return 1;
495 }
496 #else
497 /* This check is needed to avoid cache aliasing when PAT is enabled */
498 static 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 
504         if (!pat_enabled)
505                 return 1;
506 
507         while (cursor < to) {
508                 if (!devmem_is_allowed(pfn))
509                         return 0;
510                 cursor += PAGE_SIZE;
511                 pfn++;
512         }
513         return 1;
514 }
515 #endif /* CONFIG_STRICT_DEVMEM */
516 
517 int phys_mem_access_prot_allowed(struct file *file, unsigned long pfn,
518                                 unsigned long size, pgprot_t *vma_prot)
519 {
520         unsigned long flags = _PAGE_CACHE_WB;
521 
522         if (!range_is_allowed(pfn, size))
523                 return 0;
524 
525         if (file->f_flags & O_DSYNC)
526                 flags = _PAGE_CACHE_UC_MINUS;
527 
528 #ifdef CONFIG_X86_32
529         /*
530          * On the PPro and successors, the MTRRs are used to set
531          * memory types for physical addresses outside main memory,
532          * so blindly setting UC or PWT on those pages is wrong.
533          * For Pentiums and earlier, the surround logic should disable
534          * caching for the high addresses through the KEN pin, but
535          * we maintain the tradition of paranoia in this code.
536          */
537         if (!pat_enabled &&
538             !(boot_cpu_has(X86_FEATURE_MTRR) ||
539               boot_cpu_has(X86_FEATURE_K6_MTRR) ||
540               boot_cpu_has(X86_FEATURE_CYRIX_ARR) ||
541               boot_cpu_has(X86_FEATURE_CENTAUR_MCR)) &&
542             (pfn << PAGE_SHIFT) >= __pa(high_memory)) {
543                 flags = _PAGE_CACHE_UC;
544         }
545 #endif
546 
547         *vma_prot = __pgprot((pgprot_val(*vma_prot) & ~_PAGE_CACHE_MASK) |
548                              flags);
549         return 1;
550 }
551 
552 /*
553  * Change the memory type for the physial address range in kernel identity
554  * mapping space if that range is a part of identity map.
555  */
556 int kernel_map_sync_memtype(u64 base, unsigned long size, unsigned long flags)
557 {
558         unsigned long id_sz;
559 
560         if (base > __pa(high_memory-1))
561                 return 0;
562 
563         /*
564          * some areas in the middle of the kernel identity range
565          * are not mapped, like the PCI space.
566          */
567         if (!page_is_ram(base >> PAGE_SHIFT))
568                 return 0;
569 
570         id_sz = (__pa(high_memory-1) <= base + size) ?
571                                 __pa(high_memory) - base :
572                                 size;
573 
574         if (ioremap_change_attr((unsigned long)__va(base), id_sz, flags) < 0) {
575                 printk(KERN_INFO "%s:%d ioremap_change_attr failed %s "
576                         "for [mem %#010Lx-%#010Lx]\n",
577                         current->comm, current->pid,
578                         cattr_name(flags),
579                         base, (unsigned long long)(base + size-1));
580                 return -EINVAL;
581         }
582         return 0;
583 }
584 
585 /*
586  * Internal interface to reserve a range of physical memory with prot.
587  * Reserved non RAM regions only and after successful reserve_memtype,
588  * this func also keeps identity mapping (if any) in sync with this new prot.
589  */
590 static int reserve_pfn_range(u64 paddr, unsigned long size, pgprot_t *vma_prot,
591                                 int strict_prot)
592 {
593         int is_ram = 0;
594         int ret;
595         unsigned long want_flags = (pgprot_val(*vma_prot) & _PAGE_CACHE_MASK);
596         unsigned long flags = want_flags;
597 
598         is_ram = pat_pagerange_is_ram(paddr, paddr + size);
599 
600         /*
601          * reserve_pfn_range() for RAM pages. We do not refcount to keep
602          * track of number of mappings of RAM pages. We can assert that
603          * the type requested matches the type of first page in the range.
604          */
605         if (is_ram) {
606                 if (!pat_enabled)
607                         return 0;
608 
609                 flags = lookup_memtype(paddr);
610                 if (want_flags != flags) {
611                         printk(KERN_WARNING "%s:%d map pfn RAM range req %s for [mem %#010Lx-%#010Lx], got %s\n",
612                                 current->comm, current->pid,
613                                 cattr_name(want_flags),
614                                 (unsigned long long)paddr,
615                                 (unsigned long long)(paddr + size - 1),
616                                 cattr_name(flags));
617                         *vma_prot = __pgprot((pgprot_val(*vma_prot) &
618                                               (~_PAGE_CACHE_MASK)) |
619                                              flags);
620                 }
621                 return 0;
622         }
623 
624         ret = reserve_memtype(paddr, paddr + size, want_flags, &flags);
625         if (ret)
626                 return ret;
627 
628         if (flags != want_flags) {
629                 if (strict_prot ||
630                     !is_new_memtype_allowed(paddr, size, want_flags, flags)) {
631                         free_memtype(paddr, paddr + size);
632                         printk(KERN_ERR "%s:%d map pfn expected mapping type %s"
633                                 " for [mem %#010Lx-%#010Lx], got %s\n",
634                                 current->comm, current->pid,
635                                 cattr_name(want_flags),
636                                 (unsigned long long)paddr,
637                                 (unsigned long long)(paddr + size - 1),
638                                 cattr_name(flags));
639                         return -EINVAL;
640                 }
641                 /*
642                  * We allow returning different type than the one requested in
643                  * non strict case.
644                  */
645                 *vma_prot = __pgprot((pgprot_val(*vma_prot) &
646                                       (~_PAGE_CACHE_MASK)) |
647                                      flags);
648         }
649 
650         if (kernel_map_sync_memtype(paddr, size, flags) < 0) {
651                 free_memtype(paddr, paddr + size);
652                 return -EINVAL;
653         }
654         return 0;
655 }
656 
657 /*
658  * Internal interface to free a range of physical memory.
659  * Frees non RAM regions only.
660  */
661 static void free_pfn_range(u64 paddr, unsigned long size)
662 {
663         int is_ram;
664 
665         is_ram = pat_pagerange_is_ram(paddr, paddr + size);
666         if (is_ram == 0)
667                 free_memtype(paddr, paddr + size);
668 }
669 
670 /*
671  * track_pfn_copy is called when vma that is covering the pfnmap gets
672  * copied through copy_page_range().
673  *
674  * If the vma has a linear pfn mapping for the entire range, we get the prot
675  * from pte and reserve the entire vma range with single reserve_pfn_range call.
676  */
677 int track_pfn_copy(struct vm_area_struct *vma)
678 {
679         resource_size_t paddr;
680         unsigned long prot;
681         unsigned long vma_size = vma->vm_end - vma->vm_start;
682         pgprot_t pgprot;
683 
684         if (vma->vm_flags & VM_PAT) {
685                 /*
686                  * reserve the whole chunk covered by vma. We need the
687                  * starting address and protection from pte.
688                  */
689                 if (follow_phys(vma, vma->vm_start, 0, &prot, &paddr)) {
690                         WARN_ON_ONCE(1);
691                         return -EINVAL;
692                 }
693                 pgprot = __pgprot(prot);
694                 return reserve_pfn_range(paddr, vma_size, &pgprot, 1);
695         }
696 
697         return 0;
698 }
699 
700 /*
701  * prot is passed in as a parameter for the new mapping. If the vma has a
702  * linear pfn mapping for the entire range reserve the entire vma range with
703  * single reserve_pfn_range call.
704  */
705 int track_pfn_remap(struct vm_area_struct *vma, pgprot_t *prot,
706                     unsigned long pfn, unsigned long addr, unsigned long size)
707 {
708         resource_size_t paddr = (resource_size_t)pfn << PAGE_SHIFT;
709         unsigned long flags;
710 
711         /* reserve the whole chunk starting from paddr */
712         if (addr == vma->vm_start && size == (vma->vm_end - vma->vm_start)) {
713                 int ret;
714 
715                 ret = reserve_pfn_range(paddr, size, prot, 0);
716                 if (!ret)
717                         vma->vm_flags |= VM_PAT;
718                 return ret;
719         }
720 
721         if (!pat_enabled)
722                 return 0;
723 
724         /*
725          * For anything smaller than the vma size we set prot based on the
726          * lookup.
727          */
728         flags = lookup_memtype(paddr);
729 
730         /* Check memtype for the remaining pages */
731         while (size > PAGE_SIZE) {
732                 size -= PAGE_SIZE;
733                 paddr += PAGE_SIZE;
734                 if (flags != lookup_memtype(paddr))
735                         return -EINVAL;
736         }
737 
738         *prot = __pgprot((pgprot_val(vma->vm_page_prot) & (~_PAGE_CACHE_MASK)) |
739                          flags);
740 
741         return 0;
742 }
743 
744 int track_pfn_insert(struct vm_area_struct *vma, pgprot_t *prot,
745                      unsigned long pfn)
746 {
747         unsigned long flags;
748 
749         if (!pat_enabled)
750                 return 0;
751 
752         /* Set prot based on lookup */
753         flags = lookup_memtype((resource_size_t)pfn << PAGE_SHIFT);
754         *prot = __pgprot((pgprot_val(vma->vm_page_prot) & (~_PAGE_CACHE_MASK)) |
755                          flags);
756 
757         return 0;
758 }
759 
760 /*
761  * untrack_pfn is called while unmapping a pfnmap for a region.
762  * untrack can be called for a specific region indicated by pfn and size or
763  * can be for the entire vma (in which case pfn, size are zero).
764  */
765 void untrack_pfn(struct vm_area_struct *vma, unsigned long pfn,
766                  unsigned long size)
767 {
768         resource_size_t paddr;
769         unsigned long prot;
770 
771         if (!(vma->vm_flags & VM_PAT))
772                 return;
773 
774         /* free the chunk starting from pfn or the whole chunk */
775         paddr = (resource_size_t)pfn << PAGE_SHIFT;
776         if (!paddr && !size) {
777                 if (follow_phys(vma, vma->vm_start, 0, &prot, &paddr)) {
778                         WARN_ON_ONCE(1);
779                         return;
780                 }
781 
782                 size = vma->vm_end - vma->vm_start;
783         }
784         free_pfn_range(paddr, size);
785         vma->vm_flags &= ~VM_PAT;
786 }
787 
788 pgprot_t pgprot_writecombine(pgprot_t prot)
789 {
790         if (pat_enabled)
791                 return __pgprot(pgprot_val(prot) | _PAGE_CACHE_WC);
792         else
793                 return pgprot_noncached(prot);
794 }
795 EXPORT_SYMBOL_GPL(pgprot_writecombine);
796 
797 #if defined(CONFIG_DEBUG_FS) && defined(CONFIG_X86_PAT)
798 
799 static struct memtype *memtype_get_idx(loff_t pos)
800 {
801         struct memtype *print_entry;
802         int ret;
803 
804         print_entry  = kzalloc(sizeof(struct memtype), GFP_KERNEL);
805         if (!print_entry)
806                 return NULL;
807 
808         spin_lock(&memtype_lock);
809         ret = rbt_memtype_copy_nth_element(print_entry, pos);
810         spin_unlock(&memtype_lock);
811 
812         if (!ret) {
813                 return print_entry;
814         } else {
815                 kfree(print_entry);
816                 return NULL;
817         }
818 }
819 
820 static void *memtype_seq_start(struct seq_file *seq, loff_t *pos)
821 {
822         if (*pos == 0) {
823                 ++*pos;
824                 seq_printf(seq, "PAT memtype list:\n");
825         }
826 
827         return memtype_get_idx(*pos);
828 }
829 
830 static void *memtype_seq_next(struct seq_file *seq, void *v, loff_t *pos)
831 {
832         ++*pos;
833         return memtype_get_idx(*pos);
834 }
835 
836 static void memtype_seq_stop(struct seq_file *seq, void *v)
837 {
838 }
839 
840 static int memtype_seq_show(struct seq_file *seq, void *v)
841 {
842         struct memtype *print_entry = (struct memtype *)v;
843 
844         seq_printf(seq, "%s @ 0x%Lx-0x%Lx\n", cattr_name(print_entry->type),
845                         print_entry->start, print_entry->end);
846         kfree(print_entry);
847 
848         return 0;
849 }
850 
851 static const struct seq_operations memtype_seq_ops = {
852         .start = memtype_seq_start,
853         .next  = memtype_seq_next,
854         .stop  = memtype_seq_stop,
855         .show  = memtype_seq_show,
856 };
857 
858 static int memtype_seq_open(struct inode *inode, struct file *file)
859 {
860         return seq_open(file, &memtype_seq_ops);
861 }
862 
863 static const struct file_operations memtype_fops = {
864         .open    = memtype_seq_open,
865         .read    = seq_read,
866         .llseek  = seq_lseek,
867         .release = seq_release,
868 };
869 
870 static int __init pat_memtype_list_init(void)
871 {
872         if (pat_enabled) {
873                 debugfs_create_file("pat_memtype_list", S_IRUSR,
874                                     arch_debugfs_dir, NULL, &memtype_fops);
875         }
876         return 0;
877 }
878 
879 late_initcall(pat_memtype_list_init);
880 
881 #endif /* CONFIG_DEBUG_FS && CONFIG_X86_PAT */
882 

~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

kernel.org | git.kernel.org | LWN.net | Project Home | Wiki (Japanese) | Wiki (English) | SVN repository | Mail admin

Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.

osdn.jp