2 * Handle caching attributes in page tables (PAT)
4 * Authors: Venkatesh Pallipadi <venkatesh.pallipadi@intel.com>
5 * Suresh B Siddha <suresh.b.siddha@intel.com>
7 * Loosely based on earlier PAT patchset from Eric Biederman and Andi Kleen.
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>
18 #include <linux/rbtree.h>
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>
33 #include "pat_internal.h"
34 #include "mm_internal.h"
37 #define pr_fmt(fmt) "" fmt
39 static bool boot_cpu_done;
41 static int __read_mostly __pat_enabled = IS_ENABLED(CONFIG_X86_PAT);
43 static inline void pat_disable(const char *reason)
46 pr_info("x86/PAT: %s\n", reason);
49 static int __init nopat(char *str)
51 pat_disable("PAT support disabled.");
54 early_param("nopat", nopat);
56 bool pat_enabled(void)
58 return !!__pat_enabled;
60 EXPORT_SYMBOL_GPL(pat_enabled);
64 static int __init pat_debug_setup(char *str)
69 __setup("debugpat", pat_debug_setup);
73 * X86 PAT uses page flags arch_1 and uncached together to keep track of
74 * memory type of pages that have backing page struct.
76 * X86 PAT supports 4 different memory types:
77 * - _PAGE_CACHE_MODE_WB
78 * - _PAGE_CACHE_MODE_WC
79 * - _PAGE_CACHE_MODE_UC_MINUS
80 * - _PAGE_CACHE_MODE_WT
82 * _PAGE_CACHE_MODE_WB is the default type.
86 #define _PGMT_WC (1UL << PG_arch_1)
87 #define _PGMT_UC_MINUS (1UL << PG_uncached)
88 #define _PGMT_WT (1UL << PG_uncached | 1UL << PG_arch_1)
89 #define _PGMT_MASK (1UL << PG_uncached | 1UL << PG_arch_1)
90 #define _PGMT_CLEAR_MASK (~_PGMT_MASK)
92 static inline enum page_cache_mode get_page_memtype(struct page *pg)
94 unsigned long pg_flags = pg->flags & _PGMT_MASK;
96 if (pg_flags == _PGMT_WB)
97 return _PAGE_CACHE_MODE_WB;
98 else if (pg_flags == _PGMT_WC)
99 return _PAGE_CACHE_MODE_WC;
100 else if (pg_flags == _PGMT_UC_MINUS)
101 return _PAGE_CACHE_MODE_UC_MINUS;
103 return _PAGE_CACHE_MODE_WT;
106 static inline void set_page_memtype(struct page *pg,
107 enum page_cache_mode memtype)
109 unsigned long memtype_flags;
110 unsigned long old_flags;
111 unsigned long new_flags;
114 case _PAGE_CACHE_MODE_WC:
115 memtype_flags = _PGMT_WC;
117 case _PAGE_CACHE_MODE_UC_MINUS:
118 memtype_flags = _PGMT_UC_MINUS;
120 case _PAGE_CACHE_MODE_WT:
121 memtype_flags = _PGMT_WT;
123 case _PAGE_CACHE_MODE_WB:
125 memtype_flags = _PGMT_WB;
130 old_flags = pg->flags;
131 new_flags = (old_flags & _PGMT_CLEAR_MASK) | memtype_flags;
132 } while (cmpxchg(&pg->flags, old_flags, new_flags) != old_flags);
135 static inline enum page_cache_mode get_page_memtype(struct page *pg)
139 static inline void set_page_memtype(struct page *pg,
140 enum page_cache_mode memtype)
146 PAT_UC = 0, /* uncached */
147 PAT_WC = 1, /* Write combining */
148 PAT_WT = 4, /* Write Through */
149 PAT_WP = 5, /* Write Protected */
150 PAT_WB = 6, /* Write Back (default) */
151 PAT_UC_MINUS = 7, /* UC, but can be overriden by MTRR */
154 #define CM(c) (_PAGE_CACHE_MODE_ ## c)
156 static enum page_cache_mode pat_get_cache_mode(unsigned pat_val, char *msg)
158 enum page_cache_mode cache;
162 case PAT_UC: cache = CM(UC); cache_mode = "UC "; break;
163 case PAT_WC: cache = CM(WC); cache_mode = "WC "; break;
164 case PAT_WT: cache = CM(WT); cache_mode = "WT "; break;
165 case PAT_WP: cache = CM(WP); cache_mode = "WP "; break;
166 case PAT_WB: cache = CM(WB); cache_mode = "WB "; break;
167 case PAT_UC_MINUS: cache = CM(UC_MINUS); cache_mode = "UC- "; break;
168 default: cache = CM(WB); cache_mode = "WB "; break;
171 memcpy(msg, cache_mode, 4);
179 * Update the cache mode to pgprot translation tables according to PAT
181 * Using lower indices is preferred, so we start with highest index.
183 void pat_init_cache_modes(u64 pat)
185 enum page_cache_mode cache;
190 for (i = 7; i >= 0; i--) {
191 cache = pat_get_cache_mode((pat >> (i * 8)) & 7,
193 update_cache_mode_entry(i, cache);
195 pr_info("x86/PAT: Configuration [0-7]: %s\n", pat_msg);
198 #define PAT(x, y) ((u64)PAT_ ## y << ((x)*8))
200 static void pat_bsp_init(u64 pat)
205 pat_disable("PAT not supported by CPU.");
212 rdmsrl(MSR_IA32_CR_PAT, tmp_pat);
214 pat_disable("PAT MSR is 0, disabled.");
218 wrmsrl(MSR_IA32_CR_PAT, pat);
221 pat_init_cache_modes(pat);
224 static void pat_ap_init(u64 pat)
231 * If this happens we are on a secondary CPU, but switched to
232 * PAT on the boot CPU. We have no way to undo PAT.
234 panic("x86/PAT: PAT enabled, but not supported by secondary CPU\n");
237 wrmsrl(MSR_IA32_CR_PAT, pat);
243 struct cpuinfo_x86 *c = &boot_cpu_data;
245 if (!pat_enabled()) {
247 * No PAT. Emulate the PAT table that corresponds to the two
248 * cache bits, PWT (Write Through) and PCD (Cache Disable). This
249 * setup is the same as the BIOS default setup when the system
250 * has PAT but the "nopat" boot option has been specified. This
251 * emulated PAT table is used when MSR_IA32_CR_PAT returns 0.
258 * 00 0 WB : _PAGE_CACHE_MODE_WB
259 * 01 1 WT : _PAGE_CACHE_MODE_WT
260 * 10 2 UC-: _PAGE_CACHE_MODE_UC_MINUS
261 * 11 3 UC : _PAGE_CACHE_MODE_UC
263 * NOTE: When WC or WP is used, it is redirected to UC- per
264 * the default setup in __cachemode2pte_tbl[].
266 pat = PAT(0, WB) | PAT(1, WT) | PAT(2, UC_MINUS) | PAT(3, UC) |
267 PAT(4, WB) | PAT(5, WT) | PAT(6, UC_MINUS) | PAT(7, UC);
269 } else if ((c->x86_vendor == X86_VENDOR_INTEL) &&
270 (((c->x86 == 0x6) && (c->x86_model <= 0xd)) ||
271 ((c->x86 == 0xf) && (c->x86_model <= 0x6)))) {
273 * PAT support with the lower four entries. Intel Pentium 2,
274 * 3, M, and 4 are affected by PAT errata, which makes the
275 * upper four entries unusable. To be on the safe side, we don't
283 * 000 0 WB : _PAGE_CACHE_MODE_WB
284 * 001 1 WC : _PAGE_CACHE_MODE_WC
285 * 010 2 UC-: _PAGE_CACHE_MODE_UC_MINUS
286 * 011 3 UC : _PAGE_CACHE_MODE_UC
289 * NOTE: When WT or WP is used, it is redirected to UC- per
290 * the default setup in __cachemode2pte_tbl[].
292 pat = PAT(0, WB) | PAT(1, WC) | PAT(2, UC_MINUS) | PAT(3, UC) |
293 PAT(4, WB) | PAT(5, WC) | PAT(6, UC_MINUS) | PAT(7, UC);
296 * Full PAT support. We put WT in slot 7 to improve
297 * robustness in the presence of errata that might cause
298 * the high PAT bit to be ignored. This way, a buggy slot 7
299 * access will hit slot 3, and slot 3 is UC, so at worst
300 * we lose performance without causing a correctness issue.
301 * Pentium 4 erratum N46 is an example for such an erratum,
302 * although we try not to use PAT at all on affected CPUs.
309 * 000 0 WB : _PAGE_CACHE_MODE_WB
310 * 001 1 WC : _PAGE_CACHE_MODE_WC
311 * 010 2 UC-: _PAGE_CACHE_MODE_UC_MINUS
312 * 011 3 UC : _PAGE_CACHE_MODE_UC
313 * 100 4 WB : Reserved
314 * 101 5 WC : Reserved
315 * 110 6 UC-: Reserved
316 * 111 7 WT : _PAGE_CACHE_MODE_WT
318 * The reserved slots are unused, but mapped to their
319 * corresponding types in the presence of PAT errata.
321 pat = PAT(0, WB) | PAT(1, WC) | PAT(2, UC_MINUS) | PAT(3, UC) |
322 PAT(4, WB) | PAT(5, WC) | PAT(6, UC_MINUS) | PAT(7, WT);
325 if (!boot_cpu_done) {
327 boot_cpu_done = true;
335 static DEFINE_SPINLOCK(memtype_lock); /* protects memtype accesses */
338 * Does intersection of PAT memory type and MTRR memory type and returns
339 * the resulting memory type as PAT understands it.
340 * (Type in pat and mtrr will not have same value)
341 * The intersection is based on "Effective Memory Type" tables in IA-32
344 static unsigned long pat_x_mtrr_type(u64 start, u64 end,
345 enum page_cache_mode req_type)
348 * Look for MTRR hint to get the effective type in case where PAT
351 if (req_type == _PAGE_CACHE_MODE_WB) {
352 u8 mtrr_type, uniform;
354 mtrr_type = mtrr_type_lookup(start, end, &uniform);
355 if (mtrr_type != MTRR_TYPE_WRBACK)
356 return _PAGE_CACHE_MODE_UC_MINUS;
358 return _PAGE_CACHE_MODE_WB;
364 struct pagerange_state {
365 unsigned long cur_pfn;
371 pagerange_is_ram_callback(unsigned long initial_pfn, unsigned long total_nr_pages, void *arg)
373 struct pagerange_state *state = arg;
375 state->not_ram |= initial_pfn > state->cur_pfn;
376 state->ram |= total_nr_pages > 0;
377 state->cur_pfn = initial_pfn + total_nr_pages;
379 return state->ram && state->not_ram;
382 static int pat_pagerange_is_ram(resource_size_t start, resource_size_t end)
385 unsigned long start_pfn = start >> PAGE_SHIFT;
386 unsigned long end_pfn = (end + PAGE_SIZE - 1) >> PAGE_SHIFT;
387 struct pagerange_state state = {start_pfn, 0, 0};
390 * For legacy reasons, physical address range in the legacy ISA
391 * region is tracked as non-RAM. This will allow users of
392 * /dev/mem to map portions of legacy ISA region, even when
393 * some of those portions are listed(or not even listed) with
394 * different e820 types(RAM/reserved/..)
396 if (start_pfn < ISA_END_ADDRESS >> PAGE_SHIFT)
397 start_pfn = ISA_END_ADDRESS >> PAGE_SHIFT;
399 if (start_pfn < end_pfn) {
400 ret = walk_system_ram_range(start_pfn, end_pfn - start_pfn,
401 &state, pagerange_is_ram_callback);
404 return (ret > 0) ? -1 : (state.ram ? 1 : 0);
408 * For RAM pages, we use page flags to mark the pages with appropriate type.
409 * The page flags are limited to four types, WB (default), WC, WT and UC-.
410 * WP request fails with -EINVAL, and UC gets redirected to UC-. Setting
411 * a new memory type is only allowed for a page mapped with the default WB
414 * Here we do two passes:
415 * - Find the memtype of all the pages in the range, look for any conflicts.
416 * - In case of no conflicts, set the new memtype for pages in the range.
418 static int reserve_ram_pages_type(u64 start, u64 end,
419 enum page_cache_mode req_type,
420 enum page_cache_mode *new_type)
425 if (req_type == _PAGE_CACHE_MODE_WP) {
427 *new_type = _PAGE_CACHE_MODE_UC_MINUS;
431 if (req_type == _PAGE_CACHE_MODE_UC) {
432 /* We do not support strong UC */
434 req_type = _PAGE_CACHE_MODE_UC_MINUS;
437 for (pfn = (start >> PAGE_SHIFT); pfn < (end >> PAGE_SHIFT); ++pfn) {
438 enum page_cache_mode type;
440 page = pfn_to_page(pfn);
441 type = get_page_memtype(page);
442 if (type != _PAGE_CACHE_MODE_WB) {
443 pr_info("x86/PAT: reserve_ram_pages_type failed [mem %#010Lx-%#010Lx], track 0x%x, req 0x%x\n",
444 start, end - 1, type, req_type);
453 *new_type = req_type;
455 for (pfn = (start >> PAGE_SHIFT); pfn < (end >> PAGE_SHIFT); ++pfn) {
456 page = pfn_to_page(pfn);
457 set_page_memtype(page, req_type);
462 static int free_ram_pages_type(u64 start, u64 end)
467 for (pfn = (start >> PAGE_SHIFT); pfn < (end >> PAGE_SHIFT); ++pfn) {
468 page = pfn_to_page(pfn);
469 set_page_memtype(page, _PAGE_CACHE_MODE_WB);
475 * req_type typically has one of the:
476 * - _PAGE_CACHE_MODE_WB
477 * - _PAGE_CACHE_MODE_WC
478 * - _PAGE_CACHE_MODE_UC_MINUS
479 * - _PAGE_CACHE_MODE_UC
480 * - _PAGE_CACHE_MODE_WT
482 * If new_type is NULL, function will return an error if it cannot reserve the
483 * region with req_type. If new_type is non-NULL, function will return
484 * available type in new_type in case of no error. In case of any error
485 * it will return a negative return value.
487 int reserve_memtype(u64 start, u64 end, enum page_cache_mode req_type,
488 enum page_cache_mode *new_type)
491 enum page_cache_mode actual_type;
495 BUG_ON(start >= end); /* end is exclusive */
497 if (!pat_enabled()) {
498 /* This is identical to page table setting without PAT */
500 *new_type = req_type;
504 /* Low ISA region is always mapped WB in page table. No need to track */
505 if (x86_platform.is_untracked_pat_range(start, end)) {
507 *new_type = _PAGE_CACHE_MODE_WB;
512 * Call mtrr_lookup to get the type hint. This is an
513 * optimization for /dev/mem mmap'ers into WB memory (BIOS
514 * tools and ACPI tools). Use WB request for WB memory and use
515 * UC_MINUS otherwise.
517 actual_type = pat_x_mtrr_type(start, end, req_type);
520 *new_type = actual_type;
522 is_range_ram = pat_pagerange_is_ram(start, end);
523 if (is_range_ram == 1) {
525 err = reserve_ram_pages_type(start, end, req_type, new_type);
528 } else if (is_range_ram < 0) {
532 new = kzalloc(sizeof(struct memtype), GFP_KERNEL);
538 new->type = actual_type;
540 spin_lock(&memtype_lock);
542 err = rbt_memtype_check_insert(new, new_type);
544 pr_info("x86/PAT: reserve_memtype failed [mem %#010Lx-%#010Lx], track %s, req %s\n",
546 cattr_name(new->type), cattr_name(req_type));
548 spin_unlock(&memtype_lock);
553 spin_unlock(&memtype_lock);
555 dprintk("reserve_memtype added [mem %#010Lx-%#010Lx], track %s, req %s, ret %s\n",
556 start, end - 1, cattr_name(new->type), cattr_name(req_type),
557 new_type ? cattr_name(*new_type) : "-");
562 int free_memtype(u64 start, u64 end)
566 struct memtype *entry;
571 /* Low ISA region is always mapped WB. No need to track */
572 if (x86_platform.is_untracked_pat_range(start, end))
575 is_range_ram = pat_pagerange_is_ram(start, end);
576 if (is_range_ram == 1) {
578 err = free_ram_pages_type(start, end);
581 } else if (is_range_ram < 0) {
585 spin_lock(&memtype_lock);
586 entry = rbt_memtype_erase(start, end);
587 spin_unlock(&memtype_lock);
590 pr_info("x86/PAT: %s:%d freeing invalid memtype [mem %#010Lx-%#010Lx]\n",
591 current->comm, current->pid, start, end - 1);
597 dprintk("free_memtype request [mem %#010Lx-%#010Lx]\n", start, end - 1);
604 * lookup_memtype - Looksup the memory type for a physical address
605 * @paddr: physical address of which memory type needs to be looked up
607 * Only to be called when PAT is enabled
609 * Returns _PAGE_CACHE_MODE_WB, _PAGE_CACHE_MODE_WC, _PAGE_CACHE_MODE_UC_MINUS
610 * or _PAGE_CACHE_MODE_WT.
612 static enum page_cache_mode lookup_memtype(u64 paddr)
614 enum page_cache_mode rettype = _PAGE_CACHE_MODE_WB;
615 struct memtype *entry;
617 if (x86_platform.is_untracked_pat_range(paddr, paddr + PAGE_SIZE))
620 if (pat_pagerange_is_ram(paddr, paddr + PAGE_SIZE)) {
623 page = pfn_to_page(paddr >> PAGE_SHIFT);
624 return get_page_memtype(page);
627 spin_lock(&memtype_lock);
629 entry = rbt_memtype_lookup(paddr);
631 rettype = entry->type;
633 rettype = _PAGE_CACHE_MODE_UC_MINUS;
635 spin_unlock(&memtype_lock);
640 * io_reserve_memtype - Request a memory type mapping for a region of memory
641 * @start: start (physical address) of the region
642 * @end: end (physical address) of the region
643 * @type: A pointer to memtype, with requested type. On success, requested
644 * or any other compatible type that was available for the region is returned
646 * On success, returns 0
647 * On failure, returns non-zero
649 int io_reserve_memtype(resource_size_t start, resource_size_t end,
650 enum page_cache_mode *type)
652 resource_size_t size = end - start;
653 enum page_cache_mode req_type = *type;
654 enum page_cache_mode new_type;
657 WARN_ON_ONCE(iomem_map_sanity_check(start, size));
659 ret = reserve_memtype(start, end, req_type, &new_type);
663 if (!is_new_memtype_allowed(start, size, req_type, new_type))
666 if (kernel_map_sync_memtype(start, size, new_type) < 0)
673 free_memtype(start, end);
680 * io_free_memtype - Release a memory type mapping for a region of memory
681 * @start: start (physical address) of the region
682 * @end: end (physical address) of the region
684 void io_free_memtype(resource_size_t start, resource_size_t end)
686 free_memtype(start, end);
689 pgprot_t phys_mem_access_prot(struct file *file, unsigned long pfn,
690 unsigned long size, pgprot_t vma_prot)
695 #ifdef CONFIG_STRICT_DEVMEM
696 /* This check is done in drivers/char/mem.c in case of STRICT_DEVMEM */
697 static inline int range_is_allowed(unsigned long pfn, unsigned long size)
702 /* This check is needed to avoid cache aliasing when PAT is enabled */
703 static inline int range_is_allowed(unsigned long pfn, unsigned long size)
705 u64 from = ((u64)pfn) << PAGE_SHIFT;
706 u64 to = from + size;
712 while (cursor < to) {
713 if (!devmem_is_allowed(pfn)) {
714 pr_info("x86/PAT: Program %s tried to access /dev/mem between [mem %#010Lx-%#010Lx], PAT prevents it\n",
715 current->comm, from, to - 1);
723 #endif /* CONFIG_STRICT_DEVMEM */
725 int phys_mem_access_prot_allowed(struct file *file, unsigned long pfn,
726 unsigned long size, pgprot_t *vma_prot)
728 enum page_cache_mode pcm = _PAGE_CACHE_MODE_WB;
730 if (!range_is_allowed(pfn, size))
733 if (file->f_flags & O_DSYNC)
734 pcm = _PAGE_CACHE_MODE_UC_MINUS;
738 * On the PPro and successors, the MTRRs are used to set
739 * memory types for physical addresses outside main memory,
740 * so blindly setting UC or PWT on those pages is wrong.
741 * For Pentiums and earlier, the surround logic should disable
742 * caching for the high addresses through the KEN pin, but
743 * we maintain the tradition of paranoia in this code.
745 if (!pat_enabled() &&
746 !(boot_cpu_has(X86_FEATURE_MTRR) ||
747 boot_cpu_has(X86_FEATURE_K6_MTRR) ||
748 boot_cpu_has(X86_FEATURE_CYRIX_ARR) ||
749 boot_cpu_has(X86_FEATURE_CENTAUR_MCR)) &&
750 (pfn << PAGE_SHIFT) >= __pa(high_memory)) {
751 pcm = _PAGE_CACHE_MODE_UC;
755 *vma_prot = __pgprot((pgprot_val(*vma_prot) & ~_PAGE_CACHE_MASK) |
756 cachemode2protval(pcm));
761 * Change the memory type for the physial address range in kernel identity
762 * mapping space if that range is a part of identity map.
764 int kernel_map_sync_memtype(u64 base, unsigned long size,
765 enum page_cache_mode pcm)
769 if (base > __pa(high_memory-1))
773 * some areas in the middle of the kernel identity range
774 * are not mapped, like the PCI space.
776 if (!page_is_ram(base >> PAGE_SHIFT))
779 id_sz = (__pa(high_memory-1) <= base + size) ?
780 __pa(high_memory) - base :
783 if (ioremap_change_attr((unsigned long)__va(base), id_sz, pcm) < 0) {
784 pr_info("x86/PAT: %s:%d ioremap_change_attr failed %s for [mem %#010Lx-%#010Lx]\n",
785 current->comm, current->pid,
787 base, (unsigned long long)(base + size-1));
794 * Internal interface to reserve a range of physical memory with prot.
795 * Reserved non RAM regions only and after successful reserve_memtype,
796 * this func also keeps identity mapping (if any) in sync with this new prot.
798 static int reserve_pfn_range(u64 paddr, unsigned long size, pgprot_t *vma_prot,
803 enum page_cache_mode want_pcm = pgprot2cachemode(*vma_prot);
804 enum page_cache_mode pcm = want_pcm;
806 is_ram = pat_pagerange_is_ram(paddr, paddr + size);
809 * reserve_pfn_range() for RAM pages. We do not refcount to keep
810 * track of number of mappings of RAM pages. We can assert that
811 * the type requested matches the type of first page in the range.
817 pcm = lookup_memtype(paddr);
818 if (want_pcm != pcm) {
819 pr_warn("x86/PAT: %s:%d map pfn RAM range req %s for [mem %#010Lx-%#010Lx], got %s\n",
820 current->comm, current->pid,
821 cattr_name(want_pcm),
822 (unsigned long long)paddr,
823 (unsigned long long)(paddr + size - 1),
825 *vma_prot = __pgprot((pgprot_val(*vma_prot) &
826 (~_PAGE_CACHE_MASK)) |
827 cachemode2protval(pcm));
832 ret = reserve_memtype(paddr, paddr + size, want_pcm, &pcm);
836 if (pcm != want_pcm) {
838 !is_new_memtype_allowed(paddr, size, want_pcm, pcm)) {
839 free_memtype(paddr, paddr + size);
840 pr_err("x86/PAT: %s:%d map pfn expected mapping type %s for [mem %#010Lx-%#010Lx], got %s\n",
841 current->comm, current->pid,
842 cattr_name(want_pcm),
843 (unsigned long long)paddr,
844 (unsigned long long)(paddr + size - 1),
849 * We allow returning different type than the one requested in
852 *vma_prot = __pgprot((pgprot_val(*vma_prot) &
853 (~_PAGE_CACHE_MASK)) |
854 cachemode2protval(pcm));
857 if (kernel_map_sync_memtype(paddr, size, pcm) < 0) {
858 free_memtype(paddr, paddr + size);
865 * Internal interface to free a range of physical memory.
866 * Frees non RAM regions only.
868 static void free_pfn_range(u64 paddr, unsigned long size)
872 is_ram = pat_pagerange_is_ram(paddr, paddr + size);
874 free_memtype(paddr, paddr + size);
878 * track_pfn_copy is called when vma that is covering the pfnmap gets
879 * copied through copy_page_range().
881 * If the vma has a linear pfn mapping for the entire range, we get the prot
882 * from pte and reserve the entire vma range with single reserve_pfn_range call.
884 int track_pfn_copy(struct vm_area_struct *vma)
886 resource_size_t paddr;
888 unsigned long vma_size = vma->vm_end - vma->vm_start;
891 if (vma->vm_flags & VM_PAT) {
893 * reserve the whole chunk covered by vma. We need the
894 * starting address and protection from pte.
896 if (follow_phys(vma, vma->vm_start, 0, &prot, &paddr)) {
900 pgprot = __pgprot(prot);
901 return reserve_pfn_range(paddr, vma_size, &pgprot, 1);
908 * prot is passed in as a parameter for the new mapping. If the vma has a
909 * linear pfn mapping for the entire range reserve the entire vma range with
910 * single reserve_pfn_range call.
912 int track_pfn_remap(struct vm_area_struct *vma, pgprot_t *prot,
913 unsigned long pfn, unsigned long addr, unsigned long size)
915 resource_size_t paddr = (resource_size_t)pfn << PAGE_SHIFT;
916 enum page_cache_mode pcm;
918 /* reserve the whole chunk starting from paddr */
919 if (addr == vma->vm_start && size == (vma->vm_end - vma->vm_start)) {
922 ret = reserve_pfn_range(paddr, size, prot, 0);
924 vma->vm_flags |= VM_PAT;
932 * For anything smaller than the vma size we set prot based on the
935 pcm = lookup_memtype(paddr);
937 /* Check memtype for the remaining pages */
938 while (size > PAGE_SIZE) {
941 if (pcm != lookup_memtype(paddr))
945 *prot = __pgprot((pgprot_val(vma->vm_page_prot) & (~_PAGE_CACHE_MASK)) |
946 cachemode2protval(pcm));
951 int track_pfn_insert(struct vm_area_struct *vma, pgprot_t *prot,
954 enum page_cache_mode pcm;
959 /* Set prot based on lookup */
960 pcm = lookup_memtype((resource_size_t)pfn << PAGE_SHIFT);
961 *prot = __pgprot((pgprot_val(vma->vm_page_prot) & (~_PAGE_CACHE_MASK)) |
962 cachemode2protval(pcm));
968 * untrack_pfn is called while unmapping a pfnmap for a region.
969 * untrack can be called for a specific region indicated by pfn and size or
970 * can be for the entire vma (in which case pfn, size are zero).
972 void untrack_pfn(struct vm_area_struct *vma, unsigned long pfn,
975 resource_size_t paddr;
978 if (!(vma->vm_flags & VM_PAT))
981 /* free the chunk starting from pfn or the whole chunk */
982 paddr = (resource_size_t)pfn << PAGE_SHIFT;
983 if (!paddr && !size) {
984 if (follow_phys(vma, vma->vm_start, 0, &prot, &paddr)) {
989 size = vma->vm_end - vma->vm_start;
991 free_pfn_range(paddr, size);
992 vma->vm_flags &= ~VM_PAT;
995 pgprot_t pgprot_writecombine(pgprot_t prot)
997 return __pgprot(pgprot_val(prot) |
998 cachemode2protval(_PAGE_CACHE_MODE_WC));
1000 EXPORT_SYMBOL_GPL(pgprot_writecombine);
1002 pgprot_t pgprot_writethrough(pgprot_t prot)
1004 return __pgprot(pgprot_val(prot) |
1005 cachemode2protval(_PAGE_CACHE_MODE_WT));
1007 EXPORT_SYMBOL_GPL(pgprot_writethrough);
1009 #if defined(CONFIG_DEBUG_FS) && defined(CONFIG_X86_PAT)
1011 static struct memtype *memtype_get_idx(loff_t pos)
1013 struct memtype *print_entry;
1016 print_entry = kzalloc(sizeof(struct memtype), GFP_KERNEL);
1020 spin_lock(&memtype_lock);
1021 ret = rbt_memtype_copy_nth_element(print_entry, pos);
1022 spin_unlock(&memtype_lock);
1032 static void *memtype_seq_start(struct seq_file *seq, loff_t *pos)
1036 seq_puts(seq, "PAT memtype list:\n");
1039 return memtype_get_idx(*pos);
1042 static void *memtype_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1045 return memtype_get_idx(*pos);
1048 static void memtype_seq_stop(struct seq_file *seq, void *v)
1052 static int memtype_seq_show(struct seq_file *seq, void *v)
1054 struct memtype *print_entry = (struct memtype *)v;
1056 seq_printf(seq, "%s @ 0x%Lx-0x%Lx\n", cattr_name(print_entry->type),
1057 print_entry->start, print_entry->end);
1063 static const struct seq_operations memtype_seq_ops = {
1064 .start = memtype_seq_start,
1065 .next = memtype_seq_next,
1066 .stop = memtype_seq_stop,
1067 .show = memtype_seq_show,
1070 static int memtype_seq_open(struct inode *inode, struct file *file)
1072 return seq_open(file, &memtype_seq_ops);
1075 static const struct file_operations memtype_fops = {
1076 .open = memtype_seq_open,
1078 .llseek = seq_lseek,
1079 .release = seq_release,
1082 static int __init pat_memtype_list_init(void)
1084 if (pat_enabled()) {
1085 debugfs_create_file("pat_memtype_list", S_IRUSR,
1086 arch_debugfs_dir, NULL, &memtype_fops);
1091 late_initcall(pat_memtype_list_init);
1093 #endif /* CONFIG_DEBUG_FS && CONFIG_X86_PAT */