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/ioport.h>
14 #include <linux/kernel.h>
15 #include <linux/pfn_t.h>
16 #include <linux/slab.h>
19 #include <linux/rbtree.h>
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/api.h>
34 #include "pat_internal.h"
35 #include "mm_internal.h"
38 #define pr_fmt(fmt) "" fmt
40 static bool boot_cpu_done;
42 static int __read_mostly __pat_enabled = IS_ENABLED(CONFIG_X86_PAT);
43 static void init_cache_modes(void);
45 void pat_disable(const char *reason)
51 WARN_ONCE(1, "x86/PAT: PAT cannot be disabled after initialization\n");
56 pr_info("x86/PAT: %s\n", reason);
61 static int __init nopat(char *str)
63 pat_disable("PAT support disabled.");
66 early_param("nopat", nopat);
68 static bool __read_mostly __pat_initialized = false;
70 bool pat_enabled(void)
72 return __pat_initialized;
74 EXPORT_SYMBOL_GPL(pat_enabled);
78 static int __init pat_debug_setup(char *str)
83 __setup("debugpat", pat_debug_setup);
87 * X86 PAT uses page flags arch_1 and uncached together to keep track of
88 * memory type of pages that have backing page struct.
90 * X86 PAT supports 4 different memory types:
91 * - _PAGE_CACHE_MODE_WB
92 * - _PAGE_CACHE_MODE_WC
93 * - _PAGE_CACHE_MODE_UC_MINUS
94 * - _PAGE_CACHE_MODE_WT
96 * _PAGE_CACHE_MODE_WB is the default type.
100 #define _PGMT_WC (1UL << PG_arch_1)
101 #define _PGMT_UC_MINUS (1UL << PG_uncached)
102 #define _PGMT_WT (1UL << PG_uncached | 1UL << PG_arch_1)
103 #define _PGMT_MASK (1UL << PG_uncached | 1UL << PG_arch_1)
104 #define _PGMT_CLEAR_MASK (~_PGMT_MASK)
106 static inline enum page_cache_mode get_page_memtype(struct page *pg)
108 unsigned long pg_flags = pg->flags & _PGMT_MASK;
110 if (pg_flags == _PGMT_WB)
111 return _PAGE_CACHE_MODE_WB;
112 else if (pg_flags == _PGMT_WC)
113 return _PAGE_CACHE_MODE_WC;
114 else if (pg_flags == _PGMT_UC_MINUS)
115 return _PAGE_CACHE_MODE_UC_MINUS;
117 return _PAGE_CACHE_MODE_WT;
120 static inline void set_page_memtype(struct page *pg,
121 enum page_cache_mode memtype)
123 unsigned long memtype_flags;
124 unsigned long old_flags;
125 unsigned long new_flags;
128 case _PAGE_CACHE_MODE_WC:
129 memtype_flags = _PGMT_WC;
131 case _PAGE_CACHE_MODE_UC_MINUS:
132 memtype_flags = _PGMT_UC_MINUS;
134 case _PAGE_CACHE_MODE_WT:
135 memtype_flags = _PGMT_WT;
137 case _PAGE_CACHE_MODE_WB:
139 memtype_flags = _PGMT_WB;
144 old_flags = pg->flags;
145 new_flags = (old_flags & _PGMT_CLEAR_MASK) | memtype_flags;
146 } while (cmpxchg(&pg->flags, old_flags, new_flags) != old_flags);
149 static inline enum page_cache_mode get_page_memtype(struct page *pg)
153 static inline void set_page_memtype(struct page *pg,
154 enum page_cache_mode memtype)
160 PAT_UC = 0, /* uncached */
161 PAT_WC = 1, /* Write combining */
162 PAT_WT = 4, /* Write Through */
163 PAT_WP = 5, /* Write Protected */
164 PAT_WB = 6, /* Write Back (default) */
165 PAT_UC_MINUS = 7, /* UC, but can be overridden by MTRR */
168 #define CM(c) (_PAGE_CACHE_MODE_ ## c)
170 static enum page_cache_mode pat_get_cache_mode(unsigned pat_val, char *msg)
172 enum page_cache_mode cache;
176 case PAT_UC: cache = CM(UC); cache_mode = "UC "; break;
177 case PAT_WC: cache = CM(WC); cache_mode = "WC "; break;
178 case PAT_WT: cache = CM(WT); cache_mode = "WT "; break;
179 case PAT_WP: cache = CM(WP); cache_mode = "WP "; break;
180 case PAT_WB: cache = CM(WB); cache_mode = "WB "; break;
181 case PAT_UC_MINUS: cache = CM(UC_MINUS); cache_mode = "UC- "; break;
182 default: cache = CM(WB); cache_mode = "WB "; break;
185 memcpy(msg, cache_mode, 4);
193 * Update the cache mode to pgprot translation tables according to PAT
195 * Using lower indices is preferred, so we start with highest index.
197 static void __init_cache_modes(u64 pat)
199 enum page_cache_mode cache;
204 for (i = 7; i >= 0; i--) {
205 cache = pat_get_cache_mode((pat >> (i * 8)) & 7,
207 update_cache_mode_entry(i, cache);
209 pr_info("x86/PAT: Configuration [0-7]: %s\n", pat_msg);
212 #define PAT(x, y) ((u64)PAT_ ## y << ((x)*8))
214 static void pat_bsp_init(u64 pat)
218 if (!boot_cpu_has(X86_FEATURE_PAT)) {
219 pat_disable("PAT not supported by CPU.");
223 rdmsrl(MSR_IA32_CR_PAT, tmp_pat);
225 pat_disable("PAT MSR is 0, disabled.");
229 wrmsrl(MSR_IA32_CR_PAT, pat);
230 __pat_initialized = true;
232 __init_cache_modes(pat);
235 static void pat_ap_init(u64 pat)
237 if (!this_cpu_has(X86_FEATURE_PAT)) {
239 * If this happens we are on a secondary CPU, but switched to
240 * PAT on the boot CPU. We have no way to undo PAT.
242 panic("x86/PAT: PAT enabled, but not supported by secondary CPU\n");
245 wrmsrl(MSR_IA32_CR_PAT, pat);
248 static void init_cache_modes(void)
251 static int init_cm_done;
256 if (boot_cpu_has(X86_FEATURE_PAT)) {
258 * CPU supports PAT. Set PAT table to be consistent with
259 * PAT MSR. This case supports "nopat" boot option, and
260 * virtual machine environments which support PAT without
261 * MTRRs. In specific, Xen has unique setup to PAT MSR.
263 * If PAT MSR returns 0, it is considered invalid and emulates
266 rdmsrl(MSR_IA32_CR_PAT, pat);
271 * No PAT. Emulate the PAT table that corresponds to the two
272 * cache bits, PWT (Write Through) and PCD (Cache Disable).
273 * This setup is also the same as the BIOS default setup.
280 * 00 0 WB : _PAGE_CACHE_MODE_WB
281 * 01 1 WT : _PAGE_CACHE_MODE_WT
282 * 10 2 UC-: _PAGE_CACHE_MODE_UC_MINUS
283 * 11 3 UC : _PAGE_CACHE_MODE_UC
285 * NOTE: When WC or WP is used, it is redirected to UC- per
286 * the default setup in __cachemode2pte_tbl[].
288 pat = PAT(0, WB) | PAT(1, WT) | PAT(2, UC_MINUS) | PAT(3, UC) |
289 PAT(4, WB) | PAT(5, WT) | PAT(6, UC_MINUS) | PAT(7, UC);
292 __init_cache_modes(pat);
298 * pat_init - Initialize PAT MSR and PAT table
300 * This function initializes PAT MSR and PAT table with an OS-defined value
301 * to enable additional cache attributes, WC and WT.
303 * This function must be called on all CPUs using the specific sequence of
304 * operations defined in Intel SDM. mtrr_rendezvous_handler() provides this
310 struct cpuinfo_x86 *c = &boot_cpu_data;
312 if (!__pat_enabled) {
317 if ((c->x86_vendor == X86_VENDOR_INTEL) &&
318 (((c->x86 == 0x6) && (c->x86_model <= 0xd)) ||
319 ((c->x86 == 0xf) && (c->x86_model <= 0x6)))) {
321 * PAT support with the lower four entries. Intel Pentium 2,
322 * 3, M, and 4 are affected by PAT errata, which makes the
323 * upper four entries unusable. To be on the safe side, we don't
331 * 000 0 WB : _PAGE_CACHE_MODE_WB
332 * 001 1 WC : _PAGE_CACHE_MODE_WC
333 * 010 2 UC-: _PAGE_CACHE_MODE_UC_MINUS
334 * 011 3 UC : _PAGE_CACHE_MODE_UC
337 * NOTE: When WT or WP is used, it is redirected to UC- per
338 * the default setup in __cachemode2pte_tbl[].
340 pat = PAT(0, WB) | PAT(1, WC) | PAT(2, UC_MINUS) | PAT(3, UC) |
341 PAT(4, WB) | PAT(5, WC) | PAT(6, UC_MINUS) | PAT(7, UC);
344 * Full PAT support. We put WT in slot 7 to improve
345 * robustness in the presence of errata that might cause
346 * the high PAT bit to be ignored. This way, a buggy slot 7
347 * access will hit slot 3, and slot 3 is UC, so at worst
348 * we lose performance without causing a correctness issue.
349 * Pentium 4 erratum N46 is an example for such an erratum,
350 * although we try not to use PAT at all on affected CPUs.
357 * 000 0 WB : _PAGE_CACHE_MODE_WB
358 * 001 1 WC : _PAGE_CACHE_MODE_WC
359 * 010 2 UC-: _PAGE_CACHE_MODE_UC_MINUS
360 * 011 3 UC : _PAGE_CACHE_MODE_UC
361 * 100 4 WB : Reserved
362 * 101 5 WC : Reserved
363 * 110 6 UC-: Reserved
364 * 111 7 WT : _PAGE_CACHE_MODE_WT
366 * The reserved slots are unused, but mapped to their
367 * corresponding types in the presence of PAT errata.
369 pat = PAT(0, WB) | PAT(1, WC) | PAT(2, UC_MINUS) | PAT(3, UC) |
370 PAT(4, WB) | PAT(5, WC) | PAT(6, UC_MINUS) | PAT(7, WT);
373 if (!boot_cpu_done) {
375 boot_cpu_done = true;
383 static DEFINE_SPINLOCK(memtype_lock); /* protects memtype accesses */
386 * Does intersection of PAT memory type and MTRR memory type and returns
387 * the resulting memory type as PAT understands it.
388 * (Type in pat and mtrr will not have same value)
389 * The intersection is based on "Effective Memory Type" tables in IA-32
392 static unsigned long pat_x_mtrr_type(u64 start, u64 end,
393 enum page_cache_mode req_type)
396 * Look for MTRR hint to get the effective type in case where PAT
399 if (req_type == _PAGE_CACHE_MODE_WB) {
400 u8 mtrr_type, uniform;
402 mtrr_type = mtrr_type_lookup(start, end, &uniform);
403 if (mtrr_type != MTRR_TYPE_WRBACK)
404 return _PAGE_CACHE_MODE_UC_MINUS;
406 return _PAGE_CACHE_MODE_WB;
412 struct pagerange_state {
413 unsigned long cur_pfn;
419 pagerange_is_ram_callback(unsigned long initial_pfn, unsigned long total_nr_pages, void *arg)
421 struct pagerange_state *state = arg;
423 state->not_ram |= initial_pfn > state->cur_pfn;
424 state->ram |= total_nr_pages > 0;
425 state->cur_pfn = initial_pfn + total_nr_pages;
427 return state->ram && state->not_ram;
430 static int pat_pagerange_is_ram(resource_size_t start, resource_size_t end)
433 unsigned long start_pfn = start >> PAGE_SHIFT;
434 unsigned long end_pfn = (end + PAGE_SIZE - 1) >> PAGE_SHIFT;
435 struct pagerange_state state = {start_pfn, 0, 0};
438 * For legacy reasons, physical address range in the legacy ISA
439 * region is tracked as non-RAM. This will allow users of
440 * /dev/mem to map portions of legacy ISA region, even when
441 * some of those portions are listed(or not even listed) with
442 * different e820 types(RAM/reserved/..)
444 if (start_pfn < ISA_END_ADDRESS >> PAGE_SHIFT)
445 start_pfn = ISA_END_ADDRESS >> PAGE_SHIFT;
447 if (start_pfn < end_pfn) {
448 ret = walk_system_ram_range(start_pfn, end_pfn - start_pfn,
449 &state, pagerange_is_ram_callback);
452 return (ret > 0) ? -1 : (state.ram ? 1 : 0);
456 * For RAM pages, we use page flags to mark the pages with appropriate type.
457 * The page flags are limited to four types, WB (default), WC, WT and UC-.
458 * WP request fails with -EINVAL, and UC gets redirected to UC-. Setting
459 * a new memory type is only allowed for a page mapped with the default WB
462 * Here we do two passes:
463 * - Find the memtype of all the pages in the range, look for any conflicts.
464 * - In case of no conflicts, set the new memtype for pages in the range.
466 static int reserve_ram_pages_type(u64 start, u64 end,
467 enum page_cache_mode req_type,
468 enum page_cache_mode *new_type)
473 if (req_type == _PAGE_CACHE_MODE_WP) {
475 *new_type = _PAGE_CACHE_MODE_UC_MINUS;
479 if (req_type == _PAGE_CACHE_MODE_UC) {
480 /* We do not support strong UC */
482 req_type = _PAGE_CACHE_MODE_UC_MINUS;
485 for (pfn = (start >> PAGE_SHIFT); pfn < (end >> PAGE_SHIFT); ++pfn) {
486 enum page_cache_mode type;
488 page = pfn_to_page(pfn);
489 type = get_page_memtype(page);
490 if (type != _PAGE_CACHE_MODE_WB) {
491 pr_info("x86/PAT: reserve_ram_pages_type failed [mem %#010Lx-%#010Lx], track 0x%x, req 0x%x\n",
492 start, end - 1, type, req_type);
501 *new_type = req_type;
503 for (pfn = (start >> PAGE_SHIFT); pfn < (end >> PAGE_SHIFT); ++pfn) {
504 page = pfn_to_page(pfn);
505 set_page_memtype(page, req_type);
510 static int free_ram_pages_type(u64 start, u64 end)
515 for (pfn = (start >> PAGE_SHIFT); pfn < (end >> PAGE_SHIFT); ++pfn) {
516 page = pfn_to_page(pfn);
517 set_page_memtype(page, _PAGE_CACHE_MODE_WB);
523 * req_type typically has one of the:
524 * - _PAGE_CACHE_MODE_WB
525 * - _PAGE_CACHE_MODE_WC
526 * - _PAGE_CACHE_MODE_UC_MINUS
527 * - _PAGE_CACHE_MODE_UC
528 * - _PAGE_CACHE_MODE_WT
530 * If new_type is NULL, function will return an error if it cannot reserve the
531 * region with req_type. If new_type is non-NULL, function will return
532 * available type in new_type in case of no error. In case of any error
533 * it will return a negative return value.
535 int reserve_memtype(u64 start, u64 end, enum page_cache_mode req_type,
536 enum page_cache_mode *new_type)
539 enum page_cache_mode actual_type;
543 BUG_ON(start >= end); /* end is exclusive */
545 if (!pat_enabled()) {
546 /* This is identical to page table setting without PAT */
548 *new_type = req_type;
552 /* Low ISA region is always mapped WB in page table. No need to track */
553 if (x86_platform.is_untracked_pat_range(start, end)) {
555 *new_type = _PAGE_CACHE_MODE_WB;
560 * Call mtrr_lookup to get the type hint. This is an
561 * optimization for /dev/mem mmap'ers into WB memory (BIOS
562 * tools and ACPI tools). Use WB request for WB memory and use
563 * UC_MINUS otherwise.
565 actual_type = pat_x_mtrr_type(start, end, req_type);
568 *new_type = actual_type;
570 is_range_ram = pat_pagerange_is_ram(start, end);
571 if (is_range_ram == 1) {
573 err = reserve_ram_pages_type(start, end, req_type, new_type);
576 } else if (is_range_ram < 0) {
580 new = kzalloc(sizeof(struct memtype), GFP_KERNEL);
586 new->type = actual_type;
588 spin_lock(&memtype_lock);
590 err = rbt_memtype_check_insert(new, new_type);
592 pr_info("x86/PAT: reserve_memtype failed [mem %#010Lx-%#010Lx], track %s, req %s\n",
594 cattr_name(new->type), cattr_name(req_type));
596 spin_unlock(&memtype_lock);
601 spin_unlock(&memtype_lock);
603 dprintk("reserve_memtype added [mem %#010Lx-%#010Lx], track %s, req %s, ret %s\n",
604 start, end - 1, cattr_name(new->type), cattr_name(req_type),
605 new_type ? cattr_name(*new_type) : "-");
610 int free_memtype(u64 start, u64 end)
614 struct memtype *entry;
619 /* Low ISA region is always mapped WB. No need to track */
620 if (x86_platform.is_untracked_pat_range(start, end))
623 is_range_ram = pat_pagerange_is_ram(start, end);
624 if (is_range_ram == 1) {
626 err = free_ram_pages_type(start, end);
629 } else if (is_range_ram < 0) {
633 spin_lock(&memtype_lock);
634 entry = rbt_memtype_erase(start, end);
635 spin_unlock(&memtype_lock);
638 pr_info("x86/PAT: %s:%d freeing invalid memtype [mem %#010Lx-%#010Lx]\n",
639 current->comm, current->pid, start, end - 1);
645 dprintk("free_memtype request [mem %#010Lx-%#010Lx]\n", start, end - 1);
652 * lookup_memtype - Looksup the memory type for a physical address
653 * @paddr: physical address of which memory type needs to be looked up
655 * Only to be called when PAT is enabled
657 * Returns _PAGE_CACHE_MODE_WB, _PAGE_CACHE_MODE_WC, _PAGE_CACHE_MODE_UC_MINUS
658 * or _PAGE_CACHE_MODE_WT.
660 static enum page_cache_mode lookup_memtype(u64 paddr)
662 enum page_cache_mode rettype = _PAGE_CACHE_MODE_WB;
663 struct memtype *entry;
665 if (x86_platform.is_untracked_pat_range(paddr, paddr + PAGE_SIZE))
668 if (pat_pagerange_is_ram(paddr, paddr + PAGE_SIZE)) {
671 page = pfn_to_page(paddr >> PAGE_SHIFT);
672 return get_page_memtype(page);
675 spin_lock(&memtype_lock);
677 entry = rbt_memtype_lookup(paddr);
679 rettype = entry->type;
681 rettype = _PAGE_CACHE_MODE_UC_MINUS;
683 spin_unlock(&memtype_lock);
688 * io_reserve_memtype - Request a memory type mapping for a region of memory
689 * @start: start (physical address) of the region
690 * @end: end (physical address) of the region
691 * @type: A pointer to memtype, with requested type. On success, requested
692 * or any other compatible type that was available for the region is returned
694 * On success, returns 0
695 * On failure, returns non-zero
697 int io_reserve_memtype(resource_size_t start, resource_size_t end,
698 enum page_cache_mode *type)
700 resource_size_t size = end - start;
701 enum page_cache_mode req_type = *type;
702 enum page_cache_mode new_type;
705 WARN_ON_ONCE(iomem_map_sanity_check(start, size));
707 ret = reserve_memtype(start, end, req_type, &new_type);
711 if (!is_new_memtype_allowed(start, size, req_type, new_type))
714 if (kernel_map_sync_memtype(start, size, new_type) < 0)
721 free_memtype(start, end);
728 * io_free_memtype - Release a memory type mapping for a region of memory
729 * @start: start (physical address) of the region
730 * @end: end (physical address) of the region
732 void io_free_memtype(resource_size_t start, resource_size_t end)
734 free_memtype(start, end);
737 int arch_io_reserve_memtype_wc(resource_size_t start, resource_size_t size)
739 enum page_cache_mode type = _PAGE_CACHE_MODE_WC;
741 return io_reserve_memtype(start, start + size, &type);
743 EXPORT_SYMBOL(arch_io_reserve_memtype_wc);
745 void arch_io_free_memtype_wc(resource_size_t start, resource_size_t size)
747 io_free_memtype(start, start + size);
749 EXPORT_SYMBOL(arch_io_free_memtype_wc);
751 pgprot_t phys_mem_access_prot(struct file *file, unsigned long pfn,
752 unsigned long size, pgprot_t vma_prot)
757 #ifdef CONFIG_STRICT_DEVMEM
758 /* This check is done in drivers/char/mem.c in case of STRICT_DEVMEM */
759 static inline int range_is_allowed(unsigned long pfn, unsigned long size)
764 /* This check is needed to avoid cache aliasing when PAT is enabled */
765 static inline int range_is_allowed(unsigned long pfn, unsigned long size)
767 u64 from = ((u64)pfn) << PAGE_SHIFT;
768 u64 to = from + size;
774 while (cursor < to) {
775 if (!devmem_is_allowed(pfn))
782 #endif /* CONFIG_STRICT_DEVMEM */
784 int phys_mem_access_prot_allowed(struct file *file, unsigned long pfn,
785 unsigned long size, pgprot_t *vma_prot)
787 enum page_cache_mode pcm = _PAGE_CACHE_MODE_WB;
789 if (!range_is_allowed(pfn, size))
792 if (file->f_flags & O_DSYNC)
793 pcm = _PAGE_CACHE_MODE_UC_MINUS;
795 *vma_prot = __pgprot((pgprot_val(*vma_prot) & ~_PAGE_CACHE_MASK) |
796 cachemode2protval(pcm));
801 * Change the memory type for the physial address range in kernel identity
802 * mapping space if that range is a part of identity map.
804 int kernel_map_sync_memtype(u64 base, unsigned long size,
805 enum page_cache_mode pcm)
809 if (base > __pa(high_memory-1))
813 * some areas in the middle of the kernel identity range
814 * are not mapped, like the PCI space.
816 if (!page_is_ram(base >> PAGE_SHIFT))
819 id_sz = (__pa(high_memory-1) <= base + size) ?
820 __pa(high_memory) - base :
823 if (ioremap_change_attr((unsigned long)__va(base), id_sz, pcm) < 0) {
824 pr_info("x86/PAT: %s:%d ioremap_change_attr failed %s for [mem %#010Lx-%#010Lx]\n",
825 current->comm, current->pid,
827 base, (unsigned long long)(base + size-1));
834 * Internal interface to reserve a range of physical memory with prot.
835 * Reserved non RAM regions only and after successful reserve_memtype,
836 * this func also keeps identity mapping (if any) in sync with this new prot.
838 static int reserve_pfn_range(u64 paddr, unsigned long size, pgprot_t *vma_prot,
843 enum page_cache_mode want_pcm = pgprot2cachemode(*vma_prot);
844 enum page_cache_mode pcm = want_pcm;
846 is_ram = pat_pagerange_is_ram(paddr, paddr + size);
849 * reserve_pfn_range() for RAM pages. We do not refcount to keep
850 * track of number of mappings of RAM pages. We can assert that
851 * the type requested matches the type of first page in the range.
857 pcm = lookup_memtype(paddr);
858 if (want_pcm != pcm) {
859 pr_warn("x86/PAT: %s:%d map pfn RAM range req %s for [mem %#010Lx-%#010Lx], got %s\n",
860 current->comm, current->pid,
861 cattr_name(want_pcm),
862 (unsigned long long)paddr,
863 (unsigned long long)(paddr + size - 1),
865 *vma_prot = __pgprot((pgprot_val(*vma_prot) &
866 (~_PAGE_CACHE_MASK)) |
867 cachemode2protval(pcm));
872 ret = reserve_memtype(paddr, paddr + size, want_pcm, &pcm);
876 if (pcm != want_pcm) {
878 !is_new_memtype_allowed(paddr, size, want_pcm, pcm)) {
879 free_memtype(paddr, paddr + size);
880 pr_err("x86/PAT: %s:%d map pfn expected mapping type %s for [mem %#010Lx-%#010Lx], got %s\n",
881 current->comm, current->pid,
882 cattr_name(want_pcm),
883 (unsigned long long)paddr,
884 (unsigned long long)(paddr + size - 1),
889 * We allow returning different type than the one requested in
892 *vma_prot = __pgprot((pgprot_val(*vma_prot) &
893 (~_PAGE_CACHE_MASK)) |
894 cachemode2protval(pcm));
897 if (kernel_map_sync_memtype(paddr, size, pcm) < 0) {
898 free_memtype(paddr, paddr + size);
905 * Internal interface to free a range of physical memory.
906 * Frees non RAM regions only.
908 static void free_pfn_range(u64 paddr, unsigned long size)
912 is_ram = pat_pagerange_is_ram(paddr, paddr + size);
914 free_memtype(paddr, paddr + size);
918 * track_pfn_copy is called when vma that is covering the pfnmap gets
919 * copied through copy_page_range().
921 * If the vma has a linear pfn mapping for the entire range, we get the prot
922 * from pte and reserve the entire vma range with single reserve_pfn_range call.
924 int track_pfn_copy(struct vm_area_struct *vma)
926 resource_size_t paddr;
928 unsigned long vma_size = vma->vm_end - vma->vm_start;
931 if (vma->vm_flags & VM_PAT) {
933 * reserve the whole chunk covered by vma. We need the
934 * starting address and protection from pte.
936 if (follow_phys(vma, vma->vm_start, 0, &prot, &paddr)) {
940 pgprot = __pgprot(prot);
941 return reserve_pfn_range(paddr, vma_size, &pgprot, 1);
948 * prot is passed in as a parameter for the new mapping. If the vma has
949 * a linear pfn mapping for the entire range, or no vma is provided,
950 * reserve the entire pfn + size range with single reserve_pfn_range
953 int track_pfn_remap(struct vm_area_struct *vma, pgprot_t *prot,
954 unsigned long pfn, unsigned long addr, unsigned long size)
956 resource_size_t paddr = (resource_size_t)pfn << PAGE_SHIFT;
957 enum page_cache_mode pcm;
959 /* reserve the whole chunk starting from paddr */
960 if (!vma || (addr == vma->vm_start
961 && size == (vma->vm_end - vma->vm_start))) {
964 ret = reserve_pfn_range(paddr, size, prot, 0);
966 vma->vm_flags |= VM_PAT;
974 * For anything smaller than the vma size we set prot based on the
977 pcm = lookup_memtype(paddr);
979 /* Check memtype for the remaining pages */
980 while (size > PAGE_SIZE) {
983 if (pcm != lookup_memtype(paddr))
987 *prot = __pgprot((pgprot_val(*prot) & (~_PAGE_CACHE_MASK)) |
988 cachemode2protval(pcm));
993 void track_pfn_insert(struct vm_area_struct *vma, pgprot_t *prot, pfn_t pfn)
995 enum page_cache_mode pcm;
1000 /* Set prot based on lookup */
1001 pcm = lookup_memtype(pfn_t_to_phys(pfn));
1002 *prot = __pgprot((pgprot_val(*prot) & (~_PAGE_CACHE_MASK)) |
1003 cachemode2protval(pcm));
1007 * untrack_pfn is called while unmapping a pfnmap for a region.
1008 * untrack can be called for a specific region indicated by pfn and size or
1009 * can be for the entire vma (in which case pfn, size are zero).
1011 void untrack_pfn(struct vm_area_struct *vma, unsigned long pfn,
1014 resource_size_t paddr;
1017 if (vma && !(vma->vm_flags & VM_PAT))
1020 /* free the chunk starting from pfn or the whole chunk */
1021 paddr = (resource_size_t)pfn << PAGE_SHIFT;
1022 if (!paddr && !size) {
1023 if (follow_phys(vma, vma->vm_start, 0, &prot, &paddr)) {
1028 size = vma->vm_end - vma->vm_start;
1030 free_pfn_range(paddr, size);
1032 vma->vm_flags &= ~VM_PAT;
1036 * untrack_pfn_moved is called, while mremapping a pfnmap for a new region,
1037 * with the old vma after its pfnmap page table has been removed. The new
1038 * vma has a new pfnmap to the same pfn & cache type with VM_PAT set.
1040 void untrack_pfn_moved(struct vm_area_struct *vma)
1042 vma->vm_flags &= ~VM_PAT;
1045 pgprot_t pgprot_writecombine(pgprot_t prot)
1047 return __pgprot(pgprot_val(prot) |
1048 cachemode2protval(_PAGE_CACHE_MODE_WC));
1050 EXPORT_SYMBOL_GPL(pgprot_writecombine);
1052 pgprot_t pgprot_writethrough(pgprot_t prot)
1054 return __pgprot(pgprot_val(prot) |
1055 cachemode2protval(_PAGE_CACHE_MODE_WT));
1057 EXPORT_SYMBOL_GPL(pgprot_writethrough);
1059 #if defined(CONFIG_DEBUG_FS) && defined(CONFIG_X86_PAT)
1061 static struct memtype *memtype_get_idx(loff_t pos)
1063 struct memtype *print_entry;
1066 print_entry = kzalloc(sizeof(struct memtype), GFP_KERNEL);
1070 spin_lock(&memtype_lock);
1071 ret = rbt_memtype_copy_nth_element(print_entry, pos);
1072 spin_unlock(&memtype_lock);
1082 static void *memtype_seq_start(struct seq_file *seq, loff_t *pos)
1086 seq_puts(seq, "PAT memtype list:\n");
1089 return memtype_get_idx(*pos);
1092 static void *memtype_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1095 return memtype_get_idx(*pos);
1098 static void memtype_seq_stop(struct seq_file *seq, void *v)
1102 static int memtype_seq_show(struct seq_file *seq, void *v)
1104 struct memtype *print_entry = (struct memtype *)v;
1106 seq_printf(seq, "%s @ 0x%Lx-0x%Lx\n", cattr_name(print_entry->type),
1107 print_entry->start, print_entry->end);
1113 static const struct seq_operations memtype_seq_ops = {
1114 .start = memtype_seq_start,
1115 .next = memtype_seq_next,
1116 .stop = memtype_seq_stop,
1117 .show = memtype_seq_show,
1120 static int memtype_seq_open(struct inode *inode, struct file *file)
1122 return seq_open(file, &memtype_seq_ops);
1125 static const struct file_operations memtype_fops = {
1126 .open = memtype_seq_open,
1128 .llseek = seq_lseek,
1129 .release = seq_release,
1132 static int __init pat_memtype_list_init(void)
1134 if (pat_enabled()) {
1135 debugfs_create_file("pat_memtype_list", S_IRUSR,
1136 arch_debugfs_dir, NULL, &memtype_fops);
1141 late_initcall(pat_memtype_list_init);
1143 #endif /* CONFIG_DEBUG_FS && CONFIG_X86_PAT */