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 __read_mostly boot_cpu_done;
41 static bool __read_mostly pat_disabled = !IS_ENABLED(CONFIG_X86_PAT);
42 static bool __read_mostly pat_initialized;
43 static bool __read_mostly init_cm_done;
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);
59 static int __init nopat(char *str)
61 pat_disable("PAT support disabled.");
64 early_param("nopat", nopat);
66 bool pat_enabled(void)
68 return pat_initialized;
70 EXPORT_SYMBOL_GPL(pat_enabled);
74 static int __init pat_debug_setup(char *str)
79 __setup("debugpat", pat_debug_setup);
83 * X86 PAT uses page flags arch_1 and uncached together to keep track of
84 * memory type of pages that have backing page struct.
86 * X86 PAT supports 4 different memory types:
87 * - _PAGE_CACHE_MODE_WB
88 * - _PAGE_CACHE_MODE_WC
89 * - _PAGE_CACHE_MODE_UC_MINUS
90 * - _PAGE_CACHE_MODE_WT
92 * _PAGE_CACHE_MODE_WB is the default type.
96 #define _PGMT_WC (1UL << PG_arch_1)
97 #define _PGMT_UC_MINUS (1UL << PG_uncached)
98 #define _PGMT_WT (1UL << PG_uncached | 1UL << PG_arch_1)
99 #define _PGMT_MASK (1UL << PG_uncached | 1UL << PG_arch_1)
100 #define _PGMT_CLEAR_MASK (~_PGMT_MASK)
102 static inline enum page_cache_mode get_page_memtype(struct page *pg)
104 unsigned long pg_flags = pg->flags & _PGMT_MASK;
106 if (pg_flags == _PGMT_WB)
107 return _PAGE_CACHE_MODE_WB;
108 else if (pg_flags == _PGMT_WC)
109 return _PAGE_CACHE_MODE_WC;
110 else if (pg_flags == _PGMT_UC_MINUS)
111 return _PAGE_CACHE_MODE_UC_MINUS;
113 return _PAGE_CACHE_MODE_WT;
116 static inline void set_page_memtype(struct page *pg,
117 enum page_cache_mode memtype)
119 unsigned long memtype_flags;
120 unsigned long old_flags;
121 unsigned long new_flags;
124 case _PAGE_CACHE_MODE_WC:
125 memtype_flags = _PGMT_WC;
127 case _PAGE_CACHE_MODE_UC_MINUS:
128 memtype_flags = _PGMT_UC_MINUS;
130 case _PAGE_CACHE_MODE_WT:
131 memtype_flags = _PGMT_WT;
133 case _PAGE_CACHE_MODE_WB:
135 memtype_flags = _PGMT_WB;
140 old_flags = pg->flags;
141 new_flags = (old_flags & _PGMT_CLEAR_MASK) | memtype_flags;
142 } while (cmpxchg(&pg->flags, old_flags, new_flags) != old_flags);
145 static inline enum page_cache_mode get_page_memtype(struct page *pg)
149 static inline void set_page_memtype(struct page *pg,
150 enum page_cache_mode memtype)
156 PAT_UC = 0, /* uncached */
157 PAT_WC = 1, /* Write combining */
158 PAT_WT = 4, /* Write Through */
159 PAT_WP = 5, /* Write Protected */
160 PAT_WB = 6, /* Write Back (default) */
161 PAT_UC_MINUS = 7, /* UC, but can be overridden by MTRR */
164 #define CM(c) (_PAGE_CACHE_MODE_ ## c)
166 static enum page_cache_mode pat_get_cache_mode(unsigned pat_val, char *msg)
168 enum page_cache_mode cache;
172 case PAT_UC: cache = CM(UC); cache_mode = "UC "; break;
173 case PAT_WC: cache = CM(WC); cache_mode = "WC "; break;
174 case PAT_WT: cache = CM(WT); cache_mode = "WT "; break;
175 case PAT_WP: cache = CM(WP); cache_mode = "WP "; break;
176 case PAT_WB: cache = CM(WB); cache_mode = "WB "; break;
177 case PAT_UC_MINUS: cache = CM(UC_MINUS); cache_mode = "UC- "; break;
178 default: cache = CM(WB); cache_mode = "WB "; break;
181 memcpy(msg, cache_mode, 4);
189 * Update the cache mode to pgprot translation tables according to PAT
191 * Using lower indices is preferred, so we start with highest index.
193 static void __init_cache_modes(u64 pat)
195 enum page_cache_mode cache;
200 for (i = 7; i >= 0; i--) {
201 cache = pat_get_cache_mode((pat >> (i * 8)) & 7,
203 update_cache_mode_entry(i, cache);
205 pr_info("x86/PAT: Configuration [0-7]: %s\n", pat_msg);
210 #define PAT(x, y) ((u64)PAT_ ## y << ((x)*8))
212 static void pat_bsp_init(u64 pat)
216 if (!boot_cpu_has(X86_FEATURE_PAT)) {
217 pat_disable("PAT not supported by CPU.");
221 rdmsrl(MSR_IA32_CR_PAT, tmp_pat);
223 pat_disable("PAT MSR is 0, disabled.");
227 wrmsrl(MSR_IA32_CR_PAT, pat);
228 pat_initialized = true;
230 __init_cache_modes(pat);
233 static void pat_ap_init(u64 pat)
235 if (!boot_cpu_has(X86_FEATURE_PAT)) {
237 * If this happens we are on a secondary CPU, but switched to
238 * PAT on the boot CPU. We have no way to undo PAT.
240 panic("x86/PAT: PAT enabled, but not supported by secondary CPU\n");
243 wrmsrl(MSR_IA32_CR_PAT, pat);
246 void init_cache_modes(void)
253 if (boot_cpu_has(X86_FEATURE_PAT)) {
255 * CPU supports PAT. Set PAT table to be consistent with
256 * PAT MSR. This case supports "nopat" boot option, and
257 * virtual machine environments which support PAT without
258 * MTRRs. In specific, Xen has unique setup to PAT MSR.
260 * If PAT MSR returns 0, it is considered invalid and emulates
263 rdmsrl(MSR_IA32_CR_PAT, pat);
268 * No PAT. Emulate the PAT table that corresponds to the two
269 * cache bits, PWT (Write Through) and PCD (Cache Disable).
270 * This setup is also the same as the BIOS default setup.
277 * 00 0 WB : _PAGE_CACHE_MODE_WB
278 * 01 1 WT : _PAGE_CACHE_MODE_WT
279 * 10 2 UC-: _PAGE_CACHE_MODE_UC_MINUS
280 * 11 3 UC : _PAGE_CACHE_MODE_UC
282 * NOTE: When WC or WP is used, it is redirected to UC- per
283 * the default setup in __cachemode2pte_tbl[].
285 pat = PAT(0, WB) | PAT(1, WT) | PAT(2, UC_MINUS) | PAT(3, UC) |
286 PAT(4, WB) | PAT(5, WT) | PAT(6, UC_MINUS) | PAT(7, UC);
289 __init_cache_modes(pat);
293 * pat_init - Initialize PAT MSR and PAT table
295 * This function initializes PAT MSR and PAT table with an OS-defined value
296 * to enable additional cache attributes, WC, WT and WP.
298 * This function must be called on all CPUs using the specific sequence of
299 * operations defined in Intel SDM. mtrr_rendezvous_handler() provides this
305 struct cpuinfo_x86 *c = &boot_cpu_data;
310 if ((c->x86_vendor == X86_VENDOR_INTEL) &&
311 (((c->x86 == 0x6) && (c->x86_model <= 0xd)) ||
312 ((c->x86 == 0xf) && (c->x86_model <= 0x6)))) {
314 * PAT support with the lower four entries. Intel Pentium 2,
315 * 3, M, and 4 are affected by PAT errata, which makes the
316 * upper four entries unusable. To be on the safe side, we don't
324 * 000 0 WB : _PAGE_CACHE_MODE_WB
325 * 001 1 WC : _PAGE_CACHE_MODE_WC
326 * 010 2 UC-: _PAGE_CACHE_MODE_UC_MINUS
327 * 011 3 UC : _PAGE_CACHE_MODE_UC
330 * NOTE: When WT or WP is used, it is redirected to UC- per
331 * the default setup in __cachemode2pte_tbl[].
333 pat = PAT(0, WB) | PAT(1, WC) | PAT(2, UC_MINUS) | PAT(3, UC) |
334 PAT(4, WB) | PAT(5, WC) | PAT(6, UC_MINUS) | PAT(7, UC);
337 * Full PAT support. We put WT in slot 7 to improve
338 * robustness in the presence of errata that might cause
339 * the high PAT bit to be ignored. This way, a buggy slot 7
340 * access will hit slot 3, and slot 3 is UC, so at worst
341 * we lose performance without causing a correctness issue.
342 * Pentium 4 erratum N46 is an example for such an erratum,
343 * although we try not to use PAT at all on affected CPUs.
350 * 000 0 WB : _PAGE_CACHE_MODE_WB
351 * 001 1 WC : _PAGE_CACHE_MODE_WC
352 * 010 2 UC-: _PAGE_CACHE_MODE_UC_MINUS
353 * 011 3 UC : _PAGE_CACHE_MODE_UC
354 * 100 4 WB : Reserved
355 * 101 5 WP : _PAGE_CACHE_MODE_WP
356 * 110 6 UC-: Reserved
357 * 111 7 WT : _PAGE_CACHE_MODE_WT
359 * The reserved slots are unused, but mapped to their
360 * corresponding types in the presence of PAT errata.
362 pat = PAT(0, WB) | PAT(1, WC) | PAT(2, UC_MINUS) | PAT(3, UC) |
363 PAT(4, WB) | PAT(5, WP) | PAT(6, UC_MINUS) | PAT(7, WT);
366 if (!boot_cpu_done) {
368 boot_cpu_done = true;
376 static DEFINE_SPINLOCK(memtype_lock); /* protects memtype accesses */
379 * Does intersection of PAT memory type and MTRR memory type and returns
380 * the resulting memory type as PAT understands it.
381 * (Type in pat and mtrr will not have same value)
382 * The intersection is based on "Effective Memory Type" tables in IA-32
385 static unsigned long pat_x_mtrr_type(u64 start, u64 end,
386 enum page_cache_mode req_type)
389 * Look for MTRR hint to get the effective type in case where PAT
392 if (req_type == _PAGE_CACHE_MODE_WB) {
393 u8 mtrr_type, uniform;
395 mtrr_type = mtrr_type_lookup(start, end, &uniform);
396 if (mtrr_type != MTRR_TYPE_WRBACK)
397 return _PAGE_CACHE_MODE_UC_MINUS;
399 return _PAGE_CACHE_MODE_WB;
405 struct pagerange_state {
406 unsigned long cur_pfn;
412 pagerange_is_ram_callback(unsigned long initial_pfn, unsigned long total_nr_pages, void *arg)
414 struct pagerange_state *state = arg;
416 state->not_ram |= initial_pfn > state->cur_pfn;
417 state->ram |= total_nr_pages > 0;
418 state->cur_pfn = initial_pfn + total_nr_pages;
420 return state->ram && state->not_ram;
423 static int pat_pagerange_is_ram(resource_size_t start, resource_size_t end)
426 unsigned long start_pfn = start >> PAGE_SHIFT;
427 unsigned long end_pfn = (end + PAGE_SIZE - 1) >> PAGE_SHIFT;
428 struct pagerange_state state = {start_pfn, 0, 0};
431 * For legacy reasons, physical address range in the legacy ISA
432 * region is tracked as non-RAM. This will allow users of
433 * /dev/mem to map portions of legacy ISA region, even when
434 * some of those portions are listed(or not even listed) with
435 * different e820 types(RAM/reserved/..)
437 if (start_pfn < ISA_END_ADDRESS >> PAGE_SHIFT)
438 start_pfn = ISA_END_ADDRESS >> PAGE_SHIFT;
440 if (start_pfn < end_pfn) {
441 ret = walk_system_ram_range(start_pfn, end_pfn - start_pfn,
442 &state, pagerange_is_ram_callback);
445 return (ret > 0) ? -1 : (state.ram ? 1 : 0);
449 * For RAM pages, we use page flags to mark the pages with appropriate type.
450 * The page flags are limited to four types, WB (default), WC, WT and UC-.
451 * WP request fails with -EINVAL, and UC gets redirected to UC-. Setting
452 * a new memory type is only allowed for a page mapped with the default WB
455 * Here we do two passes:
456 * - Find the memtype of all the pages in the range, look for any conflicts.
457 * - In case of no conflicts, set the new memtype for pages in the range.
459 static int reserve_ram_pages_type(u64 start, u64 end,
460 enum page_cache_mode req_type,
461 enum page_cache_mode *new_type)
466 if (req_type == _PAGE_CACHE_MODE_WP) {
468 *new_type = _PAGE_CACHE_MODE_UC_MINUS;
472 if (req_type == _PAGE_CACHE_MODE_UC) {
473 /* We do not support strong UC */
475 req_type = _PAGE_CACHE_MODE_UC_MINUS;
478 for (pfn = (start >> PAGE_SHIFT); pfn < (end >> PAGE_SHIFT); ++pfn) {
479 enum page_cache_mode type;
481 page = pfn_to_page(pfn);
482 type = get_page_memtype(page);
483 if (type != _PAGE_CACHE_MODE_WB) {
484 pr_info("x86/PAT: reserve_ram_pages_type failed [mem %#010Lx-%#010Lx], track 0x%x, req 0x%x\n",
485 start, end - 1, type, req_type);
494 *new_type = req_type;
496 for (pfn = (start >> PAGE_SHIFT); pfn < (end >> PAGE_SHIFT); ++pfn) {
497 page = pfn_to_page(pfn);
498 set_page_memtype(page, req_type);
503 static int free_ram_pages_type(u64 start, u64 end)
508 for (pfn = (start >> PAGE_SHIFT); pfn < (end >> PAGE_SHIFT); ++pfn) {
509 page = pfn_to_page(pfn);
510 set_page_memtype(page, _PAGE_CACHE_MODE_WB);
516 * req_type typically has one of the:
517 * - _PAGE_CACHE_MODE_WB
518 * - _PAGE_CACHE_MODE_WC
519 * - _PAGE_CACHE_MODE_UC_MINUS
520 * - _PAGE_CACHE_MODE_UC
521 * - _PAGE_CACHE_MODE_WT
523 * If new_type is NULL, function will return an error if it cannot reserve the
524 * region with req_type. If new_type is non-NULL, function will return
525 * available type in new_type in case of no error. In case of any error
526 * it will return a negative return value.
528 int reserve_memtype(u64 start, u64 end, enum page_cache_mode req_type,
529 enum page_cache_mode *new_type)
532 enum page_cache_mode actual_type;
536 BUG_ON(start >= end); /* end is exclusive */
538 if (!pat_enabled()) {
539 /* This is identical to page table setting without PAT */
541 *new_type = req_type;
545 /* Low ISA region is always mapped WB in page table. No need to track */
546 if (x86_platform.is_untracked_pat_range(start, end)) {
548 *new_type = _PAGE_CACHE_MODE_WB;
553 * Call mtrr_lookup to get the type hint. This is an
554 * optimization for /dev/mem mmap'ers into WB memory (BIOS
555 * tools and ACPI tools). Use WB request for WB memory and use
556 * UC_MINUS otherwise.
558 actual_type = pat_x_mtrr_type(start, end, req_type);
561 *new_type = actual_type;
563 is_range_ram = pat_pagerange_is_ram(start, end);
564 if (is_range_ram == 1) {
566 err = reserve_ram_pages_type(start, end, req_type, new_type);
569 } else if (is_range_ram < 0) {
573 new = kzalloc(sizeof(struct memtype), GFP_KERNEL);
579 new->type = actual_type;
581 spin_lock(&memtype_lock);
583 err = rbt_memtype_check_insert(new, new_type);
585 pr_info("x86/PAT: reserve_memtype failed [mem %#010Lx-%#010Lx], track %s, req %s\n",
587 cattr_name(new->type), cattr_name(req_type));
589 spin_unlock(&memtype_lock);
594 spin_unlock(&memtype_lock);
596 dprintk("reserve_memtype added [mem %#010Lx-%#010Lx], track %s, req %s, ret %s\n",
597 start, end - 1, cattr_name(new->type), cattr_name(req_type),
598 new_type ? cattr_name(*new_type) : "-");
603 int free_memtype(u64 start, u64 end)
607 struct memtype *entry;
612 /* Low ISA region is always mapped WB. No need to track */
613 if (x86_platform.is_untracked_pat_range(start, end))
616 is_range_ram = pat_pagerange_is_ram(start, end);
617 if (is_range_ram == 1) {
619 err = free_ram_pages_type(start, end);
622 } else if (is_range_ram < 0) {
626 spin_lock(&memtype_lock);
627 entry = rbt_memtype_erase(start, end);
628 spin_unlock(&memtype_lock);
631 pr_info("x86/PAT: %s:%d freeing invalid memtype [mem %#010Lx-%#010Lx]\n",
632 current->comm, current->pid, start, end - 1);
638 dprintk("free_memtype request [mem %#010Lx-%#010Lx]\n", start, end - 1);
645 * lookup_memtype - Looksup the memory type for a physical address
646 * @paddr: physical address of which memory type needs to be looked up
648 * Only to be called when PAT is enabled
650 * Returns _PAGE_CACHE_MODE_WB, _PAGE_CACHE_MODE_WC, _PAGE_CACHE_MODE_UC_MINUS
651 * or _PAGE_CACHE_MODE_WT.
653 static enum page_cache_mode lookup_memtype(u64 paddr)
655 enum page_cache_mode rettype = _PAGE_CACHE_MODE_WB;
656 struct memtype *entry;
658 if (x86_platform.is_untracked_pat_range(paddr, paddr + PAGE_SIZE))
661 if (pat_pagerange_is_ram(paddr, paddr + PAGE_SIZE)) {
664 page = pfn_to_page(paddr >> PAGE_SHIFT);
665 return get_page_memtype(page);
668 spin_lock(&memtype_lock);
670 entry = rbt_memtype_lookup(paddr);
672 rettype = entry->type;
674 rettype = _PAGE_CACHE_MODE_UC_MINUS;
676 spin_unlock(&memtype_lock);
681 * io_reserve_memtype - Request a memory type mapping for a region of memory
682 * @start: start (physical address) of the region
683 * @end: end (physical address) of the region
684 * @type: A pointer to memtype, with requested type. On success, requested
685 * or any other compatible type that was available for the region is returned
687 * On success, returns 0
688 * On failure, returns non-zero
690 int io_reserve_memtype(resource_size_t start, resource_size_t end,
691 enum page_cache_mode *type)
693 resource_size_t size = end - start;
694 enum page_cache_mode req_type = *type;
695 enum page_cache_mode new_type;
698 WARN_ON_ONCE(iomem_map_sanity_check(start, size));
700 ret = reserve_memtype(start, end, req_type, &new_type);
704 if (!is_new_memtype_allowed(start, size, req_type, new_type))
707 if (kernel_map_sync_memtype(start, size, new_type) < 0)
714 free_memtype(start, end);
721 * io_free_memtype - Release a memory type mapping for a region of memory
722 * @start: start (physical address) of the region
723 * @end: end (physical address) of the region
725 void io_free_memtype(resource_size_t start, resource_size_t end)
727 free_memtype(start, end);
730 int arch_io_reserve_memtype_wc(resource_size_t start, resource_size_t size)
732 enum page_cache_mode type = _PAGE_CACHE_MODE_WC;
734 return io_reserve_memtype(start, start + size, &type);
736 EXPORT_SYMBOL(arch_io_reserve_memtype_wc);
738 void arch_io_free_memtype_wc(resource_size_t start, resource_size_t size)
740 io_free_memtype(start, start + size);
742 EXPORT_SYMBOL(arch_io_free_memtype_wc);
744 pgprot_t phys_mem_access_prot(struct file *file, unsigned long pfn,
745 unsigned long size, pgprot_t vma_prot)
747 if (!phys_mem_access_encrypted(pfn << PAGE_SHIFT, size))
748 vma_prot = pgprot_decrypted(vma_prot);
753 #ifdef CONFIG_STRICT_DEVMEM
754 /* This check is done in drivers/char/mem.c in case of STRICT_DEVMEM */
755 static inline int range_is_allowed(unsigned long pfn, unsigned long size)
760 /* This check is needed to avoid cache aliasing when PAT is enabled */
761 static inline int range_is_allowed(unsigned long pfn, unsigned long size)
763 u64 from = ((u64)pfn) << PAGE_SHIFT;
764 u64 to = from + size;
770 while (cursor < to) {
771 if (!devmem_is_allowed(pfn))
778 #endif /* CONFIG_STRICT_DEVMEM */
780 int phys_mem_access_prot_allowed(struct file *file, unsigned long pfn,
781 unsigned long size, pgprot_t *vma_prot)
783 enum page_cache_mode pcm = _PAGE_CACHE_MODE_WB;
785 if (!range_is_allowed(pfn, size))
788 if (file->f_flags & O_DSYNC)
789 pcm = _PAGE_CACHE_MODE_UC_MINUS;
791 *vma_prot = __pgprot((pgprot_val(*vma_prot) & ~_PAGE_CACHE_MASK) |
792 cachemode2protval(pcm));
797 * Change the memory type for the physial address range in kernel identity
798 * mapping space if that range is a part of identity map.
800 int kernel_map_sync_memtype(u64 base, unsigned long size,
801 enum page_cache_mode pcm)
805 if (base > __pa(high_memory-1))
809 * some areas in the middle of the kernel identity range
810 * are not mapped, like the PCI space.
812 if (!page_is_ram(base >> PAGE_SHIFT))
815 id_sz = (__pa(high_memory-1) <= base + size) ?
816 __pa(high_memory) - base :
819 if (ioremap_change_attr((unsigned long)__va(base), id_sz, pcm) < 0) {
820 pr_info("x86/PAT: %s:%d ioremap_change_attr failed %s for [mem %#010Lx-%#010Lx]\n",
821 current->comm, current->pid,
823 base, (unsigned long long)(base + size-1));
830 * Internal interface to reserve a range of physical memory with prot.
831 * Reserved non RAM regions only and after successful reserve_memtype,
832 * this func also keeps identity mapping (if any) in sync with this new prot.
834 static int reserve_pfn_range(u64 paddr, unsigned long size, pgprot_t *vma_prot,
839 enum page_cache_mode want_pcm = pgprot2cachemode(*vma_prot);
840 enum page_cache_mode pcm = want_pcm;
842 is_ram = pat_pagerange_is_ram(paddr, paddr + size);
845 * reserve_pfn_range() for RAM pages. We do not refcount to keep
846 * track of number of mappings of RAM pages. We can assert that
847 * the type requested matches the type of first page in the range.
853 pcm = lookup_memtype(paddr);
854 if (want_pcm != pcm) {
855 pr_warn("x86/PAT: %s:%d map pfn RAM range req %s for [mem %#010Lx-%#010Lx], got %s\n",
856 current->comm, current->pid,
857 cattr_name(want_pcm),
858 (unsigned long long)paddr,
859 (unsigned long long)(paddr + size - 1),
861 *vma_prot = __pgprot((pgprot_val(*vma_prot) &
862 (~_PAGE_CACHE_MASK)) |
863 cachemode2protval(pcm));
868 ret = reserve_memtype(paddr, paddr + size, want_pcm, &pcm);
872 if (pcm != want_pcm) {
874 !is_new_memtype_allowed(paddr, size, want_pcm, pcm)) {
875 free_memtype(paddr, paddr + size);
876 pr_err("x86/PAT: %s:%d map pfn expected mapping type %s for [mem %#010Lx-%#010Lx], got %s\n",
877 current->comm, current->pid,
878 cattr_name(want_pcm),
879 (unsigned long long)paddr,
880 (unsigned long long)(paddr + size - 1),
885 * We allow returning different type than the one requested in
888 *vma_prot = __pgprot((pgprot_val(*vma_prot) &
889 (~_PAGE_CACHE_MASK)) |
890 cachemode2protval(pcm));
893 if (kernel_map_sync_memtype(paddr, size, pcm) < 0) {
894 free_memtype(paddr, paddr + size);
901 * Internal interface to free a range of physical memory.
902 * Frees non RAM regions only.
904 static void free_pfn_range(u64 paddr, unsigned long size)
908 is_ram = pat_pagerange_is_ram(paddr, paddr + size);
910 free_memtype(paddr, paddr + size);
914 * track_pfn_copy is called when vma that is covering the pfnmap gets
915 * copied through copy_page_range().
917 * If the vma has a linear pfn mapping for the entire range, we get the prot
918 * from pte and reserve the entire vma range with single reserve_pfn_range call.
920 int track_pfn_copy(struct vm_area_struct *vma)
922 resource_size_t paddr;
924 unsigned long vma_size = vma->vm_end - vma->vm_start;
927 if (vma->vm_flags & VM_PAT) {
929 * reserve the whole chunk covered by vma. We need the
930 * starting address and protection from pte.
932 if (follow_phys(vma, vma->vm_start, 0, &prot, &paddr)) {
936 pgprot = __pgprot(prot);
937 return reserve_pfn_range(paddr, vma_size, &pgprot, 1);
944 * prot is passed in as a parameter for the new mapping. If the vma has
945 * a linear pfn mapping for the entire range, or no vma is provided,
946 * reserve the entire pfn + size range with single reserve_pfn_range
949 int track_pfn_remap(struct vm_area_struct *vma, pgprot_t *prot,
950 unsigned long pfn, unsigned long addr, unsigned long size)
952 resource_size_t paddr = (resource_size_t)pfn << PAGE_SHIFT;
953 enum page_cache_mode pcm;
955 /* reserve the whole chunk starting from paddr */
956 if (!vma || (addr == vma->vm_start
957 && size == (vma->vm_end - vma->vm_start))) {
960 ret = reserve_pfn_range(paddr, size, prot, 0);
962 vma->vm_flags |= VM_PAT;
970 * For anything smaller than the vma size we set prot based on the
973 pcm = lookup_memtype(paddr);
975 /* Check memtype for the remaining pages */
976 while (size > PAGE_SIZE) {
979 if (pcm != lookup_memtype(paddr))
983 *prot = __pgprot((pgprot_val(*prot) & (~_PAGE_CACHE_MASK)) |
984 cachemode2protval(pcm));
989 void track_pfn_insert(struct vm_area_struct *vma, pgprot_t *prot, pfn_t pfn)
991 enum page_cache_mode pcm;
996 /* Set prot based on lookup */
997 pcm = lookup_memtype(pfn_t_to_phys(pfn));
998 *prot = __pgprot((pgprot_val(*prot) & (~_PAGE_CACHE_MASK)) |
999 cachemode2protval(pcm));
1003 * untrack_pfn is called while unmapping a pfnmap for a region.
1004 * untrack can be called for a specific region indicated by pfn and size or
1005 * can be for the entire vma (in which case pfn, size are zero).
1007 void untrack_pfn(struct vm_area_struct *vma, unsigned long pfn,
1010 resource_size_t paddr;
1013 if (vma && !(vma->vm_flags & VM_PAT))
1016 /* free the chunk starting from pfn or the whole chunk */
1017 paddr = (resource_size_t)pfn << PAGE_SHIFT;
1018 if (!paddr && !size) {
1019 if (follow_phys(vma, vma->vm_start, 0, &prot, &paddr)) {
1024 size = vma->vm_end - vma->vm_start;
1026 free_pfn_range(paddr, size);
1028 vma->vm_flags &= ~VM_PAT;
1032 * untrack_pfn_moved is called, while mremapping a pfnmap for a new region,
1033 * with the old vma after its pfnmap page table has been removed. The new
1034 * vma has a new pfnmap to the same pfn & cache type with VM_PAT set.
1036 void untrack_pfn_moved(struct vm_area_struct *vma)
1038 vma->vm_flags &= ~VM_PAT;
1041 pgprot_t pgprot_writecombine(pgprot_t prot)
1043 return __pgprot(pgprot_val(prot) |
1044 cachemode2protval(_PAGE_CACHE_MODE_WC));
1046 EXPORT_SYMBOL_GPL(pgprot_writecombine);
1048 pgprot_t pgprot_writethrough(pgprot_t prot)
1050 return __pgprot(pgprot_val(prot) |
1051 cachemode2protval(_PAGE_CACHE_MODE_WT));
1053 EXPORT_SYMBOL_GPL(pgprot_writethrough);
1055 #if defined(CONFIG_DEBUG_FS) && defined(CONFIG_X86_PAT)
1057 static struct memtype *memtype_get_idx(loff_t pos)
1059 struct memtype *print_entry;
1062 print_entry = kzalloc(sizeof(struct memtype), GFP_KERNEL);
1066 spin_lock(&memtype_lock);
1067 ret = rbt_memtype_copy_nth_element(print_entry, pos);
1068 spin_unlock(&memtype_lock);
1078 static void *memtype_seq_start(struct seq_file *seq, loff_t *pos)
1082 seq_puts(seq, "PAT memtype list:\n");
1085 return memtype_get_idx(*pos);
1088 static void *memtype_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1091 return memtype_get_idx(*pos);
1094 static void memtype_seq_stop(struct seq_file *seq, void *v)
1098 static int memtype_seq_show(struct seq_file *seq, void *v)
1100 struct memtype *print_entry = (struct memtype *)v;
1102 seq_printf(seq, "%s @ 0x%Lx-0x%Lx\n", cattr_name(print_entry->type),
1103 print_entry->start, print_entry->end);
1109 static const struct seq_operations memtype_seq_ops = {
1110 .start = memtype_seq_start,
1111 .next = memtype_seq_next,
1112 .stop = memtype_seq_stop,
1113 .show = memtype_seq_show,
1116 static int memtype_seq_open(struct inode *inode, struct file *file)
1118 return seq_open(file, &memtype_seq_ops);
1121 static const struct file_operations memtype_fops = {
1122 .open = memtype_seq_open,
1124 .llseek = seq_lseek,
1125 .release = seq_release,
1128 static int __init pat_memtype_list_init(void)
1130 if (pat_enabled()) {
1131 debugfs_create_file("pat_memtype_list", S_IRUSR,
1132 arch_debugfs_dir, NULL, &memtype_fops);
1137 late_initcall(pat_memtype_list_init);
1139 #endif /* CONFIG_DEBUG_FS && CONFIG_X86_PAT */