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/pfn_t.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);
42 static void init_cache_modes(void);
44 void pat_disable(const char *reason)
50 WARN_ONCE(1, "x86/PAT: PAT cannot be disabled after initialization\n");
55 pr_info("x86/PAT: %s\n", reason);
60 static int __init nopat(char *str)
62 pat_disable("PAT support disabled.");
65 early_param("nopat", nopat);
67 bool pat_enabled(void)
69 return !!__pat_enabled;
71 EXPORT_SYMBOL_GPL(pat_enabled);
75 static int __init pat_debug_setup(char *str)
80 __setup("debugpat", pat_debug_setup);
84 * X86 PAT uses page flags arch_1 and uncached together to keep track of
85 * memory type of pages that have backing page struct.
87 * X86 PAT supports 4 different memory types:
88 * - _PAGE_CACHE_MODE_WB
89 * - _PAGE_CACHE_MODE_WC
90 * - _PAGE_CACHE_MODE_UC_MINUS
91 * - _PAGE_CACHE_MODE_WT
93 * _PAGE_CACHE_MODE_WB is the default type.
97 #define _PGMT_WC (1UL << PG_arch_1)
98 #define _PGMT_UC_MINUS (1UL << PG_uncached)
99 #define _PGMT_WT (1UL << PG_uncached | 1UL << PG_arch_1)
100 #define _PGMT_MASK (1UL << PG_uncached | 1UL << PG_arch_1)
101 #define _PGMT_CLEAR_MASK (~_PGMT_MASK)
103 static inline enum page_cache_mode get_page_memtype(struct page *pg)
105 unsigned long pg_flags = pg->flags & _PGMT_MASK;
107 if (pg_flags == _PGMT_WB)
108 return _PAGE_CACHE_MODE_WB;
109 else if (pg_flags == _PGMT_WC)
110 return _PAGE_CACHE_MODE_WC;
111 else if (pg_flags == _PGMT_UC_MINUS)
112 return _PAGE_CACHE_MODE_UC_MINUS;
114 return _PAGE_CACHE_MODE_WT;
117 static inline void set_page_memtype(struct page *pg,
118 enum page_cache_mode memtype)
120 unsigned long memtype_flags;
121 unsigned long old_flags;
122 unsigned long new_flags;
125 case _PAGE_CACHE_MODE_WC:
126 memtype_flags = _PGMT_WC;
128 case _PAGE_CACHE_MODE_UC_MINUS:
129 memtype_flags = _PGMT_UC_MINUS;
131 case _PAGE_CACHE_MODE_WT:
132 memtype_flags = _PGMT_WT;
134 case _PAGE_CACHE_MODE_WB:
136 memtype_flags = _PGMT_WB;
141 old_flags = pg->flags;
142 new_flags = (old_flags & _PGMT_CLEAR_MASK) | memtype_flags;
143 } while (cmpxchg(&pg->flags, old_flags, new_flags) != old_flags);
146 static inline enum page_cache_mode get_page_memtype(struct page *pg)
150 static inline void set_page_memtype(struct page *pg,
151 enum page_cache_mode memtype)
157 PAT_UC = 0, /* uncached */
158 PAT_WC = 1, /* Write combining */
159 PAT_WT = 4, /* Write Through */
160 PAT_WP = 5, /* Write Protected */
161 PAT_WB = 6, /* Write Back (default) */
162 PAT_UC_MINUS = 7, /* UC, but can be overridden by MTRR */
165 #define CM(c) (_PAGE_CACHE_MODE_ ## c)
167 static enum page_cache_mode pat_get_cache_mode(unsigned pat_val, char *msg)
169 enum page_cache_mode cache;
173 case PAT_UC: cache = CM(UC); cache_mode = "UC "; break;
174 case PAT_WC: cache = CM(WC); cache_mode = "WC "; break;
175 case PAT_WT: cache = CM(WT); cache_mode = "WT "; break;
176 case PAT_WP: cache = CM(WP); cache_mode = "WP "; break;
177 case PAT_WB: cache = CM(WB); cache_mode = "WB "; break;
178 case PAT_UC_MINUS: cache = CM(UC_MINUS); cache_mode = "UC- "; break;
179 default: cache = CM(WB); cache_mode = "WB "; break;
182 memcpy(msg, cache_mode, 4);
190 * Update the cache mode to pgprot translation tables according to PAT
192 * Using lower indices is preferred, so we start with highest index.
194 static void __init_cache_modes(u64 pat)
196 enum page_cache_mode cache;
201 for (i = 7; i >= 0; i--) {
202 cache = pat_get_cache_mode((pat >> (i * 8)) & 7,
204 update_cache_mode_entry(i, cache);
206 pr_info("x86/PAT: Configuration [0-7]: %s\n", pat_msg);
209 #define PAT(x, y) ((u64)PAT_ ## y << ((x)*8))
211 static void pat_bsp_init(u64 pat)
215 if (!boot_cpu_has(X86_FEATURE_PAT)) {
216 pat_disable("PAT not supported by CPU.");
220 rdmsrl(MSR_IA32_CR_PAT, tmp_pat);
222 pat_disable("PAT MSR is 0, disabled.");
226 wrmsrl(MSR_IA32_CR_PAT, pat);
228 __init_cache_modes(pat);
231 static void pat_ap_init(u64 pat)
233 if (!boot_cpu_has(X86_FEATURE_PAT)) {
235 * If this happens we are on a secondary CPU, but switched to
236 * PAT on the boot CPU. We have no way to undo PAT.
238 panic("x86/PAT: PAT enabled, but not supported by secondary CPU\n");
241 wrmsrl(MSR_IA32_CR_PAT, pat);
244 static void init_cache_modes(void)
247 static int init_cm_done;
252 if (boot_cpu_has(X86_FEATURE_PAT)) {
254 * CPU supports PAT. Set PAT table to be consistent with
255 * PAT MSR. This case supports "nopat" boot option, and
256 * virtual machine environments which support PAT without
257 * MTRRs. In specific, Xen has unique setup to PAT MSR.
259 * If PAT MSR returns 0, it is considered invalid and emulates
262 rdmsrl(MSR_IA32_CR_PAT, pat);
267 * No PAT. Emulate the PAT table that corresponds to the two
268 * cache bits, PWT (Write Through) and PCD (Cache Disable).
269 * This setup is also the same as the BIOS default setup.
276 * 00 0 WB : _PAGE_CACHE_MODE_WB
277 * 01 1 WT : _PAGE_CACHE_MODE_WT
278 * 10 2 UC-: _PAGE_CACHE_MODE_UC_MINUS
279 * 11 3 UC : _PAGE_CACHE_MODE_UC
281 * NOTE: When WC or WP is used, it is redirected to UC- per
282 * the default setup in __cachemode2pte_tbl[].
284 pat = PAT(0, WB) | PAT(1, WT) | PAT(2, UC_MINUS) | PAT(3, UC) |
285 PAT(4, WB) | PAT(5, WT) | PAT(6, UC_MINUS) | PAT(7, UC);
288 __init_cache_modes(pat);
294 * pat_init - Initialize PAT MSR and PAT table
296 * This function initializes PAT MSR and PAT table with an OS-defined value
297 * to enable additional cache attributes, WC and WT.
299 * This function must be called on all CPUs using the specific sequence of
300 * operations defined in Intel SDM. mtrr_rendezvous_handler() provides this
306 struct cpuinfo_x86 *c = &boot_cpu_data;
308 if (!pat_enabled()) {
313 if ((c->x86_vendor == X86_VENDOR_INTEL) &&
314 (((c->x86 == 0x6) && (c->x86_model <= 0xd)) ||
315 ((c->x86 == 0xf) && (c->x86_model <= 0x6)))) {
317 * PAT support with the lower four entries. Intel Pentium 2,
318 * 3, M, and 4 are affected by PAT errata, which makes the
319 * upper four entries unusable. To be on the safe side, we don't
327 * 000 0 WB : _PAGE_CACHE_MODE_WB
328 * 001 1 WC : _PAGE_CACHE_MODE_WC
329 * 010 2 UC-: _PAGE_CACHE_MODE_UC_MINUS
330 * 011 3 UC : _PAGE_CACHE_MODE_UC
333 * NOTE: When WT or WP is used, it is redirected to UC- per
334 * the default setup in __cachemode2pte_tbl[].
336 pat = PAT(0, WB) | PAT(1, WC) | PAT(2, UC_MINUS) | PAT(3, UC) |
337 PAT(4, WB) | PAT(5, WC) | PAT(6, UC_MINUS) | PAT(7, UC);
340 * Full PAT support. We put WT in slot 7 to improve
341 * robustness in the presence of errata that might cause
342 * the high PAT bit to be ignored. This way, a buggy slot 7
343 * access will hit slot 3, and slot 3 is UC, so at worst
344 * we lose performance without causing a correctness issue.
345 * Pentium 4 erratum N46 is an example for such an erratum,
346 * although we try not to use PAT at all on affected CPUs.
353 * 000 0 WB : _PAGE_CACHE_MODE_WB
354 * 001 1 WC : _PAGE_CACHE_MODE_WC
355 * 010 2 UC-: _PAGE_CACHE_MODE_UC_MINUS
356 * 011 3 UC : _PAGE_CACHE_MODE_UC
357 * 100 4 WB : Reserved
358 * 101 5 WC : Reserved
359 * 110 6 UC-: Reserved
360 * 111 7 WT : _PAGE_CACHE_MODE_WT
362 * The reserved slots are unused, but mapped to their
363 * corresponding types in the presence of PAT errata.
365 pat = PAT(0, WB) | PAT(1, WC) | PAT(2, UC_MINUS) | PAT(3, UC) |
366 PAT(4, WB) | PAT(5, WC) | PAT(6, UC_MINUS) | PAT(7, WT);
369 if (!boot_cpu_done) {
371 boot_cpu_done = true;
379 static DEFINE_SPINLOCK(memtype_lock); /* protects memtype accesses */
382 * Does intersection of PAT memory type and MTRR memory type and returns
383 * the resulting memory type as PAT understands it.
384 * (Type in pat and mtrr will not have same value)
385 * The intersection is based on "Effective Memory Type" tables in IA-32
388 static unsigned long pat_x_mtrr_type(u64 start, u64 end,
389 enum page_cache_mode req_type)
392 * Look for MTRR hint to get the effective type in case where PAT
395 if (req_type == _PAGE_CACHE_MODE_WB) {
396 u8 mtrr_type, uniform;
398 mtrr_type = mtrr_type_lookup(start, end, &uniform);
399 if (mtrr_type != MTRR_TYPE_WRBACK)
400 return _PAGE_CACHE_MODE_UC_MINUS;
402 return _PAGE_CACHE_MODE_WB;
408 struct pagerange_state {
409 unsigned long cur_pfn;
415 pagerange_is_ram_callback(unsigned long initial_pfn, unsigned long total_nr_pages, void *arg)
417 struct pagerange_state *state = arg;
419 state->not_ram |= initial_pfn > state->cur_pfn;
420 state->ram |= total_nr_pages > 0;
421 state->cur_pfn = initial_pfn + total_nr_pages;
423 return state->ram && state->not_ram;
426 static int pat_pagerange_is_ram(resource_size_t start, resource_size_t end)
429 unsigned long start_pfn = start >> PAGE_SHIFT;
430 unsigned long end_pfn = (end + PAGE_SIZE - 1) >> PAGE_SHIFT;
431 struct pagerange_state state = {start_pfn, 0, 0};
434 * For legacy reasons, physical address range in the legacy ISA
435 * region is tracked as non-RAM. This will allow users of
436 * /dev/mem to map portions of legacy ISA region, even when
437 * some of those portions are listed(or not even listed) with
438 * different e820 types(RAM/reserved/..)
440 if (start_pfn < ISA_END_ADDRESS >> PAGE_SHIFT)
441 start_pfn = ISA_END_ADDRESS >> PAGE_SHIFT;
443 if (start_pfn < end_pfn) {
444 ret = walk_system_ram_range(start_pfn, end_pfn - start_pfn,
445 &state, pagerange_is_ram_callback);
448 return (ret > 0) ? -1 : (state.ram ? 1 : 0);
452 * For RAM pages, we use page flags to mark the pages with appropriate type.
453 * The page flags are limited to four types, WB (default), WC, WT and UC-.
454 * WP request fails with -EINVAL, and UC gets redirected to UC-. Setting
455 * a new memory type is only allowed for a page mapped with the default WB
458 * Here we do two passes:
459 * - Find the memtype of all the pages in the range, look for any conflicts.
460 * - In case of no conflicts, set the new memtype for pages in the range.
462 static int reserve_ram_pages_type(u64 start, u64 end,
463 enum page_cache_mode req_type,
464 enum page_cache_mode *new_type)
469 if (req_type == _PAGE_CACHE_MODE_WP) {
471 *new_type = _PAGE_CACHE_MODE_UC_MINUS;
475 if (req_type == _PAGE_CACHE_MODE_UC) {
476 /* We do not support strong UC */
478 req_type = _PAGE_CACHE_MODE_UC_MINUS;
481 for (pfn = (start >> PAGE_SHIFT); pfn < (end >> PAGE_SHIFT); ++pfn) {
482 enum page_cache_mode type;
484 page = pfn_to_page(pfn);
485 type = get_page_memtype(page);
486 if (type != _PAGE_CACHE_MODE_WB) {
487 pr_info("x86/PAT: reserve_ram_pages_type failed [mem %#010Lx-%#010Lx], track 0x%x, req 0x%x\n",
488 start, end - 1, type, req_type);
497 *new_type = req_type;
499 for (pfn = (start >> PAGE_SHIFT); pfn < (end >> PAGE_SHIFT); ++pfn) {
500 page = pfn_to_page(pfn);
501 set_page_memtype(page, req_type);
506 static int free_ram_pages_type(u64 start, u64 end)
511 for (pfn = (start >> PAGE_SHIFT); pfn < (end >> PAGE_SHIFT); ++pfn) {
512 page = pfn_to_page(pfn);
513 set_page_memtype(page, _PAGE_CACHE_MODE_WB);
519 * req_type typically has one of the:
520 * - _PAGE_CACHE_MODE_WB
521 * - _PAGE_CACHE_MODE_WC
522 * - _PAGE_CACHE_MODE_UC_MINUS
523 * - _PAGE_CACHE_MODE_UC
524 * - _PAGE_CACHE_MODE_WT
526 * If new_type is NULL, function will return an error if it cannot reserve the
527 * region with req_type. If new_type is non-NULL, function will return
528 * available type in new_type in case of no error. In case of any error
529 * it will return a negative return value.
531 int reserve_memtype(u64 start, u64 end, enum page_cache_mode req_type,
532 enum page_cache_mode *new_type)
535 enum page_cache_mode actual_type;
539 BUG_ON(start >= end); /* end is exclusive */
541 if (!pat_enabled()) {
542 /* This is identical to page table setting without PAT */
544 *new_type = req_type;
548 /* Low ISA region is always mapped WB in page table. No need to track */
549 if (x86_platform.is_untracked_pat_range(start, end)) {
551 *new_type = _PAGE_CACHE_MODE_WB;
556 * Call mtrr_lookup to get the type hint. This is an
557 * optimization for /dev/mem mmap'ers into WB memory (BIOS
558 * tools and ACPI tools). Use WB request for WB memory and use
559 * UC_MINUS otherwise.
561 actual_type = pat_x_mtrr_type(start, end, req_type);
564 *new_type = actual_type;
566 is_range_ram = pat_pagerange_is_ram(start, end);
567 if (is_range_ram == 1) {
569 err = reserve_ram_pages_type(start, end, req_type, new_type);
572 } else if (is_range_ram < 0) {
576 new = kzalloc(sizeof(struct memtype), GFP_KERNEL);
582 new->type = actual_type;
584 spin_lock(&memtype_lock);
586 err = rbt_memtype_check_insert(new, new_type);
588 pr_info("x86/PAT: reserve_memtype failed [mem %#010Lx-%#010Lx], track %s, req %s\n",
590 cattr_name(new->type), cattr_name(req_type));
592 spin_unlock(&memtype_lock);
597 spin_unlock(&memtype_lock);
599 dprintk("reserve_memtype added [mem %#010Lx-%#010Lx], track %s, req %s, ret %s\n",
600 start, end - 1, cattr_name(new->type), cattr_name(req_type),
601 new_type ? cattr_name(*new_type) : "-");
606 int free_memtype(u64 start, u64 end)
610 struct memtype *entry;
615 /* Low ISA region is always mapped WB. No need to track */
616 if (x86_platform.is_untracked_pat_range(start, end))
619 is_range_ram = pat_pagerange_is_ram(start, end);
620 if (is_range_ram == 1) {
622 err = free_ram_pages_type(start, end);
625 } else if (is_range_ram < 0) {
629 spin_lock(&memtype_lock);
630 entry = rbt_memtype_erase(start, end);
631 spin_unlock(&memtype_lock);
634 pr_info("x86/PAT: %s:%d freeing invalid memtype [mem %#010Lx-%#010Lx]\n",
635 current->comm, current->pid, start, end - 1);
641 dprintk("free_memtype request [mem %#010Lx-%#010Lx]\n", start, end - 1);
648 * lookup_memtype - Looksup the memory type for a physical address
649 * @paddr: physical address of which memory type needs to be looked up
651 * Only to be called when PAT is enabled
653 * Returns _PAGE_CACHE_MODE_WB, _PAGE_CACHE_MODE_WC, _PAGE_CACHE_MODE_UC_MINUS
654 * or _PAGE_CACHE_MODE_WT.
656 static enum page_cache_mode lookup_memtype(u64 paddr)
658 enum page_cache_mode rettype = _PAGE_CACHE_MODE_WB;
659 struct memtype *entry;
661 if (x86_platform.is_untracked_pat_range(paddr, paddr + PAGE_SIZE))
664 if (pat_pagerange_is_ram(paddr, paddr + PAGE_SIZE)) {
667 page = pfn_to_page(paddr >> PAGE_SHIFT);
668 return get_page_memtype(page);
671 spin_lock(&memtype_lock);
673 entry = rbt_memtype_lookup(paddr);
675 rettype = entry->type;
677 rettype = _PAGE_CACHE_MODE_UC_MINUS;
679 spin_unlock(&memtype_lock);
684 * io_reserve_memtype - Request a memory type mapping for a region of memory
685 * @start: start (physical address) of the region
686 * @end: end (physical address) of the region
687 * @type: A pointer to memtype, with requested type. On success, requested
688 * or any other compatible type that was available for the region is returned
690 * On success, returns 0
691 * On failure, returns non-zero
693 int io_reserve_memtype(resource_size_t start, resource_size_t end,
694 enum page_cache_mode *type)
696 resource_size_t size = end - start;
697 enum page_cache_mode req_type = *type;
698 enum page_cache_mode new_type;
701 WARN_ON_ONCE(iomem_map_sanity_check(start, size));
703 ret = reserve_memtype(start, end, req_type, &new_type);
707 if (!is_new_memtype_allowed(start, size, req_type, new_type))
710 if (kernel_map_sync_memtype(start, size, new_type) < 0)
717 free_memtype(start, end);
724 * io_free_memtype - Release a memory type mapping for a region of memory
725 * @start: start (physical address) of the region
726 * @end: end (physical address) of the region
728 void io_free_memtype(resource_size_t start, resource_size_t end)
730 free_memtype(start, end);
733 pgprot_t phys_mem_access_prot(struct file *file, unsigned long pfn,
734 unsigned long size, pgprot_t vma_prot)
739 #ifdef CONFIG_STRICT_DEVMEM
740 /* This check is done in drivers/char/mem.c in case of STRICT_DEVMEM */
741 static inline int range_is_allowed(unsigned long pfn, unsigned long size)
746 /* This check is needed to avoid cache aliasing when PAT is enabled */
747 static inline int range_is_allowed(unsigned long pfn, unsigned long size)
749 u64 from = ((u64)pfn) << PAGE_SHIFT;
750 u64 to = from + size;
756 while (cursor < to) {
757 if (!devmem_is_allowed(pfn))
764 #endif /* CONFIG_STRICT_DEVMEM */
766 int phys_mem_access_prot_allowed(struct file *file, unsigned long pfn,
767 unsigned long size, pgprot_t *vma_prot)
769 enum page_cache_mode pcm = _PAGE_CACHE_MODE_WB;
771 if (!range_is_allowed(pfn, size))
774 if (file->f_flags & O_DSYNC)
775 pcm = _PAGE_CACHE_MODE_UC_MINUS;
777 *vma_prot = __pgprot((pgprot_val(*vma_prot) & ~_PAGE_CACHE_MASK) |
778 cachemode2protval(pcm));
783 * Change the memory type for the physial address range in kernel identity
784 * mapping space if that range is a part of identity map.
786 int kernel_map_sync_memtype(u64 base, unsigned long size,
787 enum page_cache_mode pcm)
791 if (base > __pa(high_memory-1))
795 * some areas in the middle of the kernel identity range
796 * are not mapped, like the PCI space.
798 if (!page_is_ram(base >> PAGE_SHIFT))
801 id_sz = (__pa(high_memory-1) <= base + size) ?
802 __pa(high_memory) - base :
805 if (ioremap_change_attr((unsigned long)__va(base), id_sz, pcm) < 0) {
806 pr_info("x86/PAT: %s:%d ioremap_change_attr failed %s for [mem %#010Lx-%#010Lx]\n",
807 current->comm, current->pid,
809 base, (unsigned long long)(base + size-1));
816 * Internal interface to reserve a range of physical memory with prot.
817 * Reserved non RAM regions only and after successful reserve_memtype,
818 * this func also keeps identity mapping (if any) in sync with this new prot.
820 static int reserve_pfn_range(u64 paddr, unsigned long size, pgprot_t *vma_prot,
825 enum page_cache_mode want_pcm = pgprot2cachemode(*vma_prot);
826 enum page_cache_mode pcm = want_pcm;
828 is_ram = pat_pagerange_is_ram(paddr, paddr + size);
831 * reserve_pfn_range() for RAM pages. We do not refcount to keep
832 * track of number of mappings of RAM pages. We can assert that
833 * the type requested matches the type of first page in the range.
839 pcm = lookup_memtype(paddr);
840 if (want_pcm != pcm) {
841 pr_warn("x86/PAT: %s:%d map pfn RAM range req %s for [mem %#010Lx-%#010Lx], got %s\n",
842 current->comm, current->pid,
843 cattr_name(want_pcm),
844 (unsigned long long)paddr,
845 (unsigned long long)(paddr + size - 1),
847 *vma_prot = __pgprot((pgprot_val(*vma_prot) &
848 (~_PAGE_CACHE_MASK)) |
849 cachemode2protval(pcm));
854 ret = reserve_memtype(paddr, paddr + size, want_pcm, &pcm);
858 if (pcm != want_pcm) {
860 !is_new_memtype_allowed(paddr, size, want_pcm, pcm)) {
861 free_memtype(paddr, paddr + size);
862 pr_err("x86/PAT: %s:%d map pfn expected mapping type %s for [mem %#010Lx-%#010Lx], got %s\n",
863 current->comm, current->pid,
864 cattr_name(want_pcm),
865 (unsigned long long)paddr,
866 (unsigned long long)(paddr + size - 1),
871 * We allow returning different type than the one requested in
874 *vma_prot = __pgprot((pgprot_val(*vma_prot) &
875 (~_PAGE_CACHE_MASK)) |
876 cachemode2protval(pcm));
879 if (kernel_map_sync_memtype(paddr, size, pcm) < 0) {
880 free_memtype(paddr, paddr + size);
887 * Internal interface to free a range of physical memory.
888 * Frees non RAM regions only.
890 static void free_pfn_range(u64 paddr, unsigned long size)
894 is_ram = pat_pagerange_is_ram(paddr, paddr + size);
896 free_memtype(paddr, paddr + size);
900 * track_pfn_copy is called when vma that is covering the pfnmap gets
901 * copied through copy_page_range().
903 * If the vma has a linear pfn mapping for the entire range, we get the prot
904 * from pte and reserve the entire vma range with single reserve_pfn_range call.
906 int track_pfn_copy(struct vm_area_struct *vma)
908 resource_size_t paddr;
910 unsigned long vma_size = vma->vm_end - vma->vm_start;
913 if (vma->vm_flags & VM_PAT) {
915 * reserve the whole chunk covered by vma. We need the
916 * starting address and protection from pte.
918 if (follow_phys(vma, vma->vm_start, 0, &prot, &paddr)) {
922 pgprot = __pgprot(prot);
923 return reserve_pfn_range(paddr, vma_size, &pgprot, 1);
930 * prot is passed in as a parameter for the new mapping. If the vma has a
931 * linear pfn mapping for the entire range reserve the entire vma range with
932 * single reserve_pfn_range call.
934 int track_pfn_remap(struct vm_area_struct *vma, pgprot_t *prot,
935 unsigned long pfn, unsigned long addr, unsigned long size)
937 resource_size_t paddr = (resource_size_t)pfn << PAGE_SHIFT;
938 enum page_cache_mode pcm;
940 /* reserve the whole chunk starting from paddr */
941 if (addr == vma->vm_start && size == (vma->vm_end - vma->vm_start)) {
944 ret = reserve_pfn_range(paddr, size, prot, 0);
946 vma->vm_flags |= VM_PAT;
954 * For anything smaller than the vma size we set prot based on the
957 pcm = lookup_memtype(paddr);
959 /* Check memtype for the remaining pages */
960 while (size > PAGE_SIZE) {
963 if (pcm != lookup_memtype(paddr))
967 *prot = __pgprot((pgprot_val(*prot) & (~_PAGE_CACHE_MASK)) |
968 cachemode2protval(pcm));
973 int track_pfn_insert(struct vm_area_struct *vma, pgprot_t *prot,
976 enum page_cache_mode pcm;
981 /* Set prot based on lookup */
982 pcm = lookup_memtype(pfn_t_to_phys(pfn));
983 *prot = __pgprot((pgprot_val(*prot) & (~_PAGE_CACHE_MASK)) |
984 cachemode2protval(pcm));
990 * untrack_pfn is called while unmapping a pfnmap for a region.
991 * untrack can be called for a specific region indicated by pfn and size or
992 * can be for the entire vma (in which case pfn, size are zero).
994 void untrack_pfn(struct vm_area_struct *vma, unsigned long pfn,
997 resource_size_t paddr;
1000 if (!(vma->vm_flags & VM_PAT))
1003 /* free the chunk starting from pfn or the whole chunk */
1004 paddr = (resource_size_t)pfn << PAGE_SHIFT;
1005 if (!paddr && !size) {
1006 if (follow_phys(vma, vma->vm_start, 0, &prot, &paddr)) {
1011 size = vma->vm_end - vma->vm_start;
1013 free_pfn_range(paddr, size);
1014 vma->vm_flags &= ~VM_PAT;
1018 * untrack_pfn_moved is called, while mremapping a pfnmap for a new region,
1019 * with the old vma after its pfnmap page table has been removed. The new
1020 * vma has a new pfnmap to the same pfn & cache type with VM_PAT set.
1022 void untrack_pfn_moved(struct vm_area_struct *vma)
1024 vma->vm_flags &= ~VM_PAT;
1027 pgprot_t pgprot_writecombine(pgprot_t prot)
1029 return __pgprot(pgprot_val(prot) |
1030 cachemode2protval(_PAGE_CACHE_MODE_WC));
1032 EXPORT_SYMBOL_GPL(pgprot_writecombine);
1034 pgprot_t pgprot_writethrough(pgprot_t prot)
1036 return __pgprot(pgprot_val(prot) |
1037 cachemode2protval(_PAGE_CACHE_MODE_WT));
1039 EXPORT_SYMBOL_GPL(pgprot_writethrough);
1041 #if defined(CONFIG_DEBUG_FS) && defined(CONFIG_X86_PAT)
1043 static struct memtype *memtype_get_idx(loff_t pos)
1045 struct memtype *print_entry;
1048 print_entry = kzalloc(sizeof(struct memtype), GFP_KERNEL);
1052 spin_lock(&memtype_lock);
1053 ret = rbt_memtype_copy_nth_element(print_entry, pos);
1054 spin_unlock(&memtype_lock);
1064 static void *memtype_seq_start(struct seq_file *seq, loff_t *pos)
1068 seq_puts(seq, "PAT memtype list:\n");
1071 return memtype_get_idx(*pos);
1074 static void *memtype_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1077 return memtype_get_idx(*pos);
1080 static void memtype_seq_stop(struct seq_file *seq, void *v)
1084 static int memtype_seq_show(struct seq_file *seq, void *v)
1086 struct memtype *print_entry = (struct memtype *)v;
1088 seq_printf(seq, "%s @ 0x%Lx-0x%Lx\n", cattr_name(print_entry->type),
1089 print_entry->start, print_entry->end);
1095 static const struct seq_operations memtype_seq_ops = {
1096 .start = memtype_seq_start,
1097 .next = memtype_seq_next,
1098 .stop = memtype_seq_stop,
1099 .show = memtype_seq_show,
1102 static int memtype_seq_open(struct inode *inode, struct file *file)
1104 return seq_open(file, &memtype_seq_ops);
1107 static const struct file_operations memtype_fops = {
1108 .open = memtype_seq_open,
1110 .llseek = seq_lseek,
1111 .release = seq_release,
1114 static int __init pat_memtype_list_init(void)
1116 if (pat_enabled()) {
1117 debugfs_create_file("pat_memtype_list", S_IRUSR,
1118 arch_debugfs_dir, NULL, &memtype_fops);
1123 late_initcall(pat_memtype_list_init);
1125 #endif /* CONFIG_DEBUG_FS && CONFIG_X86_PAT */