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 __read_mostly boot_cpu_done;
40 static bool __read_mostly pat_disabled = !IS_ENABLED(CONFIG_X86_PAT);
41 static bool __read_mostly pat_initialized;
42 static bool __read_mostly init_cm_done;
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);
58 static int __init nopat(char *str)
60 pat_disable("PAT support disabled.");
63 early_param("nopat", nopat);
65 bool pat_enabled(void)
67 return pat_initialized;
69 EXPORT_SYMBOL_GPL(pat_enabled);
73 static int __init pat_debug_setup(char *str)
78 __setup("debugpat", pat_debug_setup);
82 * X86 PAT uses page flags arch_1 and uncached together to keep track of
83 * memory type of pages that have backing page struct.
85 * X86 PAT supports 4 different memory types:
86 * - _PAGE_CACHE_MODE_WB
87 * - _PAGE_CACHE_MODE_WC
88 * - _PAGE_CACHE_MODE_UC_MINUS
89 * - _PAGE_CACHE_MODE_WT
91 * _PAGE_CACHE_MODE_WB is the default type.
95 #define _PGMT_WC (1UL << PG_arch_1)
96 #define _PGMT_UC_MINUS (1UL << PG_uncached)
97 #define _PGMT_WT (1UL << PG_uncached | 1UL << PG_arch_1)
98 #define _PGMT_MASK (1UL << PG_uncached | 1UL << PG_arch_1)
99 #define _PGMT_CLEAR_MASK (~_PGMT_MASK)
101 static inline enum page_cache_mode get_page_memtype(struct page *pg)
103 unsigned long pg_flags = pg->flags & _PGMT_MASK;
105 if (pg_flags == _PGMT_WB)
106 return _PAGE_CACHE_MODE_WB;
107 else if (pg_flags == _PGMT_WC)
108 return _PAGE_CACHE_MODE_WC;
109 else if (pg_flags == _PGMT_UC_MINUS)
110 return _PAGE_CACHE_MODE_UC_MINUS;
112 return _PAGE_CACHE_MODE_WT;
115 static inline void set_page_memtype(struct page *pg,
116 enum page_cache_mode memtype)
118 unsigned long memtype_flags;
119 unsigned long old_flags;
120 unsigned long new_flags;
123 case _PAGE_CACHE_MODE_WC:
124 memtype_flags = _PGMT_WC;
126 case _PAGE_CACHE_MODE_UC_MINUS:
127 memtype_flags = _PGMT_UC_MINUS;
129 case _PAGE_CACHE_MODE_WT:
130 memtype_flags = _PGMT_WT;
132 case _PAGE_CACHE_MODE_WB:
134 memtype_flags = _PGMT_WB;
139 old_flags = pg->flags;
140 new_flags = (old_flags & _PGMT_CLEAR_MASK) | memtype_flags;
141 } while (cmpxchg(&pg->flags, old_flags, new_flags) != old_flags);
144 static inline enum page_cache_mode get_page_memtype(struct page *pg)
148 static inline void set_page_memtype(struct page *pg,
149 enum page_cache_mode memtype)
155 PAT_UC = 0, /* uncached */
156 PAT_WC = 1, /* Write combining */
157 PAT_WT = 4, /* Write Through */
158 PAT_WP = 5, /* Write Protected */
159 PAT_WB = 6, /* Write Back (default) */
160 PAT_UC_MINUS = 7, /* UC, but can be overriden by MTRR */
163 #define CM(c) (_PAGE_CACHE_MODE_ ## c)
165 static enum page_cache_mode pat_get_cache_mode(unsigned pat_val, char *msg)
167 enum page_cache_mode cache;
171 case PAT_UC: cache = CM(UC); cache_mode = "UC "; break;
172 case PAT_WC: cache = CM(WC); cache_mode = "WC "; break;
173 case PAT_WT: cache = CM(WT); cache_mode = "WT "; break;
174 case PAT_WP: cache = CM(WP); cache_mode = "WP "; break;
175 case PAT_WB: cache = CM(WB); cache_mode = "WB "; break;
176 case PAT_UC_MINUS: cache = CM(UC_MINUS); cache_mode = "UC- "; break;
177 default: cache = CM(WB); cache_mode = "WB "; break;
180 memcpy(msg, cache_mode, 4);
188 * Update the cache mode to pgprot translation tables according to PAT
190 * Using lower indices is preferred, so we start with highest index.
192 static void __init_cache_modes(u64 pat)
194 enum page_cache_mode cache;
199 for (i = 7; i >= 0; i--) {
200 cache = pat_get_cache_mode((pat >> (i * 8)) & 7,
202 update_cache_mode_entry(i, cache);
204 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);
227 pat_initialized = true;
229 __init_cache_modes(pat);
232 static void pat_ap_init(u64 pat)
234 if (!boot_cpu_has(X86_FEATURE_PAT)) {
236 * If this happens we are on a secondary CPU, but switched to
237 * PAT on the boot CPU. We have no way to undo PAT.
239 panic("x86/PAT: PAT enabled, but not supported by secondary CPU\n");
242 wrmsrl(MSR_IA32_CR_PAT, pat);
245 void init_cache_modes(void)
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);
292 * pat_init - Initialize PAT MSR and PAT table
294 * This function initializes PAT MSR and PAT table with an OS-defined value
295 * to enable additional cache attributes, WC and WT.
297 * This function must be called on all CPUs using the specific sequence of
298 * operations defined in Intel SDM. mtrr_rendezvous_handler() provides this
304 struct cpuinfo_x86 *c = &boot_cpu_data;
309 if ((c->x86_vendor == X86_VENDOR_INTEL) &&
310 (((c->x86 == 0x6) && (c->x86_model <= 0xd)) ||
311 ((c->x86 == 0xf) && (c->x86_model <= 0x6)))) {
313 * PAT support with the lower four entries. Intel Pentium 2,
314 * 3, M, and 4 are affected by PAT errata, which makes the
315 * upper four entries unusable. To be on the safe side, we don't
323 * 000 0 WB : _PAGE_CACHE_MODE_WB
324 * 001 1 WC : _PAGE_CACHE_MODE_WC
325 * 010 2 UC-: _PAGE_CACHE_MODE_UC_MINUS
326 * 011 3 UC : _PAGE_CACHE_MODE_UC
329 * NOTE: When WT or WP is used, it is redirected to UC- per
330 * the default setup in __cachemode2pte_tbl[].
332 pat = PAT(0, WB) | PAT(1, WC) | PAT(2, UC_MINUS) | PAT(3, UC) |
333 PAT(4, WB) | PAT(5, WC) | PAT(6, UC_MINUS) | PAT(7, UC);
336 * Full PAT support. We put WT in slot 7 to improve
337 * robustness in the presence of errata that might cause
338 * the high PAT bit to be ignored. This way, a buggy slot 7
339 * access will hit slot 3, and slot 3 is UC, so at worst
340 * we lose performance without causing a correctness issue.
341 * Pentium 4 erratum N46 is an example for such an erratum,
342 * although we try not to use PAT at all on affected CPUs.
349 * 000 0 WB : _PAGE_CACHE_MODE_WB
350 * 001 1 WC : _PAGE_CACHE_MODE_WC
351 * 010 2 UC-: _PAGE_CACHE_MODE_UC_MINUS
352 * 011 3 UC : _PAGE_CACHE_MODE_UC
353 * 100 4 WB : Reserved
354 * 101 5 WC : Reserved
355 * 110 6 UC-: Reserved
356 * 111 7 WT : _PAGE_CACHE_MODE_WT
358 * The reserved slots are unused, but mapped to their
359 * corresponding types in the presence of PAT errata.
361 pat = PAT(0, WB) | PAT(1, WC) | PAT(2, UC_MINUS) | PAT(3, UC) |
362 PAT(4, WB) | PAT(5, WC) | PAT(6, UC_MINUS) | PAT(7, WT);
365 if (!boot_cpu_done) {
367 boot_cpu_done = true;
375 static DEFINE_SPINLOCK(memtype_lock); /* protects memtype accesses */
378 * Does intersection of PAT memory type and MTRR memory type and returns
379 * the resulting memory type as PAT understands it.
380 * (Type in pat and mtrr will not have same value)
381 * The intersection is based on "Effective Memory Type" tables in IA-32
384 static unsigned long pat_x_mtrr_type(u64 start, u64 end,
385 enum page_cache_mode req_type)
388 * Look for MTRR hint to get the effective type in case where PAT
391 if (req_type == _PAGE_CACHE_MODE_WB) {
392 u8 mtrr_type, uniform;
394 mtrr_type = mtrr_type_lookup(start, end, &uniform);
395 if (mtrr_type != MTRR_TYPE_WRBACK)
396 return _PAGE_CACHE_MODE_UC_MINUS;
398 return _PAGE_CACHE_MODE_WB;
404 struct pagerange_state {
405 unsigned long cur_pfn;
411 pagerange_is_ram_callback(unsigned long initial_pfn, unsigned long total_nr_pages, void *arg)
413 struct pagerange_state *state = arg;
415 state->not_ram |= initial_pfn > state->cur_pfn;
416 state->ram |= total_nr_pages > 0;
417 state->cur_pfn = initial_pfn + total_nr_pages;
419 return state->ram && state->not_ram;
422 static int pat_pagerange_is_ram(resource_size_t start, resource_size_t end)
425 unsigned long start_pfn = start >> PAGE_SHIFT;
426 unsigned long end_pfn = (end + PAGE_SIZE - 1) >> PAGE_SHIFT;
427 struct pagerange_state state = {start_pfn, 0, 0};
430 * For legacy reasons, physical address range in the legacy ISA
431 * region is tracked as non-RAM. This will allow users of
432 * /dev/mem to map portions of legacy ISA region, even when
433 * some of those portions are listed(or not even listed) with
434 * different e820 types(RAM/reserved/..)
436 if (start_pfn < ISA_END_ADDRESS >> PAGE_SHIFT)
437 start_pfn = ISA_END_ADDRESS >> PAGE_SHIFT;
439 if (start_pfn < end_pfn) {
440 ret = walk_system_ram_range(start_pfn, end_pfn - start_pfn,
441 &state, pagerange_is_ram_callback);
444 return (ret > 0) ? -1 : (state.ram ? 1 : 0);
448 * For RAM pages, we use page flags to mark the pages with appropriate type.
449 * The page flags are limited to four types, WB (default), WC, WT and UC-.
450 * WP request fails with -EINVAL, and UC gets redirected to UC-. Setting
451 * a new memory type is only allowed for a page mapped with the default WB
454 * Here we do two passes:
455 * - Find the memtype of all the pages in the range, look for any conflicts.
456 * - In case of no conflicts, set the new memtype for pages in the range.
458 static int reserve_ram_pages_type(u64 start, u64 end,
459 enum page_cache_mode req_type,
460 enum page_cache_mode *new_type)
465 if (req_type == _PAGE_CACHE_MODE_WP) {
467 *new_type = _PAGE_CACHE_MODE_UC_MINUS;
471 if (req_type == _PAGE_CACHE_MODE_UC) {
472 /* We do not support strong UC */
474 req_type = _PAGE_CACHE_MODE_UC_MINUS;
477 for (pfn = (start >> PAGE_SHIFT); pfn < (end >> PAGE_SHIFT); ++pfn) {
478 enum page_cache_mode type;
480 page = pfn_to_page(pfn);
481 type = get_page_memtype(page);
482 if (type != _PAGE_CACHE_MODE_WB) {
483 pr_info("x86/PAT: reserve_ram_pages_type failed [mem %#010Lx-%#010Lx], track 0x%x, req 0x%x\n",
484 start, end - 1, type, req_type);
493 *new_type = req_type;
495 for (pfn = (start >> PAGE_SHIFT); pfn < (end >> PAGE_SHIFT); ++pfn) {
496 page = pfn_to_page(pfn);
497 set_page_memtype(page, req_type);
502 static int free_ram_pages_type(u64 start, u64 end)
507 for (pfn = (start >> PAGE_SHIFT); pfn < (end >> PAGE_SHIFT); ++pfn) {
508 page = pfn_to_page(pfn);
509 set_page_memtype(page, _PAGE_CACHE_MODE_WB);
515 * req_type typically has one of the:
516 * - _PAGE_CACHE_MODE_WB
517 * - _PAGE_CACHE_MODE_WC
518 * - _PAGE_CACHE_MODE_UC_MINUS
519 * - _PAGE_CACHE_MODE_UC
520 * - _PAGE_CACHE_MODE_WT
522 * If new_type is NULL, function will return an error if it cannot reserve the
523 * region with req_type. If new_type is non-NULL, function will return
524 * available type in new_type in case of no error. In case of any error
525 * it will return a negative return value.
527 int reserve_memtype(u64 start, u64 end, enum page_cache_mode req_type,
528 enum page_cache_mode *new_type)
531 enum page_cache_mode actual_type;
535 BUG_ON(start >= end); /* end is exclusive */
537 if (!pat_enabled()) {
538 /* This is identical to page table setting without PAT */
540 *new_type = req_type;
544 /* Low ISA region is always mapped WB in page table. No need to track */
545 if (x86_platform.is_untracked_pat_range(start, end)) {
547 *new_type = _PAGE_CACHE_MODE_WB;
552 * Call mtrr_lookup to get the type hint. This is an
553 * optimization for /dev/mem mmap'ers into WB memory (BIOS
554 * tools and ACPI tools). Use WB request for WB memory and use
555 * UC_MINUS otherwise.
557 actual_type = pat_x_mtrr_type(start, end, req_type);
560 *new_type = actual_type;
562 is_range_ram = pat_pagerange_is_ram(start, end);
563 if (is_range_ram == 1) {
565 err = reserve_ram_pages_type(start, end, req_type, new_type);
568 } else if (is_range_ram < 0) {
572 new = kzalloc(sizeof(struct memtype), GFP_KERNEL);
578 new->type = actual_type;
580 spin_lock(&memtype_lock);
582 err = rbt_memtype_check_insert(new, new_type);
584 pr_info("x86/PAT: reserve_memtype failed [mem %#010Lx-%#010Lx], track %s, req %s\n",
586 cattr_name(new->type), cattr_name(req_type));
588 spin_unlock(&memtype_lock);
593 spin_unlock(&memtype_lock);
595 dprintk("reserve_memtype added [mem %#010Lx-%#010Lx], track %s, req %s, ret %s\n",
596 start, end - 1, cattr_name(new->type), cattr_name(req_type),
597 new_type ? cattr_name(*new_type) : "-");
602 int free_memtype(u64 start, u64 end)
606 struct memtype *entry;
611 /* Low ISA region is always mapped WB. No need to track */
612 if (x86_platform.is_untracked_pat_range(start, end))
615 is_range_ram = pat_pagerange_is_ram(start, end);
616 if (is_range_ram == 1) {
618 err = free_ram_pages_type(start, end);
621 } else if (is_range_ram < 0) {
625 spin_lock(&memtype_lock);
626 entry = rbt_memtype_erase(start, end);
627 spin_unlock(&memtype_lock);
630 pr_info("x86/PAT: %s:%d freeing invalid memtype [mem %#010Lx-%#010Lx]\n",
631 current->comm, current->pid, start, end - 1);
637 dprintk("free_memtype request [mem %#010Lx-%#010Lx]\n", start, end - 1);
644 * lookup_memtype - Looksup the memory type for a physical address
645 * @paddr: physical address of which memory type needs to be looked up
647 * Only to be called when PAT is enabled
649 * Returns _PAGE_CACHE_MODE_WB, _PAGE_CACHE_MODE_WC, _PAGE_CACHE_MODE_UC_MINUS
650 * or _PAGE_CACHE_MODE_WT.
652 static enum page_cache_mode lookup_memtype(u64 paddr)
654 enum page_cache_mode rettype = _PAGE_CACHE_MODE_WB;
655 struct memtype *entry;
657 if (x86_platform.is_untracked_pat_range(paddr, paddr + PAGE_SIZE))
660 if (pat_pagerange_is_ram(paddr, paddr + PAGE_SIZE)) {
663 page = pfn_to_page(paddr >> PAGE_SHIFT);
664 return get_page_memtype(page);
667 spin_lock(&memtype_lock);
669 entry = rbt_memtype_lookup(paddr);
671 rettype = entry->type;
673 rettype = _PAGE_CACHE_MODE_UC_MINUS;
675 spin_unlock(&memtype_lock);
680 * io_reserve_memtype - Request 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
683 * @type: A pointer to memtype, with requested type. On success, requested
684 * or any other compatible type that was available for the region is returned
686 * On success, returns 0
687 * On failure, returns non-zero
689 int io_reserve_memtype(resource_size_t start, resource_size_t end,
690 enum page_cache_mode *type)
692 resource_size_t size = end - start;
693 enum page_cache_mode req_type = *type;
694 enum page_cache_mode new_type;
697 WARN_ON_ONCE(iomem_map_sanity_check(start, size));
699 ret = reserve_memtype(start, end, req_type, &new_type);
703 if (!is_new_memtype_allowed(start, size, req_type, new_type))
706 if (kernel_map_sync_memtype(start, size, new_type) < 0)
713 free_memtype(start, end);
720 * io_free_memtype - Release a memory type mapping for a region of memory
721 * @start: start (physical address) of the region
722 * @end: end (physical address) of the region
724 void io_free_memtype(resource_size_t start, resource_size_t end)
726 free_memtype(start, end);
729 int arch_io_reserve_memtype_wc(resource_size_t start, resource_size_t size)
731 enum page_cache_mode type = _PAGE_CACHE_MODE_WC;
733 return io_reserve_memtype(start, start + size, &type);
735 EXPORT_SYMBOL(arch_io_reserve_memtype_wc);
737 void arch_io_free_memtype_wc(resource_size_t start, resource_size_t size)
739 io_free_memtype(start, start + size);
741 EXPORT_SYMBOL(arch_io_free_memtype_wc);
743 pgprot_t phys_mem_access_prot(struct file *file, unsigned long pfn,
744 unsigned long size, pgprot_t vma_prot)
749 #ifdef CONFIG_STRICT_DEVMEM
750 /* This check is done in drivers/char/mem.c in case of STRICT_DEVMEM */
751 static inline int range_is_allowed(unsigned long pfn, unsigned long size)
756 /* This check is needed to avoid cache aliasing when PAT is enabled */
757 static inline int range_is_allowed(unsigned long pfn, unsigned long size)
759 u64 from = ((u64)pfn) << PAGE_SHIFT;
760 u64 to = from + size;
766 while (cursor < to) {
767 if (!devmem_is_allowed(pfn))
774 #endif /* CONFIG_STRICT_DEVMEM */
776 int phys_mem_access_prot_allowed(struct file *file, unsigned long pfn,
777 unsigned long size, pgprot_t *vma_prot)
779 enum page_cache_mode pcm = _PAGE_CACHE_MODE_WB;
781 if (!range_is_allowed(pfn, size))
784 if (file->f_flags & O_DSYNC)
785 pcm = _PAGE_CACHE_MODE_UC_MINUS;
787 *vma_prot = __pgprot((pgprot_val(*vma_prot) & ~_PAGE_CACHE_MASK) |
788 cachemode2protval(pcm));
793 * Change the memory type for the physial address range in kernel identity
794 * mapping space if that range is a part of identity map.
796 int kernel_map_sync_memtype(u64 base, unsigned long size,
797 enum page_cache_mode pcm)
801 if (base > __pa(high_memory-1))
805 * some areas in the middle of the kernel identity range
806 * are not mapped, like the PCI space.
808 if (!page_is_ram(base >> PAGE_SHIFT))
811 id_sz = (__pa(high_memory-1) <= base + size) ?
812 __pa(high_memory) - base :
815 if (ioremap_change_attr((unsigned long)__va(base), id_sz, pcm) < 0) {
816 pr_info("x86/PAT: %s:%d ioremap_change_attr failed %s for [mem %#010Lx-%#010Lx]\n",
817 current->comm, current->pid,
819 base, (unsigned long long)(base + size-1));
826 * Internal interface to reserve a range of physical memory with prot.
827 * Reserved non RAM regions only and after successful reserve_memtype,
828 * this func also keeps identity mapping (if any) in sync with this new prot.
830 static int reserve_pfn_range(u64 paddr, unsigned long size, pgprot_t *vma_prot,
835 enum page_cache_mode want_pcm = pgprot2cachemode(*vma_prot);
836 enum page_cache_mode pcm = want_pcm;
838 is_ram = pat_pagerange_is_ram(paddr, paddr + size);
841 * reserve_pfn_range() for RAM pages. We do not refcount to keep
842 * track of number of mappings of RAM pages. We can assert that
843 * the type requested matches the type of first page in the range.
849 pcm = lookup_memtype(paddr);
850 if (want_pcm != pcm) {
851 pr_warn("x86/PAT: %s:%d map pfn RAM range req %s for [mem %#010Lx-%#010Lx], got %s\n",
852 current->comm, current->pid,
853 cattr_name(want_pcm),
854 (unsigned long long)paddr,
855 (unsigned long long)(paddr + size - 1),
857 *vma_prot = __pgprot((pgprot_val(*vma_prot) &
858 (~_PAGE_CACHE_MASK)) |
859 cachemode2protval(pcm));
864 ret = reserve_memtype(paddr, paddr + size, want_pcm, &pcm);
868 if (pcm != want_pcm) {
870 !is_new_memtype_allowed(paddr, size, want_pcm, pcm)) {
871 free_memtype(paddr, paddr + size);
872 pr_err("x86/PAT: %s:%d map pfn expected mapping type %s for [mem %#010Lx-%#010Lx], got %s\n",
873 current->comm, current->pid,
874 cattr_name(want_pcm),
875 (unsigned long long)paddr,
876 (unsigned long long)(paddr + size - 1),
881 * We allow returning different type than the one requested in
884 *vma_prot = __pgprot((pgprot_val(*vma_prot) &
885 (~_PAGE_CACHE_MASK)) |
886 cachemode2protval(pcm));
889 if (kernel_map_sync_memtype(paddr, size, pcm) < 0) {
890 free_memtype(paddr, paddr + size);
897 * Internal interface to free a range of physical memory.
898 * Frees non RAM regions only.
900 static void free_pfn_range(u64 paddr, unsigned long size)
904 is_ram = pat_pagerange_is_ram(paddr, paddr + size);
906 free_memtype(paddr, paddr + size);
910 * track_pfn_copy is called when vma that is covering the pfnmap gets
911 * copied through copy_page_range().
913 * If the vma has a linear pfn mapping for the entire range, we get the prot
914 * from pte and reserve the entire vma range with single reserve_pfn_range call.
916 int track_pfn_copy(struct vm_area_struct *vma)
918 resource_size_t paddr;
920 unsigned long vma_size = vma->vm_end - vma->vm_start;
923 if (vma->vm_flags & VM_PAT) {
925 * reserve the whole chunk covered by vma. We need the
926 * starting address and protection from pte.
928 if (follow_phys(vma, vma->vm_start, 0, &prot, &paddr)) {
932 pgprot = __pgprot(prot);
933 return reserve_pfn_range(paddr, vma_size, &pgprot, 1);
940 * prot is passed in as a parameter for the new mapping. If the vma has a
941 * linear pfn mapping for the entire range reserve the entire vma range with
942 * single reserve_pfn_range call.
944 int track_pfn_remap(struct vm_area_struct *vma, pgprot_t *prot,
945 unsigned long pfn, unsigned long addr, unsigned long size)
947 resource_size_t paddr = (resource_size_t)pfn << PAGE_SHIFT;
948 enum page_cache_mode pcm;
950 /* reserve the whole chunk starting from paddr */
951 if (addr == vma->vm_start && size == (vma->vm_end - vma->vm_start)) {
954 ret = reserve_pfn_range(paddr, size, prot, 0);
956 vma->vm_flags |= VM_PAT;
964 * For anything smaller than the vma size we set prot based on the
967 pcm = lookup_memtype(paddr);
969 /* Check memtype for the remaining pages */
970 while (size > PAGE_SIZE) {
973 if (pcm != lookup_memtype(paddr))
977 *prot = __pgprot((pgprot_val(vma->vm_page_prot) & (~_PAGE_CACHE_MASK)) |
978 cachemode2protval(pcm));
983 int track_pfn_insert(struct vm_area_struct *vma, pgprot_t *prot,
986 enum page_cache_mode pcm;
991 /* Set prot based on lookup */
992 pcm = lookup_memtype((resource_size_t)pfn << PAGE_SHIFT);
993 *prot = __pgprot((pgprot_val(vma->vm_page_prot) & (~_PAGE_CACHE_MASK)) |
994 cachemode2protval(pcm));
1000 * untrack_pfn is called while unmapping a pfnmap for a region.
1001 * untrack can be called for a specific region indicated by pfn and size or
1002 * can be for the entire vma (in which case pfn, size are zero).
1004 void untrack_pfn(struct vm_area_struct *vma, unsigned long pfn,
1007 resource_size_t paddr;
1010 if (!(vma->vm_flags & VM_PAT))
1013 /* free the chunk starting from pfn or the whole chunk */
1014 paddr = (resource_size_t)pfn << PAGE_SHIFT;
1015 if (!paddr && !size) {
1016 if (follow_phys(vma, vma->vm_start, 0, &prot, &paddr)) {
1021 size = vma->vm_end - vma->vm_start;
1023 free_pfn_range(paddr, size);
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 */