1 // SPDX-License-Identifier: GPL-2.0-only
3 * Handle caching attributes in page tables (PAT)
5 * Authors: Venkatesh Pallipadi <venkatesh.pallipadi@intel.com>
6 * Suresh B Siddha <suresh.b.siddha@intel.com>
8 * Loosely based on earlier PAT patchset from Eric Biederman and Andi Kleen.
11 #include <linux/seq_file.h>
12 #include <linux/memblock.h>
13 #include <linux/debugfs.h>
14 #include <linux/ioport.h>
15 #include <linux/kernel.h>
16 #include <linux/pfn_t.h>
17 #include <linux/slab.h>
20 #include <linux/rbtree.h>
22 #include <asm/cacheflush.h>
23 #include <asm/processor.h>
24 #include <asm/tlbflush.h>
25 #include <asm/x86_init.h>
26 #include <asm/pgtable.h>
27 #include <asm/fcntl.h>
28 #include <asm/e820/api.h>
35 #include "pat_internal.h"
36 #include "mm_internal.h"
37 #include "../../mm/internal.h" /* is_cow_mapping() */
40 #define pr_fmt(fmt) "" fmt
42 static bool __read_mostly boot_cpu_done;
43 static bool __read_mostly pat_disabled = !IS_ENABLED(CONFIG_X86_PAT);
44 static bool __read_mostly pat_initialized;
45 static bool __read_mostly init_cm_done;
47 void pat_disable(const char *reason)
53 WARN_ONCE(1, "x86/PAT: PAT cannot be disabled after initialization\n");
58 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 bool pat_enabled(void)
70 return pat_initialized;
72 EXPORT_SYMBOL_GPL(pat_enabled);
76 static int __init pat_debug_setup(char *str)
81 __setup("debugpat", pat_debug_setup);
85 * X86 PAT uses page flags arch_1 and uncached together to keep track of
86 * memory type of pages that have backing page struct.
88 * X86 PAT supports 4 different memory types:
89 * - _PAGE_CACHE_MODE_WB
90 * - _PAGE_CACHE_MODE_WC
91 * - _PAGE_CACHE_MODE_UC_MINUS
92 * - _PAGE_CACHE_MODE_WT
94 * _PAGE_CACHE_MODE_WB is the default type.
98 #define _PGMT_WC (1UL << PG_arch_1)
99 #define _PGMT_UC_MINUS (1UL << PG_uncached)
100 #define _PGMT_WT (1UL << PG_uncached | 1UL << PG_arch_1)
101 #define _PGMT_MASK (1UL << PG_uncached | 1UL << PG_arch_1)
102 #define _PGMT_CLEAR_MASK (~_PGMT_MASK)
104 static inline enum page_cache_mode get_page_memtype(struct page *pg)
106 unsigned long pg_flags = pg->flags & _PGMT_MASK;
108 if (pg_flags == _PGMT_WB)
109 return _PAGE_CACHE_MODE_WB;
110 else if (pg_flags == _PGMT_WC)
111 return _PAGE_CACHE_MODE_WC;
112 else if (pg_flags == _PGMT_UC_MINUS)
113 return _PAGE_CACHE_MODE_UC_MINUS;
115 return _PAGE_CACHE_MODE_WT;
118 static inline void set_page_memtype(struct page *pg,
119 enum page_cache_mode memtype)
121 unsigned long memtype_flags;
122 unsigned long old_flags;
123 unsigned long new_flags;
126 case _PAGE_CACHE_MODE_WC:
127 memtype_flags = _PGMT_WC;
129 case _PAGE_CACHE_MODE_UC_MINUS:
130 memtype_flags = _PGMT_UC_MINUS;
132 case _PAGE_CACHE_MODE_WT:
133 memtype_flags = _PGMT_WT;
135 case _PAGE_CACHE_MODE_WB:
137 memtype_flags = _PGMT_WB;
142 old_flags = pg->flags;
143 new_flags = (old_flags & _PGMT_CLEAR_MASK) | memtype_flags;
144 } while (cmpxchg(&pg->flags, old_flags, new_flags) != old_flags);
147 static inline enum page_cache_mode get_page_memtype(struct page *pg)
151 static inline void set_page_memtype(struct page *pg,
152 enum page_cache_mode memtype)
158 PAT_UC = 0, /* uncached */
159 PAT_WC = 1, /* Write combining */
160 PAT_WT = 4, /* Write Through */
161 PAT_WP = 5, /* Write Protected */
162 PAT_WB = 6, /* Write Back (default) */
163 PAT_UC_MINUS = 7, /* UC, but can be overridden by MTRR */
166 #define CM(c) (_PAGE_CACHE_MODE_ ## c)
168 static enum page_cache_mode pat_get_cache_mode(unsigned pat_val, char *msg)
170 enum page_cache_mode cache;
174 case PAT_UC: cache = CM(UC); cache_mode = "UC "; break;
175 case PAT_WC: cache = CM(WC); cache_mode = "WC "; break;
176 case PAT_WT: cache = CM(WT); cache_mode = "WT "; break;
177 case PAT_WP: cache = CM(WP); cache_mode = "WP "; break;
178 case PAT_WB: cache = CM(WB); cache_mode = "WB "; break;
179 case PAT_UC_MINUS: cache = CM(UC_MINUS); cache_mode = "UC- "; break;
180 default: cache = CM(WB); cache_mode = "WB "; break;
183 memcpy(msg, cache_mode, 4);
191 * Update the cache mode to pgprot translation tables according to PAT
193 * Using lower indices is preferred, so we start with highest index.
195 static void __init_cache_modes(u64 pat)
197 enum page_cache_mode cache;
202 for (i = 7; i >= 0; i--) {
203 cache = pat_get_cache_mode((pat >> (i * 8)) & 7,
205 update_cache_mode_entry(i, cache);
207 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 (!boot_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 void init_cache_modes(void)
255 if (boot_cpu_has(X86_FEATURE_PAT)) {
257 * CPU supports PAT. Set PAT table to be consistent with
258 * PAT MSR. This case supports "nopat" boot option, and
259 * virtual machine environments which support PAT without
260 * MTRRs. In specific, Xen has unique setup to PAT MSR.
262 * If PAT MSR returns 0, it is considered invalid and emulates
265 rdmsrl(MSR_IA32_CR_PAT, pat);
270 * No PAT. Emulate the PAT table that corresponds to the two
271 * cache bits, PWT (Write Through) and PCD (Cache Disable).
272 * This setup is also the same as the BIOS default setup.
279 * 00 0 WB : _PAGE_CACHE_MODE_WB
280 * 01 1 WT : _PAGE_CACHE_MODE_WT
281 * 10 2 UC-: _PAGE_CACHE_MODE_UC_MINUS
282 * 11 3 UC : _PAGE_CACHE_MODE_UC
284 * NOTE: When WC or WP is used, it is redirected to UC- per
285 * the default setup in __cachemode2pte_tbl[].
287 pat = PAT(0, WB) | PAT(1, WT) | PAT(2, UC_MINUS) | PAT(3, UC) |
288 PAT(4, WB) | PAT(5, WT) | PAT(6, UC_MINUS) | PAT(7, UC);
291 __init_cache_modes(pat);
295 * pat_init - Initialize PAT MSR and PAT table
297 * This function initializes PAT MSR and PAT table with an OS-defined value
298 * to enable additional cache attributes, WC, WT and WP.
300 * This function must be called on all CPUs using the specific sequence of
301 * operations defined in Intel SDM. mtrr_rendezvous_handler() provides this
307 struct cpuinfo_x86 *c = &boot_cpu_data;
312 if ((c->x86_vendor == X86_VENDOR_INTEL) &&
313 (((c->x86 == 0x6) && (c->x86_model <= 0xd)) ||
314 ((c->x86 == 0xf) && (c->x86_model <= 0x6)))) {
316 * PAT support with the lower four entries. Intel Pentium 2,
317 * 3, M, and 4 are affected by PAT errata, which makes the
318 * upper four entries unusable. To be on the safe side, we don't
326 * 000 0 WB : _PAGE_CACHE_MODE_WB
327 * 001 1 WC : _PAGE_CACHE_MODE_WC
328 * 010 2 UC-: _PAGE_CACHE_MODE_UC_MINUS
329 * 011 3 UC : _PAGE_CACHE_MODE_UC
332 * NOTE: When WT or WP is used, it is redirected to UC- per
333 * the default setup in __cachemode2pte_tbl[].
335 pat = PAT(0, WB) | PAT(1, WC) | PAT(2, UC_MINUS) | PAT(3, UC) |
336 PAT(4, WB) | PAT(5, WC) | PAT(6, UC_MINUS) | PAT(7, UC);
339 * Full PAT support. We put WT in slot 7 to improve
340 * robustness in the presence of errata that might cause
341 * the high PAT bit to be ignored. This way, a buggy slot 7
342 * access will hit slot 3, and slot 3 is UC, so at worst
343 * we lose performance without causing a correctness issue.
344 * Pentium 4 erratum N46 is an example for such an erratum,
345 * although we try not to use PAT at all on affected CPUs.
352 * 000 0 WB : _PAGE_CACHE_MODE_WB
353 * 001 1 WC : _PAGE_CACHE_MODE_WC
354 * 010 2 UC-: _PAGE_CACHE_MODE_UC_MINUS
355 * 011 3 UC : _PAGE_CACHE_MODE_UC
356 * 100 4 WB : Reserved
357 * 101 5 WP : _PAGE_CACHE_MODE_WP
358 * 110 6 UC-: Reserved
359 * 111 7 WT : _PAGE_CACHE_MODE_WT
361 * The reserved slots are unused, but mapped to their
362 * corresponding types in the presence of PAT errata.
364 pat = PAT(0, WB) | PAT(1, WC) | PAT(2, UC_MINUS) | PAT(3, UC) |
365 PAT(4, WB) | PAT(5, WP) | PAT(6, UC_MINUS) | PAT(7, WT);
368 if (!boot_cpu_done) {
370 boot_cpu_done = true;
378 static DEFINE_SPINLOCK(memtype_lock); /* protects memtype accesses */
381 * Does intersection of PAT memory type and MTRR memory type and returns
382 * the resulting memory type as PAT understands it.
383 * (Type in pat and mtrr will not have same value)
384 * The intersection is based on "Effective Memory Type" tables in IA-32
387 static unsigned long pat_x_mtrr_type(u64 start, u64 end,
388 enum page_cache_mode req_type)
391 * Look for MTRR hint to get the effective type in case where PAT
394 if (req_type == _PAGE_CACHE_MODE_WB) {
395 u8 mtrr_type, uniform;
397 mtrr_type = mtrr_type_lookup(start, end, &uniform);
398 if (mtrr_type != MTRR_TYPE_WRBACK)
399 return _PAGE_CACHE_MODE_UC_MINUS;
401 return _PAGE_CACHE_MODE_WB;
407 struct pagerange_state {
408 unsigned long cur_pfn;
414 pagerange_is_ram_callback(unsigned long initial_pfn, unsigned long total_nr_pages, void *arg)
416 struct pagerange_state *state = arg;
418 state->not_ram |= initial_pfn > state->cur_pfn;
419 state->ram |= total_nr_pages > 0;
420 state->cur_pfn = initial_pfn + total_nr_pages;
422 return state->ram && state->not_ram;
425 static int pat_pagerange_is_ram(resource_size_t start, resource_size_t end)
428 unsigned long start_pfn = start >> PAGE_SHIFT;
429 unsigned long end_pfn = (end + PAGE_SIZE - 1) >> PAGE_SHIFT;
430 struct pagerange_state state = {start_pfn, 0, 0};
433 * For legacy reasons, physical address range in the legacy ISA
434 * region is tracked as non-RAM. This will allow users of
435 * /dev/mem to map portions of legacy ISA region, even when
436 * some of those portions are listed(or not even listed) with
437 * different e820 types(RAM/reserved/..)
439 if (start_pfn < ISA_END_ADDRESS >> PAGE_SHIFT)
440 start_pfn = ISA_END_ADDRESS >> PAGE_SHIFT;
442 if (start_pfn < end_pfn) {
443 ret = walk_system_ram_range(start_pfn, end_pfn - start_pfn,
444 &state, pagerange_is_ram_callback);
447 return (ret > 0) ? -1 : (state.ram ? 1 : 0);
451 * For RAM pages, we use page flags to mark the pages with appropriate type.
452 * The page flags are limited to four types, WB (default), WC, WT and UC-.
453 * WP request fails with -EINVAL, and UC gets redirected to UC-. Setting
454 * a new memory type is only allowed for a page mapped with the default WB
457 * Here we do two passes:
458 * - Find the memtype of all the pages in the range, look for any conflicts.
459 * - In case of no conflicts, set the new memtype for pages in the range.
461 static int reserve_ram_pages_type(u64 start, u64 end,
462 enum page_cache_mode req_type,
463 enum page_cache_mode *new_type)
468 if (req_type == _PAGE_CACHE_MODE_WP) {
470 *new_type = _PAGE_CACHE_MODE_UC_MINUS;
474 if (req_type == _PAGE_CACHE_MODE_UC) {
475 /* We do not support strong UC */
477 req_type = _PAGE_CACHE_MODE_UC_MINUS;
480 for (pfn = (start >> PAGE_SHIFT); pfn < (end >> PAGE_SHIFT); ++pfn) {
481 enum page_cache_mode type;
483 page = pfn_to_page(pfn);
484 type = get_page_memtype(page);
485 if (type != _PAGE_CACHE_MODE_WB) {
486 pr_info("x86/PAT: reserve_ram_pages_type failed [mem %#010Lx-%#010Lx], track 0x%x, req 0x%x\n",
487 start, end - 1, type, req_type);
496 *new_type = req_type;
498 for (pfn = (start >> PAGE_SHIFT); pfn < (end >> PAGE_SHIFT); ++pfn) {
499 page = pfn_to_page(pfn);
500 set_page_memtype(page, req_type);
505 static int free_ram_pages_type(u64 start, u64 end)
510 for (pfn = (start >> PAGE_SHIFT); pfn < (end >> PAGE_SHIFT); ++pfn) {
511 page = pfn_to_page(pfn);
512 set_page_memtype(page, _PAGE_CACHE_MODE_WB);
517 static u64 sanitize_phys(u64 address)
520 * When changing the memtype for pages containing poison allow
521 * for a "decoy" virtual address (bit 63 clear) passed to
522 * set_memory_X(). __pa() on a "decoy" address results in a
523 * physical address with bit 63 set.
525 * Decoy addresses are not present for 32-bit builds, see
528 if (IS_ENABLED(CONFIG_X86_64))
529 return address & __PHYSICAL_MASK;
534 * req_type typically has one of the:
535 * - _PAGE_CACHE_MODE_WB
536 * - _PAGE_CACHE_MODE_WC
537 * - _PAGE_CACHE_MODE_UC_MINUS
538 * - _PAGE_CACHE_MODE_UC
539 * - _PAGE_CACHE_MODE_WT
541 * If new_type is NULL, function will return an error if it cannot reserve the
542 * region with req_type. If new_type is non-NULL, function will return
543 * available type in new_type in case of no error. In case of any error
544 * it will return a negative return value.
546 int reserve_memtype(u64 start, u64 end, enum page_cache_mode req_type,
547 enum page_cache_mode *new_type)
550 enum page_cache_mode actual_type;
554 start = sanitize_phys(start);
555 end = sanitize_phys(end);
557 WARN(1, "%s failed: [mem %#010Lx-%#010Lx], req %s\n", __func__,
558 start, end - 1, cattr_name(req_type));
562 if (!pat_enabled()) {
563 /* This is identical to page table setting without PAT */
565 *new_type = req_type;
569 /* Low ISA region is always mapped WB in page table. No need to track */
570 if (x86_platform.is_untracked_pat_range(start, end)) {
572 *new_type = _PAGE_CACHE_MODE_WB;
577 * Call mtrr_lookup to get the type hint. This is an
578 * optimization for /dev/mem mmap'ers into WB memory (BIOS
579 * tools and ACPI tools). Use WB request for WB memory and use
580 * UC_MINUS otherwise.
582 actual_type = pat_x_mtrr_type(start, end, req_type);
585 *new_type = actual_type;
587 is_range_ram = pat_pagerange_is_ram(start, end);
588 if (is_range_ram == 1) {
590 err = reserve_ram_pages_type(start, end, req_type, new_type);
593 } else if (is_range_ram < 0) {
597 new = kzalloc(sizeof(struct memtype), GFP_KERNEL);
603 new->type = actual_type;
605 spin_lock(&memtype_lock);
607 err = rbt_memtype_check_insert(new, new_type);
609 pr_info("x86/PAT: reserve_memtype failed [mem %#010Lx-%#010Lx], track %s, req %s\n",
611 cattr_name(new->type), cattr_name(req_type));
613 spin_unlock(&memtype_lock);
618 spin_unlock(&memtype_lock);
620 dprintk("reserve_memtype added [mem %#010Lx-%#010Lx], track %s, req %s, ret %s\n",
621 start, end - 1, cattr_name(new->type), cattr_name(req_type),
622 new_type ? cattr_name(*new_type) : "-");
627 int free_memtype(u64 start, u64 end)
631 struct memtype *entry;
636 start = sanitize_phys(start);
637 end = sanitize_phys(end);
639 /* Low ISA region is always mapped WB. No need to track */
640 if (x86_platform.is_untracked_pat_range(start, end))
643 is_range_ram = pat_pagerange_is_ram(start, end);
644 if (is_range_ram == 1) {
646 err = free_ram_pages_type(start, end);
649 } else if (is_range_ram < 0) {
653 spin_lock(&memtype_lock);
654 entry = rbt_memtype_erase(start, end);
655 spin_unlock(&memtype_lock);
658 pr_info("x86/PAT: %s:%d freeing invalid memtype [mem %#010Lx-%#010Lx]\n",
659 current->comm, current->pid, start, end - 1);
665 dprintk("free_memtype request [mem %#010Lx-%#010Lx]\n", start, end - 1);
672 * lookup_memtype - Looksup the memory type for a physical address
673 * @paddr: physical address of which memory type needs to be looked up
675 * Only to be called when PAT is enabled
677 * Returns _PAGE_CACHE_MODE_WB, _PAGE_CACHE_MODE_WC, _PAGE_CACHE_MODE_UC_MINUS
678 * or _PAGE_CACHE_MODE_WT.
680 static enum page_cache_mode lookup_memtype(u64 paddr)
682 enum page_cache_mode rettype = _PAGE_CACHE_MODE_WB;
683 struct memtype *entry;
685 if (x86_platform.is_untracked_pat_range(paddr, paddr + PAGE_SIZE))
688 if (pat_pagerange_is_ram(paddr, paddr + PAGE_SIZE)) {
691 page = pfn_to_page(paddr >> PAGE_SHIFT);
692 return get_page_memtype(page);
695 spin_lock(&memtype_lock);
697 entry = rbt_memtype_lookup(paddr);
699 rettype = entry->type;
701 rettype = _PAGE_CACHE_MODE_UC_MINUS;
703 spin_unlock(&memtype_lock);
708 * pat_pfn_immune_to_uc_mtrr - Check whether the PAT memory type
709 * of @pfn cannot be overridden by UC MTRR memory type.
711 * Only to be called when PAT is enabled.
713 * Returns true, if the PAT memory type of @pfn is UC, UC-, or WC.
714 * Returns false in other cases.
716 bool pat_pfn_immune_to_uc_mtrr(unsigned long pfn)
718 enum page_cache_mode cm = lookup_memtype(PFN_PHYS(pfn));
720 return cm == _PAGE_CACHE_MODE_UC ||
721 cm == _PAGE_CACHE_MODE_UC_MINUS ||
722 cm == _PAGE_CACHE_MODE_WC;
724 EXPORT_SYMBOL_GPL(pat_pfn_immune_to_uc_mtrr);
727 * io_reserve_memtype - Request a memory type mapping for a region of memory
728 * @start: start (physical address) of the region
729 * @end: end (physical address) of the region
730 * @type: A pointer to memtype, with requested type. On success, requested
731 * or any other compatible type that was available for the region is returned
733 * On success, returns 0
734 * On failure, returns non-zero
736 int io_reserve_memtype(resource_size_t start, resource_size_t end,
737 enum page_cache_mode *type)
739 resource_size_t size = end - start;
740 enum page_cache_mode req_type = *type;
741 enum page_cache_mode new_type;
744 WARN_ON_ONCE(iomem_map_sanity_check(start, size));
746 ret = reserve_memtype(start, end, req_type, &new_type);
750 if (!is_new_memtype_allowed(start, size, req_type, new_type))
753 if (kernel_map_sync_memtype(start, size, new_type) < 0)
760 free_memtype(start, end);
767 * io_free_memtype - Release a memory type mapping for a region of memory
768 * @start: start (physical address) of the region
769 * @end: end (physical address) of the region
771 void io_free_memtype(resource_size_t start, resource_size_t end)
773 free_memtype(start, end);
776 int arch_io_reserve_memtype_wc(resource_size_t start, resource_size_t size)
778 enum page_cache_mode type = _PAGE_CACHE_MODE_WC;
780 return io_reserve_memtype(start, start + size, &type);
782 EXPORT_SYMBOL(arch_io_reserve_memtype_wc);
784 void arch_io_free_memtype_wc(resource_size_t start, resource_size_t size)
786 io_free_memtype(start, start + size);
788 EXPORT_SYMBOL(arch_io_free_memtype_wc);
790 pgprot_t phys_mem_access_prot(struct file *file, unsigned long pfn,
791 unsigned long size, pgprot_t vma_prot)
793 if (!phys_mem_access_encrypted(pfn << PAGE_SHIFT, size))
794 vma_prot = pgprot_decrypted(vma_prot);
799 #ifdef CONFIG_STRICT_DEVMEM
800 /* This check is done in drivers/char/mem.c in case of STRICT_DEVMEM */
801 static inline int range_is_allowed(unsigned long pfn, unsigned long size)
806 /* This check is needed to avoid cache aliasing when PAT is enabled */
807 static inline int range_is_allowed(unsigned long pfn, unsigned long size)
809 u64 from = ((u64)pfn) << PAGE_SHIFT;
810 u64 to = from + size;
816 while (cursor < to) {
817 if (!devmem_is_allowed(pfn))
824 #endif /* CONFIG_STRICT_DEVMEM */
826 int phys_mem_access_prot_allowed(struct file *file, unsigned long pfn,
827 unsigned long size, pgprot_t *vma_prot)
829 enum page_cache_mode pcm = _PAGE_CACHE_MODE_WB;
831 if (!range_is_allowed(pfn, size))
834 if (file->f_flags & O_DSYNC)
835 pcm = _PAGE_CACHE_MODE_UC_MINUS;
837 *vma_prot = __pgprot((pgprot_val(*vma_prot) & ~_PAGE_CACHE_MASK) |
838 cachemode2protval(pcm));
843 * Change the memory type for the physial address range in kernel identity
844 * mapping space if that range is a part of identity map.
846 int kernel_map_sync_memtype(u64 base, unsigned long size,
847 enum page_cache_mode pcm)
851 if (base > __pa(high_memory-1))
855 * some areas in the middle of the kernel identity range
856 * are not mapped, like the PCI space.
858 if (!page_is_ram(base >> PAGE_SHIFT))
861 id_sz = (__pa(high_memory-1) <= base + size) ?
862 __pa(high_memory) - base :
865 if (ioremap_change_attr((unsigned long)__va(base), id_sz, pcm) < 0) {
866 pr_info("x86/PAT: %s:%d ioremap_change_attr failed %s for [mem %#010Lx-%#010Lx]\n",
867 current->comm, current->pid,
869 base, (unsigned long long)(base + size-1));
876 * Internal interface to reserve a range of physical memory with prot.
877 * Reserved non RAM regions only and after successful reserve_memtype,
878 * this func also keeps identity mapping (if any) in sync with this new prot.
880 static int reserve_pfn_range(u64 paddr, unsigned long size, pgprot_t *vma_prot,
885 enum page_cache_mode want_pcm = pgprot2cachemode(*vma_prot);
886 enum page_cache_mode pcm = want_pcm;
888 is_ram = pat_pagerange_is_ram(paddr, paddr + size);
891 * reserve_pfn_range() for RAM pages. We do not refcount to keep
892 * track of number of mappings of RAM pages. We can assert that
893 * the type requested matches the type of first page in the range.
899 pcm = lookup_memtype(paddr);
900 if (want_pcm != pcm) {
901 pr_warn("x86/PAT: %s:%d map pfn RAM range req %s for [mem %#010Lx-%#010Lx], got %s\n",
902 current->comm, current->pid,
903 cattr_name(want_pcm),
904 (unsigned long long)paddr,
905 (unsigned long long)(paddr + size - 1),
907 *vma_prot = __pgprot((pgprot_val(*vma_prot) &
908 (~_PAGE_CACHE_MASK)) |
909 cachemode2protval(pcm));
914 ret = reserve_memtype(paddr, paddr + size, want_pcm, &pcm);
918 if (pcm != want_pcm) {
920 !is_new_memtype_allowed(paddr, size, want_pcm, pcm)) {
921 free_memtype(paddr, paddr + size);
922 pr_err("x86/PAT: %s:%d map pfn expected mapping type %s for [mem %#010Lx-%#010Lx], got %s\n",
923 current->comm, current->pid,
924 cattr_name(want_pcm),
925 (unsigned long long)paddr,
926 (unsigned long long)(paddr + size - 1),
931 * We allow returning different type than the one requested in
934 *vma_prot = __pgprot((pgprot_val(*vma_prot) &
935 (~_PAGE_CACHE_MASK)) |
936 cachemode2protval(pcm));
939 if (kernel_map_sync_memtype(paddr, size, pcm) < 0) {
940 free_memtype(paddr, paddr + size);
947 * Internal interface to free a range of physical memory.
948 * Frees non RAM regions only.
950 static void free_pfn_range(u64 paddr, unsigned long size)
954 is_ram = pat_pagerange_is_ram(paddr, paddr + size);
956 free_memtype(paddr, paddr + size);
959 static int get_pat_info(struct vm_area_struct *vma, resource_size_t *paddr,
964 VM_WARN_ON_ONCE(!(vma->vm_flags & VM_PAT));
967 * We need the starting PFN and cachemode used for track_pfn_remap()
968 * that covered the whole VMA. For most mappings, we can obtain that
969 * information from the page tables. For COW mappings, we might now
970 * suddenly have anon folios mapped and follow_phys() will fail.
972 * Fallback to using vma->vm_pgoff, see remap_pfn_range_notrack(), to
973 * detect the PFN. If we need the cachemode as well, we're out of luck
974 * for now and have to fail fork().
976 if (!follow_phys(vma, vma->vm_start, 0, &prot, paddr)) {
978 *pgprot = __pgprot(prot);
981 if (is_cow_mapping(vma->vm_flags)) {
984 *paddr = (resource_size_t)vma->vm_pgoff << PAGE_SHIFT;
992 * track_pfn_copy is called when vma that is covering the pfnmap gets
993 * copied through copy_page_range().
995 * If the vma has a linear pfn mapping for the entire range, we get the prot
996 * from pte and reserve the entire vma range with single reserve_pfn_range call.
998 int track_pfn_copy(struct vm_area_struct *vma)
1000 resource_size_t paddr;
1001 unsigned long vma_size = vma->vm_end - vma->vm_start;
1004 if (vma->vm_flags & VM_PAT) {
1005 if (get_pat_info(vma, &paddr, &pgprot))
1007 /* reserve the whole chunk covered by vma. */
1008 return reserve_pfn_range(paddr, vma_size, &pgprot, 1);
1015 * prot is passed in as a parameter for the new mapping. If the vma has
1016 * a linear pfn mapping for the entire range, or no vma is provided,
1017 * reserve the entire pfn + size range with single reserve_pfn_range
1020 int track_pfn_remap(struct vm_area_struct *vma, pgprot_t *prot,
1021 unsigned long pfn, unsigned long addr, unsigned long size)
1023 resource_size_t paddr = (resource_size_t)pfn << PAGE_SHIFT;
1024 enum page_cache_mode pcm;
1026 /* reserve the whole chunk starting from paddr */
1027 if (!vma || (addr == vma->vm_start
1028 && size == (vma->vm_end - vma->vm_start))) {
1031 ret = reserve_pfn_range(paddr, size, prot, 0);
1032 if (ret == 0 && vma)
1033 vma->vm_flags |= VM_PAT;
1041 * For anything smaller than the vma size we set prot based on the
1044 pcm = lookup_memtype(paddr);
1046 /* Check memtype for the remaining pages */
1047 while (size > PAGE_SIZE) {
1050 if (pcm != lookup_memtype(paddr))
1054 *prot = __pgprot((pgprot_val(*prot) & (~_PAGE_CACHE_MASK)) |
1055 cachemode2protval(pcm));
1060 void track_pfn_insert(struct vm_area_struct *vma, pgprot_t *prot, pfn_t pfn)
1062 enum page_cache_mode pcm;
1067 /* Set prot based on lookup */
1068 pcm = lookup_memtype(pfn_t_to_phys(pfn));
1069 *prot = __pgprot((pgprot_val(*prot) & (~_PAGE_CACHE_MASK)) |
1070 cachemode2protval(pcm));
1074 * untrack_pfn is called while unmapping a pfnmap for a region.
1075 * untrack can be called for a specific region indicated by pfn and size or
1076 * can be for the entire vma (in which case pfn, size are zero).
1078 void untrack_pfn(struct vm_area_struct *vma, unsigned long pfn,
1081 resource_size_t paddr;
1083 if (vma && !(vma->vm_flags & VM_PAT))
1086 /* free the chunk starting from pfn or the whole chunk */
1087 paddr = (resource_size_t)pfn << PAGE_SHIFT;
1088 if (!paddr && !size) {
1089 if (get_pat_info(vma, &paddr, NULL))
1091 size = vma->vm_end - vma->vm_start;
1093 free_pfn_range(paddr, size);
1095 vma->vm_flags &= ~VM_PAT;
1099 * untrack_pfn_moved is called, while mremapping a pfnmap for a new region,
1100 * with the old vma after its pfnmap page table has been removed. The new
1101 * vma has a new pfnmap to the same pfn & cache type with VM_PAT set.
1103 void untrack_pfn_moved(struct vm_area_struct *vma)
1105 vma->vm_flags &= ~VM_PAT;
1108 pgprot_t pgprot_writecombine(pgprot_t prot)
1110 return __pgprot(pgprot_val(prot) |
1111 cachemode2protval(_PAGE_CACHE_MODE_WC));
1113 EXPORT_SYMBOL_GPL(pgprot_writecombine);
1115 pgprot_t pgprot_writethrough(pgprot_t prot)
1117 return __pgprot(pgprot_val(prot) |
1118 cachemode2protval(_PAGE_CACHE_MODE_WT));
1120 EXPORT_SYMBOL_GPL(pgprot_writethrough);
1122 #if defined(CONFIG_DEBUG_FS) && defined(CONFIG_X86_PAT)
1124 static struct memtype *memtype_get_idx(loff_t pos)
1126 struct memtype *print_entry;
1129 print_entry = kzalloc(sizeof(struct memtype), GFP_KERNEL);
1133 spin_lock(&memtype_lock);
1134 ret = rbt_memtype_copy_nth_element(print_entry, pos);
1135 spin_unlock(&memtype_lock);
1145 static void *memtype_seq_start(struct seq_file *seq, loff_t *pos)
1149 seq_puts(seq, "PAT memtype list:\n");
1152 return memtype_get_idx(*pos);
1155 static void *memtype_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1159 return memtype_get_idx(*pos);
1162 static void memtype_seq_stop(struct seq_file *seq, void *v)
1167 static int memtype_seq_show(struct seq_file *seq, void *v)
1169 struct memtype *print_entry = (struct memtype *)v;
1171 seq_printf(seq, "%s @ 0x%Lx-0x%Lx\n", cattr_name(print_entry->type),
1172 print_entry->start, print_entry->end);
1177 static const struct seq_operations memtype_seq_ops = {
1178 .start = memtype_seq_start,
1179 .next = memtype_seq_next,
1180 .stop = memtype_seq_stop,
1181 .show = memtype_seq_show,
1184 static int memtype_seq_open(struct inode *inode, struct file *file)
1186 return seq_open(file, &memtype_seq_ops);
1189 static const struct file_operations memtype_fops = {
1190 .open = memtype_seq_open,
1192 .llseek = seq_lseek,
1193 .release = seq_release,
1196 static int __init pat_memtype_list_init(void)
1198 if (pat_enabled()) {
1199 debugfs_create_file("pat_memtype_list", S_IRUSR,
1200 arch_debugfs_dir, NULL, &memtype_fops);
1205 late_initcall(pat_memtype_list_init);
1207 #endif /* CONFIG_DEBUG_FS && CONFIG_X86_PAT */