diff options
Diffstat (limited to 'arch/x86/mm')
-rw-r--r-- | arch/x86/mm/Makefile | 6 | ||||
-rw-r--r-- | arch/x86/mm/fault.c | 51 | ||||
-rw-r--r-- | arch/x86/mm/highmem_32.c | 3 | ||||
-rw-r--r-- | arch/x86/mm/init_64.c | 2 | ||||
-rw-r--r-- | arch/x86/mm/iomap_32.c | 27 | ||||
-rw-r--r-- | arch/x86/mm/ioremap.c | 90 | ||||
-rw-r--r-- | arch/x86/mm/kmemcheck/kmemcheck.c | 14 | ||||
-rw-r--r-- | arch/x86/mm/pageattr.c | 8 | ||||
-rw-r--r-- | arch/x86/mm/pat.c | 358 | ||||
-rw-r--r-- | arch/x86/mm/physaddr.c | 70 | ||||
-rw-r--r-- | arch/x86/mm/physaddr.h | 10 | ||||
-rw-r--r-- | arch/x86/mm/srat_32.c | 4 | ||||
-rw-r--r-- | arch/x86/mm/tlb.c | 21 |
13 files changed, 439 insertions, 225 deletions
diff --git a/arch/x86/mm/Makefile b/arch/x86/mm/Makefile index eefdeee8a871..9b5a9f59a478 100644 --- a/arch/x86/mm/Makefile +++ b/arch/x86/mm/Makefile | |||
@@ -1,5 +1,9 @@ | |||
1 | obj-y := init.o init_$(BITS).o fault.o ioremap.o extable.o pageattr.o mmap.o \ | 1 | obj-y := init.o init_$(BITS).o fault.o ioremap.o extable.o pageattr.o mmap.o \ |
2 | pat.o pgtable.o gup.o | 2 | pat.o pgtable.o physaddr.o gup.o |
3 | |||
4 | # Make sure __phys_addr has no stackprotector | ||
5 | nostackp := $(call cc-option, -fno-stack-protector) | ||
6 | CFLAGS_physaddr.o := $(nostackp) | ||
3 | 7 | ||
4 | obj-$(CONFIG_SMP) += tlb.o | 8 | obj-$(CONFIG_SMP) += tlb.o |
5 | 9 | ||
diff --git a/arch/x86/mm/fault.c b/arch/x86/mm/fault.c index bfae139182ff..775a020990a5 100644 --- a/arch/x86/mm/fault.c +++ b/arch/x86/mm/fault.c | |||
@@ -285,26 +285,25 @@ check_v8086_mode(struct pt_regs *regs, unsigned long address, | |||
285 | tsk->thread.screen_bitmap |= 1 << bit; | 285 | tsk->thread.screen_bitmap |= 1 << bit; |
286 | } | 286 | } |
287 | 287 | ||
288 | static void dump_pagetable(unsigned long address) | 288 | static bool low_pfn(unsigned long pfn) |
289 | { | 289 | { |
290 | __typeof__(pte_val(__pte(0))) page; | 290 | return pfn < max_low_pfn; |
291 | } | ||
291 | 292 | ||
292 | page = read_cr3(); | 293 | static void dump_pagetable(unsigned long address) |
293 | page = ((__typeof__(page) *) __va(page))[address >> PGDIR_SHIFT]; | 294 | { |
295 | pgd_t *base = __va(read_cr3()); | ||
296 | pgd_t *pgd = &base[pgd_index(address)]; | ||
297 | pmd_t *pmd; | ||
298 | pte_t *pte; | ||
294 | 299 | ||
295 | #ifdef CONFIG_X86_PAE | 300 | #ifdef CONFIG_X86_PAE |
296 | printk("*pdpt = %016Lx ", page); | 301 | printk("*pdpt = %016Lx ", pgd_val(*pgd)); |
297 | if ((page >> PAGE_SHIFT) < max_low_pfn | 302 | if (!low_pfn(pgd_val(*pgd) >> PAGE_SHIFT) || !pgd_present(*pgd)) |
298 | && page & _PAGE_PRESENT) { | 303 | goto out; |
299 | page &= PAGE_MASK; | ||
300 | page = ((__typeof__(page) *) __va(page))[(address >> PMD_SHIFT) | ||
301 | & (PTRS_PER_PMD - 1)]; | ||
302 | printk(KERN_CONT "*pde = %016Lx ", page); | ||
303 | page &= ~_PAGE_NX; | ||
304 | } | ||
305 | #else | ||
306 | printk("*pde = %08lx ", page); | ||
307 | #endif | 304 | #endif |
305 | pmd = pmd_offset(pud_offset(pgd, address), address); | ||
306 | printk(KERN_CONT "*pde = %0*Lx ", sizeof(*pmd) * 2, (u64)pmd_val(*pmd)); | ||
308 | 307 | ||
309 | /* | 308 | /* |
310 | * We must not directly access the pte in the highpte | 309 | * We must not directly access the pte in the highpte |
@@ -312,16 +311,12 @@ static void dump_pagetable(unsigned long address) | |||
312 | * And let's rather not kmap-atomic the pte, just in case | 311 | * And let's rather not kmap-atomic the pte, just in case |
313 | * it's allocated already: | 312 | * it's allocated already: |
314 | */ | 313 | */ |
315 | if ((page >> PAGE_SHIFT) < max_low_pfn | 314 | if (!low_pfn(pmd_pfn(*pmd)) || !pmd_present(*pmd) || pmd_large(*pmd)) |
316 | && (page & _PAGE_PRESENT) | 315 | goto out; |
317 | && !(page & _PAGE_PSE)) { | ||
318 | |||
319 | page &= PAGE_MASK; | ||
320 | page = ((__typeof__(page) *) __va(page))[(address >> PAGE_SHIFT) | ||
321 | & (PTRS_PER_PTE - 1)]; | ||
322 | printk("*pte = %0*Lx ", sizeof(page)*2, (u64)page); | ||
323 | } | ||
324 | 316 | ||
317 | pte = pte_offset_kernel(pmd, address); | ||
318 | printk("*pte = %0*Lx ", sizeof(*pte) * 2, (u64)pte_val(*pte)); | ||
319 | out: | ||
325 | printk("\n"); | 320 | printk("\n"); |
326 | } | 321 | } |
327 | 322 | ||
@@ -450,16 +445,12 @@ static int bad_address(void *p) | |||
450 | 445 | ||
451 | static void dump_pagetable(unsigned long address) | 446 | static void dump_pagetable(unsigned long address) |
452 | { | 447 | { |
453 | pgd_t *pgd; | 448 | pgd_t *base = __va(read_cr3() & PHYSICAL_PAGE_MASK); |
449 | pgd_t *pgd = base + pgd_index(address); | ||
454 | pud_t *pud; | 450 | pud_t *pud; |
455 | pmd_t *pmd; | 451 | pmd_t *pmd; |
456 | pte_t *pte; | 452 | pte_t *pte; |
457 | 453 | ||
458 | pgd = (pgd_t *)read_cr3(); | ||
459 | |||
460 | pgd = __va((unsigned long)pgd & PHYSICAL_PAGE_MASK); | ||
461 | |||
462 | pgd += pgd_index(address); | ||
463 | if (bad_address(pgd)) | 454 | if (bad_address(pgd)) |
464 | goto bad; | 455 | goto bad; |
465 | 456 | ||
diff --git a/arch/x86/mm/highmem_32.c b/arch/x86/mm/highmem_32.c index 2112ed55e7ea..63a6ba66cbe0 100644 --- a/arch/x86/mm/highmem_32.c +++ b/arch/x86/mm/highmem_32.c | |||
@@ -24,7 +24,7 @@ void kunmap(struct page *page) | |||
24 | * no global lock is needed and because the kmap code must perform a global TLB | 24 | * no global lock is needed and because the kmap code must perform a global TLB |
25 | * invalidation when the kmap pool wraps. | 25 | * invalidation when the kmap pool wraps. |
26 | * | 26 | * |
27 | * However when holding an atomic kmap is is not legal to sleep, so atomic | 27 | * However when holding an atomic kmap it is not legal to sleep, so atomic |
28 | * kmaps are appropriate for short, tight code paths only. | 28 | * kmaps are appropriate for short, tight code paths only. |
29 | */ | 29 | */ |
30 | void *kmap_atomic_prot(struct page *page, enum km_type type, pgprot_t prot) | 30 | void *kmap_atomic_prot(struct page *page, enum km_type type, pgprot_t prot) |
@@ -104,6 +104,7 @@ EXPORT_SYMBOL(kunmap); | |||
104 | EXPORT_SYMBOL(kmap_atomic); | 104 | EXPORT_SYMBOL(kmap_atomic); |
105 | EXPORT_SYMBOL(kunmap_atomic); | 105 | EXPORT_SYMBOL(kunmap_atomic); |
106 | EXPORT_SYMBOL(kmap_atomic_prot); | 106 | EXPORT_SYMBOL(kmap_atomic_prot); |
107 | EXPORT_SYMBOL(kmap_atomic_to_page); | ||
107 | 108 | ||
108 | void __init set_highmem_pages_init(void) | 109 | void __init set_highmem_pages_init(void) |
109 | { | 110 | { |
diff --git a/arch/x86/mm/init_64.c b/arch/x86/mm/init_64.c index 6176fe8f29e0..ea56b8cbb6a6 100644 --- a/arch/x86/mm/init_64.c +++ b/arch/x86/mm/init_64.c | |||
@@ -796,7 +796,7 @@ int __init reserve_bootmem_generic(unsigned long phys, unsigned long len, | |||
796 | return ret; | 796 | return ret; |
797 | 797 | ||
798 | #else | 798 | #else |
799 | reserve_bootmem(phys, len, BOOTMEM_DEFAULT); | 799 | reserve_bootmem(phys, len, flags); |
800 | #endif | 800 | #endif |
801 | 801 | ||
802 | if (phys+len <= MAX_DMA_PFN*PAGE_SIZE) { | 802 | if (phys+len <= MAX_DMA_PFN*PAGE_SIZE) { |
diff --git a/arch/x86/mm/iomap_32.c b/arch/x86/mm/iomap_32.c index fe6f84ca121e..84e236ce76ba 100644 --- a/arch/x86/mm/iomap_32.c +++ b/arch/x86/mm/iomap_32.c | |||
@@ -21,7 +21,7 @@ | |||
21 | #include <linux/module.h> | 21 | #include <linux/module.h> |
22 | #include <linux/highmem.h> | 22 | #include <linux/highmem.h> |
23 | 23 | ||
24 | int is_io_mapping_possible(resource_size_t base, unsigned long size) | 24 | static int is_io_mapping_possible(resource_size_t base, unsigned long size) |
25 | { | 25 | { |
26 | #if !defined(CONFIG_X86_PAE) && defined(CONFIG_PHYS_ADDR_T_64BIT) | 26 | #if !defined(CONFIG_X86_PAE) && defined(CONFIG_PHYS_ADDR_T_64BIT) |
27 | /* There is no way to map greater than 1 << 32 address without PAE */ | 27 | /* There is no way to map greater than 1 << 32 address without PAE */ |
@@ -30,7 +30,30 @@ int is_io_mapping_possible(resource_size_t base, unsigned long size) | |||
30 | #endif | 30 | #endif |
31 | return 1; | 31 | return 1; |
32 | } | 32 | } |
33 | EXPORT_SYMBOL_GPL(is_io_mapping_possible); | 33 | |
34 | int iomap_create_wc(resource_size_t base, unsigned long size, pgprot_t *prot) | ||
35 | { | ||
36 | unsigned long flag = _PAGE_CACHE_WC; | ||
37 | int ret; | ||
38 | |||
39 | if (!is_io_mapping_possible(base, size)) | ||
40 | return -EINVAL; | ||
41 | |||
42 | ret = io_reserve_memtype(base, base + size, &flag); | ||
43 | if (ret) | ||
44 | return ret; | ||
45 | |||
46 | *prot = __pgprot(__PAGE_KERNEL | flag); | ||
47 | return 0; | ||
48 | } | ||
49 | EXPORT_SYMBOL_GPL(iomap_create_wc); | ||
50 | |||
51 | void | ||
52 | iomap_free(resource_size_t base, unsigned long size) | ||
53 | { | ||
54 | io_free_memtype(base, base + size); | ||
55 | } | ||
56 | EXPORT_SYMBOL_GPL(iomap_free); | ||
34 | 57 | ||
35 | void *kmap_atomic_prot_pfn(unsigned long pfn, enum km_type type, pgprot_t prot) | 58 | void *kmap_atomic_prot_pfn(unsigned long pfn, enum km_type type, pgprot_t prot) |
36 | { | 59 | { |
diff --git a/arch/x86/mm/ioremap.c b/arch/x86/mm/ioremap.c index 8a450930834f..334e63ca7b2b 100644 --- a/arch/x86/mm/ioremap.c +++ b/arch/x86/mm/ioremap.c | |||
@@ -22,77 +22,7 @@ | |||
22 | #include <asm/pgalloc.h> | 22 | #include <asm/pgalloc.h> |
23 | #include <asm/pat.h> | 23 | #include <asm/pat.h> |
24 | 24 | ||
25 | static inline int phys_addr_valid(resource_size_t addr) | 25 | #include "physaddr.h" |
26 | { | ||
27 | #ifdef CONFIG_PHYS_ADDR_T_64BIT | ||
28 | return !(addr >> boot_cpu_data.x86_phys_bits); | ||
29 | #else | ||
30 | return 1; | ||
31 | #endif | ||
32 | } | ||
33 | |||
34 | #ifdef CONFIG_X86_64 | ||
35 | |||
36 | unsigned long __phys_addr(unsigned long x) | ||
37 | { | ||
38 | if (x >= __START_KERNEL_map) { | ||
39 | x -= __START_KERNEL_map; | ||
40 | VIRTUAL_BUG_ON(x >= KERNEL_IMAGE_SIZE); | ||
41 | x += phys_base; | ||
42 | } else { | ||
43 | VIRTUAL_BUG_ON(x < PAGE_OFFSET); | ||
44 | x -= PAGE_OFFSET; | ||
45 | VIRTUAL_BUG_ON(!phys_addr_valid(x)); | ||
46 | } | ||
47 | return x; | ||
48 | } | ||
49 | EXPORT_SYMBOL(__phys_addr); | ||
50 | |||
51 | bool __virt_addr_valid(unsigned long x) | ||
52 | { | ||
53 | if (x >= __START_KERNEL_map) { | ||
54 | x -= __START_KERNEL_map; | ||
55 | if (x >= KERNEL_IMAGE_SIZE) | ||
56 | return false; | ||
57 | x += phys_base; | ||
58 | } else { | ||
59 | if (x < PAGE_OFFSET) | ||
60 | return false; | ||
61 | x -= PAGE_OFFSET; | ||
62 | if (!phys_addr_valid(x)) | ||
63 | return false; | ||
64 | } | ||
65 | |||
66 | return pfn_valid(x >> PAGE_SHIFT); | ||
67 | } | ||
68 | EXPORT_SYMBOL(__virt_addr_valid); | ||
69 | |||
70 | #else | ||
71 | |||
72 | #ifdef CONFIG_DEBUG_VIRTUAL | ||
73 | unsigned long __phys_addr(unsigned long x) | ||
74 | { | ||
75 | /* VMALLOC_* aren't constants */ | ||
76 | VIRTUAL_BUG_ON(x < PAGE_OFFSET); | ||
77 | VIRTUAL_BUG_ON(__vmalloc_start_set && is_vmalloc_addr((void *) x)); | ||
78 | return x - PAGE_OFFSET; | ||
79 | } | ||
80 | EXPORT_SYMBOL(__phys_addr); | ||
81 | #endif | ||
82 | |||
83 | bool __virt_addr_valid(unsigned long x) | ||
84 | { | ||
85 | if (x < PAGE_OFFSET) | ||
86 | return false; | ||
87 | if (__vmalloc_start_set && is_vmalloc_addr((void *) x)) | ||
88 | return false; | ||
89 | if (x >= FIXADDR_START) | ||
90 | return false; | ||
91 | return pfn_valid((x - PAGE_OFFSET) >> PAGE_SHIFT); | ||
92 | } | ||
93 | EXPORT_SYMBOL(__virt_addr_valid); | ||
94 | |||
95 | #endif | ||
96 | 26 | ||
97 | int page_is_ram(unsigned long pagenr) | 27 | int page_is_ram(unsigned long pagenr) |
98 | { | 28 | { |
@@ -228,24 +158,14 @@ static void __iomem *__ioremap_caller(resource_size_t phys_addr, | |||
228 | retval = reserve_memtype(phys_addr, (u64)phys_addr + size, | 158 | retval = reserve_memtype(phys_addr, (u64)phys_addr + size, |
229 | prot_val, &new_prot_val); | 159 | prot_val, &new_prot_val); |
230 | if (retval) { | 160 | if (retval) { |
231 | pr_debug("Warning: reserve_memtype returned %d\n", retval); | 161 | printk(KERN_ERR "ioremap reserve_memtype failed %d\n", retval); |
232 | return NULL; | 162 | return NULL; |
233 | } | 163 | } |
234 | 164 | ||
235 | if (prot_val != new_prot_val) { | 165 | if (prot_val != new_prot_val) { |
236 | /* | 166 | if (!is_new_memtype_allowed(phys_addr, size, |
237 | * Do not fallback to certain memory types with certain | 167 | prot_val, new_prot_val)) { |
238 | * requested type: | 168 | printk(KERN_ERR |
239 | * - request is uc-, return cannot be write-back | ||
240 | * - request is uc-, return cannot be write-combine | ||
241 | * - request is write-combine, return cannot be write-back | ||
242 | */ | ||
243 | if ((prot_val == _PAGE_CACHE_UC_MINUS && | ||
244 | (new_prot_val == _PAGE_CACHE_WB || | ||
245 | new_prot_val == _PAGE_CACHE_WC)) || | ||
246 | (prot_val == _PAGE_CACHE_WC && | ||
247 | new_prot_val == _PAGE_CACHE_WB)) { | ||
248 | pr_debug( | ||
249 | "ioremap error for 0x%llx-0x%llx, requested 0x%lx, got 0x%lx\n", | 169 | "ioremap error for 0x%llx-0x%llx, requested 0x%lx, got 0x%lx\n", |
250 | (unsigned long long)phys_addr, | 170 | (unsigned long long)phys_addr, |
251 | (unsigned long long)(phys_addr + size), | 171 | (unsigned long long)(phys_addr + size), |
diff --git a/arch/x86/mm/kmemcheck/kmemcheck.c b/arch/x86/mm/kmemcheck/kmemcheck.c index 2c55ed098654..528bf954eb74 100644 --- a/arch/x86/mm/kmemcheck/kmemcheck.c +++ b/arch/x86/mm/kmemcheck/kmemcheck.c | |||
@@ -331,6 +331,20 @@ static void kmemcheck_read_strict(struct pt_regs *regs, | |||
331 | kmemcheck_shadow_set(shadow, size); | 331 | kmemcheck_shadow_set(shadow, size); |
332 | } | 332 | } |
333 | 333 | ||
334 | bool kmemcheck_is_obj_initialized(unsigned long addr, size_t size) | ||
335 | { | ||
336 | enum kmemcheck_shadow status; | ||
337 | void *shadow; | ||
338 | |||
339 | shadow = kmemcheck_shadow_lookup(addr); | ||
340 | if (!shadow) | ||
341 | return true; | ||
342 | |||
343 | status = kmemcheck_shadow_test(shadow, size); | ||
344 | |||
345 | return status == KMEMCHECK_SHADOW_INITIALIZED; | ||
346 | } | ||
347 | |||
334 | /* Access may cross page boundary */ | 348 | /* Access may cross page boundary */ |
335 | static void kmemcheck_read(struct pt_regs *regs, | 349 | static void kmemcheck_read(struct pt_regs *regs, |
336 | unsigned long addr, unsigned int size) | 350 | unsigned long addr, unsigned int size) |
diff --git a/arch/x86/mm/pageattr.c b/arch/x86/mm/pageattr.c index f53cfc7f963d..24952fdc7e40 100644 --- a/arch/x86/mm/pageattr.c +++ b/arch/x86/mm/pageattr.c | |||
@@ -805,6 +805,7 @@ static int change_page_attr_set_clr(unsigned long *addr, int numpages, | |||
805 | { | 805 | { |
806 | struct cpa_data cpa; | 806 | struct cpa_data cpa; |
807 | int ret, cache, checkalias; | 807 | int ret, cache, checkalias; |
808 | unsigned long baddr = 0; | ||
808 | 809 | ||
809 | /* | 810 | /* |
810 | * Check, if we are requested to change a not supported | 811 | * Check, if we are requested to change a not supported |
@@ -836,6 +837,11 @@ static int change_page_attr_set_clr(unsigned long *addr, int numpages, | |||
836 | */ | 837 | */ |
837 | WARN_ON_ONCE(1); | 838 | WARN_ON_ONCE(1); |
838 | } | 839 | } |
840 | /* | ||
841 | * Save address for cache flush. *addr is modified in the call | ||
842 | * to __change_page_attr_set_clr() below. | ||
843 | */ | ||
844 | baddr = *addr; | ||
839 | } | 845 | } |
840 | 846 | ||
841 | /* Must avoid aliasing mappings in the highmem code */ | 847 | /* Must avoid aliasing mappings in the highmem code */ |
@@ -883,7 +889,7 @@ static int change_page_attr_set_clr(unsigned long *addr, int numpages, | |||
883 | cpa_flush_array(addr, numpages, cache, | 889 | cpa_flush_array(addr, numpages, cache, |
884 | cpa.flags, pages); | 890 | cpa.flags, pages); |
885 | } else | 891 | } else |
886 | cpa_flush_range(*addr, numpages, cache); | 892 | cpa_flush_range(baddr, numpages, cache); |
887 | } else | 893 | } else |
888 | cpa_flush_all(cache); | 894 | cpa_flush_all(cache); |
889 | 895 | ||
diff --git a/arch/x86/mm/pat.c b/arch/x86/mm/pat.c index e6718bb28065..d7ebc3a10f2f 100644 --- a/arch/x86/mm/pat.c +++ b/arch/x86/mm/pat.c | |||
@@ -15,6 +15,7 @@ | |||
15 | #include <linux/gfp.h> | 15 | #include <linux/gfp.h> |
16 | #include <linux/mm.h> | 16 | #include <linux/mm.h> |
17 | #include <linux/fs.h> | 17 | #include <linux/fs.h> |
18 | #include <linux/rbtree.h> | ||
18 | 19 | ||
19 | #include <asm/cacheflush.h> | 20 | #include <asm/cacheflush.h> |
20 | #include <asm/processor.h> | 21 | #include <asm/processor.h> |
@@ -148,11 +149,10 @@ static char *cattr_name(unsigned long flags) | |||
148 | * areas). All the aliases have the same cache attributes of course. | 149 | * areas). All the aliases have the same cache attributes of course. |
149 | * Zero attributes are represented as holes. | 150 | * Zero attributes are represented as holes. |
150 | * | 151 | * |
151 | * Currently the data structure is a list because the number of mappings | 152 | * The data structure is a list that is also organized as an rbtree |
152 | * are expected to be relatively small. If this should be a problem | 153 | * sorted on the start address of memtype range. |
153 | * it could be changed to a rbtree or similar. | ||
154 | * | 154 | * |
155 | * memtype_lock protects the whole list. | 155 | * memtype_lock protects both the linear list and rbtree. |
156 | */ | 156 | */ |
157 | 157 | ||
158 | struct memtype { | 158 | struct memtype { |
@@ -160,11 +160,53 @@ struct memtype { | |||
160 | u64 end; | 160 | u64 end; |
161 | unsigned long type; | 161 | unsigned long type; |
162 | struct list_head nd; | 162 | struct list_head nd; |
163 | struct rb_node rb; | ||
163 | }; | 164 | }; |
164 | 165 | ||
166 | static struct rb_root memtype_rbroot = RB_ROOT; | ||
165 | static LIST_HEAD(memtype_list); | 167 | static LIST_HEAD(memtype_list); |
166 | static DEFINE_SPINLOCK(memtype_lock); /* protects memtype list */ | 168 | static DEFINE_SPINLOCK(memtype_lock); /* protects memtype list */ |
167 | 169 | ||
170 | static struct memtype *memtype_rb_search(struct rb_root *root, u64 start) | ||
171 | { | ||
172 | struct rb_node *node = root->rb_node; | ||
173 | struct memtype *last_lower = NULL; | ||
174 | |||
175 | while (node) { | ||
176 | struct memtype *data = container_of(node, struct memtype, rb); | ||
177 | |||
178 | if (data->start < start) { | ||
179 | last_lower = data; | ||
180 | node = node->rb_right; | ||
181 | } else if (data->start > start) { | ||
182 | node = node->rb_left; | ||
183 | } else | ||
184 | return data; | ||
185 | } | ||
186 | |||
187 | /* Will return NULL if there is no entry with its start <= start */ | ||
188 | return last_lower; | ||
189 | } | ||
190 | |||
191 | static void memtype_rb_insert(struct rb_root *root, struct memtype *data) | ||
192 | { | ||
193 | struct rb_node **new = &(root->rb_node); | ||
194 | struct rb_node *parent = NULL; | ||
195 | |||
196 | while (*new) { | ||
197 | struct memtype *this = container_of(*new, struct memtype, rb); | ||
198 | |||
199 | parent = *new; | ||
200 | if (data->start <= this->start) | ||
201 | new = &((*new)->rb_left); | ||
202 | else if (data->start > this->start) | ||
203 | new = &((*new)->rb_right); | ||
204 | } | ||
205 | |||
206 | rb_link_node(&data->rb, parent, new); | ||
207 | rb_insert_color(&data->rb, root); | ||
208 | } | ||
209 | |||
168 | /* | 210 | /* |
169 | * Does intersection of PAT memory type and MTRR memory type and returns | 211 | * Does intersection of PAT memory type and MTRR memory type and returns |
170 | * the resulting memory type as PAT understands it. | 212 | * the resulting memory type as PAT understands it. |
@@ -218,9 +260,6 @@ chk_conflict(struct memtype *new, struct memtype *entry, unsigned long *type) | |||
218 | return -EBUSY; | 260 | return -EBUSY; |
219 | } | 261 | } |
220 | 262 | ||
221 | static struct memtype *cached_entry; | ||
222 | static u64 cached_start; | ||
223 | |||
224 | static int pat_pagerange_is_ram(unsigned long start, unsigned long end) | 263 | static int pat_pagerange_is_ram(unsigned long start, unsigned long end) |
225 | { | 264 | { |
226 | int ram_page = 0, not_rampage = 0; | 265 | int ram_page = 0, not_rampage = 0; |
@@ -249,63 +288,61 @@ static int pat_pagerange_is_ram(unsigned long start, unsigned long end) | |||
249 | } | 288 | } |
250 | 289 | ||
251 | /* | 290 | /* |
252 | * For RAM pages, mark the pages as non WB memory type using | 291 | * For RAM pages, we use page flags to mark the pages with appropriate type. |
253 | * PageNonWB (PG_arch_1). We allow only one set_memory_uc() or | 292 | * Here we do two pass: |
254 | * set_memory_wc() on a RAM page at a time before marking it as WB again. | 293 | * - Find the memtype of all the pages in the range, look for any conflicts |
255 | * This is ok, because only one driver will be owning the page and | 294 | * - In case of no conflicts, set the new memtype for pages in the range |
256 | * doing set_memory_*() calls. | ||
257 | * | 295 | * |
258 | * For now, we use PageNonWB to track that the RAM page is being mapped | 296 | * Caller must hold memtype_lock for atomicity. |
259 | * as non WB. In future, we will have to use one more flag | ||
260 | * (or some other mechanism in page_struct) to distinguish between | ||
261 | * UC and WC mapping. | ||
262 | */ | 297 | */ |
263 | static int reserve_ram_pages_type(u64 start, u64 end, unsigned long req_type, | 298 | static int reserve_ram_pages_type(u64 start, u64 end, unsigned long req_type, |
264 | unsigned long *new_type) | 299 | unsigned long *new_type) |
265 | { | 300 | { |
266 | struct page *page; | 301 | struct page *page; |
267 | u64 pfn, end_pfn; | 302 | u64 pfn; |
303 | |||
304 | if (req_type == _PAGE_CACHE_UC) { | ||
305 | /* We do not support strong UC */ | ||
306 | WARN_ON_ONCE(1); | ||
307 | req_type = _PAGE_CACHE_UC_MINUS; | ||
308 | } | ||
268 | 309 | ||
269 | for (pfn = (start >> PAGE_SHIFT); pfn < (end >> PAGE_SHIFT); ++pfn) { | 310 | for (pfn = (start >> PAGE_SHIFT); pfn < (end >> PAGE_SHIFT); ++pfn) { |
270 | page = pfn_to_page(pfn); | 311 | unsigned long type; |
271 | if (page_mapped(page) || PageNonWB(page)) | ||
272 | goto out; | ||
273 | 312 | ||
274 | SetPageNonWB(page); | 313 | page = pfn_to_page(pfn); |
314 | type = get_page_memtype(page); | ||
315 | if (type != -1) { | ||
316 | printk(KERN_INFO "reserve_ram_pages_type failed " | ||
317 | "0x%Lx-0x%Lx, track 0x%lx, req 0x%lx\n", | ||
318 | start, end, type, req_type); | ||
319 | if (new_type) | ||
320 | *new_type = type; | ||
321 | |||
322 | return -EBUSY; | ||
323 | } | ||
275 | } | 324 | } |
276 | return 0; | ||
277 | 325 | ||
278 | out: | 326 | if (new_type) |
279 | end_pfn = pfn; | 327 | *new_type = req_type; |
280 | for (pfn = (start >> PAGE_SHIFT); pfn < end_pfn; ++pfn) { | 328 | |
329 | for (pfn = (start >> PAGE_SHIFT); pfn < (end >> PAGE_SHIFT); ++pfn) { | ||
281 | page = pfn_to_page(pfn); | 330 | page = pfn_to_page(pfn); |
282 | ClearPageNonWB(page); | 331 | set_page_memtype(page, req_type); |
283 | } | 332 | } |
284 | 333 | return 0; | |
285 | return -EINVAL; | ||
286 | } | 334 | } |
287 | 335 | ||
288 | static int free_ram_pages_type(u64 start, u64 end) | 336 | static int free_ram_pages_type(u64 start, u64 end) |
289 | { | 337 | { |
290 | struct page *page; | 338 | struct page *page; |
291 | u64 pfn, end_pfn; | 339 | u64 pfn; |
292 | 340 | ||
293 | for (pfn = (start >> PAGE_SHIFT); pfn < (end >> PAGE_SHIFT); ++pfn) { | 341 | for (pfn = (start >> PAGE_SHIFT); pfn < (end >> PAGE_SHIFT); ++pfn) { |
294 | page = pfn_to_page(pfn); | 342 | page = pfn_to_page(pfn); |
295 | if (page_mapped(page) || !PageNonWB(page)) | 343 | set_page_memtype(page, -1); |
296 | goto out; | ||
297 | |||
298 | ClearPageNonWB(page); | ||
299 | } | 344 | } |
300 | return 0; | 345 | return 0; |
301 | |||
302 | out: | ||
303 | end_pfn = pfn; | ||
304 | for (pfn = (start >> PAGE_SHIFT); pfn < end_pfn; ++pfn) { | ||
305 | page = pfn_to_page(pfn); | ||
306 | SetPageNonWB(page); | ||
307 | } | ||
308 | return -EINVAL; | ||
309 | } | 346 | } |
310 | 347 | ||
311 | /* | 348 | /* |
@@ -339,6 +376,8 @@ int reserve_memtype(u64 start, u64 end, unsigned long req_type, | |||
339 | if (new_type) { | 376 | if (new_type) { |
340 | if (req_type == -1) | 377 | if (req_type == -1) |
341 | *new_type = _PAGE_CACHE_WB; | 378 | *new_type = _PAGE_CACHE_WB; |
379 | else if (req_type == _PAGE_CACHE_WC) | ||
380 | *new_type = _PAGE_CACHE_UC_MINUS; | ||
342 | else | 381 | else |
343 | *new_type = req_type & _PAGE_CACHE_MASK; | 382 | *new_type = req_type & _PAGE_CACHE_MASK; |
344 | } | 383 | } |
@@ -364,11 +403,16 @@ int reserve_memtype(u64 start, u64 end, unsigned long req_type, | |||
364 | *new_type = actual_type; | 403 | *new_type = actual_type; |
365 | 404 | ||
366 | is_range_ram = pat_pagerange_is_ram(start, end); | 405 | is_range_ram = pat_pagerange_is_ram(start, end); |
367 | if (is_range_ram == 1) | 406 | if (is_range_ram == 1) { |
368 | return reserve_ram_pages_type(start, end, req_type, | 407 | |
369 | new_type); | 408 | spin_lock(&memtype_lock); |
370 | else if (is_range_ram < 0) | 409 | err = reserve_ram_pages_type(start, end, req_type, new_type); |
410 | spin_unlock(&memtype_lock); | ||
411 | |||
412 | return err; | ||
413 | } else if (is_range_ram < 0) { | ||
371 | return -EINVAL; | 414 | return -EINVAL; |
415 | } | ||
372 | 416 | ||
373 | new = kmalloc(sizeof(struct memtype), GFP_KERNEL); | 417 | new = kmalloc(sizeof(struct memtype), GFP_KERNEL); |
374 | if (!new) | 418 | if (!new) |
@@ -380,17 +424,19 @@ int reserve_memtype(u64 start, u64 end, unsigned long req_type, | |||
380 | 424 | ||
381 | spin_lock(&memtype_lock); | 425 | spin_lock(&memtype_lock); |
382 | 426 | ||
383 | if (cached_entry && start >= cached_start) | 427 | entry = memtype_rb_search(&memtype_rbroot, new->start); |
384 | entry = cached_entry; | 428 | if (likely(entry != NULL)) { |
385 | else | 429 | /* To work correctly with list_for_each_entry_continue */ |
430 | entry = list_entry(entry->nd.prev, struct memtype, nd); | ||
431 | } else { | ||
386 | entry = list_entry(&memtype_list, struct memtype, nd); | 432 | entry = list_entry(&memtype_list, struct memtype, nd); |
433 | } | ||
387 | 434 | ||
388 | /* Search for existing mapping that overlaps the current range */ | 435 | /* Search for existing mapping that overlaps the current range */ |
389 | where = NULL; | 436 | where = NULL; |
390 | list_for_each_entry_continue(entry, &memtype_list, nd) { | 437 | list_for_each_entry_continue(entry, &memtype_list, nd) { |
391 | if (end <= entry->start) { | 438 | if (end <= entry->start) { |
392 | where = entry->nd.prev; | 439 | where = entry->nd.prev; |
393 | cached_entry = list_entry(where, struct memtype, nd); | ||
394 | break; | 440 | break; |
395 | } else if (start <= entry->start) { /* end > entry->start */ | 441 | } else if (start <= entry->start) { /* end > entry->start */ |
396 | err = chk_conflict(new, entry, new_type); | 442 | err = chk_conflict(new, entry, new_type); |
@@ -398,8 +444,6 @@ int reserve_memtype(u64 start, u64 end, unsigned long req_type, | |||
398 | dprintk("Overlap at 0x%Lx-0x%Lx\n", | 444 | dprintk("Overlap at 0x%Lx-0x%Lx\n", |
399 | entry->start, entry->end); | 445 | entry->start, entry->end); |
400 | where = entry->nd.prev; | 446 | where = entry->nd.prev; |
401 | cached_entry = list_entry(where, | ||
402 | struct memtype, nd); | ||
403 | } | 447 | } |
404 | break; | 448 | break; |
405 | } else if (start < entry->end) { /* start > entry->start */ | 449 | } else if (start < entry->end) { /* start > entry->start */ |
@@ -407,8 +451,6 @@ int reserve_memtype(u64 start, u64 end, unsigned long req_type, | |||
407 | if (!err) { | 451 | if (!err) { |
408 | dprintk("Overlap at 0x%Lx-0x%Lx\n", | 452 | dprintk("Overlap at 0x%Lx-0x%Lx\n", |
409 | entry->start, entry->end); | 453 | entry->start, entry->end); |
410 | cached_entry = list_entry(entry->nd.prev, | ||
411 | struct memtype, nd); | ||
412 | 454 | ||
413 | /* | 455 | /* |
414 | * Move to right position in the linked | 456 | * Move to right position in the linked |
@@ -436,13 +478,13 @@ int reserve_memtype(u64 start, u64 end, unsigned long req_type, | |||
436 | return err; | 478 | return err; |
437 | } | 479 | } |
438 | 480 | ||
439 | cached_start = start; | ||
440 | |||
441 | if (where) | 481 | if (where) |
442 | list_add(&new->nd, where); | 482 | list_add(&new->nd, where); |
443 | else | 483 | else |
444 | list_add_tail(&new->nd, &memtype_list); | 484 | list_add_tail(&new->nd, &memtype_list); |
445 | 485 | ||
486 | memtype_rb_insert(&memtype_rbroot, new); | ||
487 | |||
446 | spin_unlock(&memtype_lock); | 488 | spin_unlock(&memtype_lock); |
447 | 489 | ||
448 | dprintk("reserve_memtype added 0x%Lx-0x%Lx, track %s, req %s, ret %s\n", | 490 | dprintk("reserve_memtype added 0x%Lx-0x%Lx, track %s, req %s, ret %s\n", |
@@ -454,7 +496,7 @@ int reserve_memtype(u64 start, u64 end, unsigned long req_type, | |||
454 | 496 | ||
455 | int free_memtype(u64 start, u64 end) | 497 | int free_memtype(u64 start, u64 end) |
456 | { | 498 | { |
457 | struct memtype *entry; | 499 | struct memtype *entry, *saved_entry; |
458 | int err = -EINVAL; | 500 | int err = -EINVAL; |
459 | int is_range_ram; | 501 | int is_range_ram; |
460 | 502 | ||
@@ -466,23 +508,58 @@ int free_memtype(u64 start, u64 end) | |||
466 | return 0; | 508 | return 0; |
467 | 509 | ||
468 | is_range_ram = pat_pagerange_is_ram(start, end); | 510 | is_range_ram = pat_pagerange_is_ram(start, end); |
469 | if (is_range_ram == 1) | 511 | if (is_range_ram == 1) { |
470 | return free_ram_pages_type(start, end); | 512 | |
471 | else if (is_range_ram < 0) | 513 | spin_lock(&memtype_lock); |
514 | err = free_ram_pages_type(start, end); | ||
515 | spin_unlock(&memtype_lock); | ||
516 | |||
517 | return err; | ||
518 | } else if (is_range_ram < 0) { | ||
472 | return -EINVAL; | 519 | return -EINVAL; |
520 | } | ||
473 | 521 | ||
474 | spin_lock(&memtype_lock); | 522 | spin_lock(&memtype_lock); |
523 | |||
524 | entry = memtype_rb_search(&memtype_rbroot, start); | ||
525 | if (unlikely(entry == NULL)) | ||
526 | goto unlock_ret; | ||
527 | |||
528 | /* | ||
529 | * Saved entry points to an entry with start same or less than what | ||
530 | * we searched for. Now go through the list in both directions to look | ||
531 | * for the entry that matches with both start and end, with list stored | ||
532 | * in sorted start address | ||
533 | */ | ||
534 | saved_entry = entry; | ||
475 | list_for_each_entry(entry, &memtype_list, nd) { | 535 | list_for_each_entry(entry, &memtype_list, nd) { |
476 | if (entry->start == start && entry->end == end) { | 536 | if (entry->start == start && entry->end == end) { |
477 | if (cached_entry == entry || cached_start == start) | 537 | rb_erase(&entry->rb, &memtype_rbroot); |
478 | cached_entry = NULL; | 538 | list_del(&entry->nd); |
539 | kfree(entry); | ||
540 | err = 0; | ||
541 | break; | ||
542 | } else if (entry->start > start) { | ||
543 | break; | ||
544 | } | ||
545 | } | ||
546 | |||
547 | if (!err) | ||
548 | goto unlock_ret; | ||
479 | 549 | ||
550 | entry = saved_entry; | ||
551 | list_for_each_entry_reverse(entry, &memtype_list, nd) { | ||
552 | if (entry->start == start && entry->end == end) { | ||
553 | rb_erase(&entry->rb, &memtype_rbroot); | ||
480 | list_del(&entry->nd); | 554 | list_del(&entry->nd); |
481 | kfree(entry); | 555 | kfree(entry); |
482 | err = 0; | 556 | err = 0; |
483 | break; | 557 | break; |
558 | } else if (entry->start < start) { | ||
559 | break; | ||
484 | } | 560 | } |
485 | } | 561 | } |
562 | unlock_ret: | ||
486 | spin_unlock(&memtype_lock); | 563 | spin_unlock(&memtype_lock); |
487 | 564 | ||
488 | if (err) { | 565 | if (err) { |
@@ -496,6 +573,101 @@ int free_memtype(u64 start, u64 end) | |||
496 | } | 573 | } |
497 | 574 | ||
498 | 575 | ||
576 | /** | ||
577 | * lookup_memtype - Looksup the memory type for a physical address | ||
578 | * @paddr: physical address of which memory type needs to be looked up | ||
579 | * | ||
580 | * Only to be called when PAT is enabled | ||
581 | * | ||
582 | * Returns _PAGE_CACHE_WB, _PAGE_CACHE_WC, _PAGE_CACHE_UC_MINUS or | ||
583 | * _PAGE_CACHE_UC | ||
584 | */ | ||
585 | static unsigned long lookup_memtype(u64 paddr) | ||
586 | { | ||
587 | int rettype = _PAGE_CACHE_WB; | ||
588 | struct memtype *entry; | ||
589 | |||
590 | if (is_ISA_range(paddr, paddr + PAGE_SIZE - 1)) | ||
591 | return rettype; | ||
592 | |||
593 | if (pat_pagerange_is_ram(paddr, paddr + PAGE_SIZE)) { | ||
594 | struct page *page; | ||
595 | spin_lock(&memtype_lock); | ||
596 | page = pfn_to_page(paddr >> PAGE_SHIFT); | ||
597 | rettype = get_page_memtype(page); | ||
598 | spin_unlock(&memtype_lock); | ||
599 | /* | ||
600 | * -1 from get_page_memtype() implies RAM page is in its | ||
601 | * default state and not reserved, and hence of type WB | ||
602 | */ | ||
603 | if (rettype == -1) | ||
604 | rettype = _PAGE_CACHE_WB; | ||
605 | |||
606 | return rettype; | ||
607 | } | ||
608 | |||
609 | spin_lock(&memtype_lock); | ||
610 | |||
611 | entry = memtype_rb_search(&memtype_rbroot, paddr); | ||
612 | if (entry != NULL) | ||
613 | rettype = entry->type; | ||
614 | else | ||
615 | rettype = _PAGE_CACHE_UC_MINUS; | ||
616 | |||
617 | spin_unlock(&memtype_lock); | ||
618 | return rettype; | ||
619 | } | ||
620 | |||
621 | /** | ||
622 | * io_reserve_memtype - Request a memory type mapping for a region of memory | ||
623 | * @start: start (physical address) of the region | ||
624 | * @end: end (physical address) of the region | ||
625 | * @type: A pointer to memtype, with requested type. On success, requested | ||
626 | * or any other compatible type that was available for the region is returned | ||
627 | * | ||
628 | * On success, returns 0 | ||
629 | * On failure, returns non-zero | ||
630 | */ | ||
631 | int io_reserve_memtype(resource_size_t start, resource_size_t end, | ||
632 | unsigned long *type) | ||
633 | { | ||
634 | resource_size_t size = end - start; | ||
635 | unsigned long req_type = *type; | ||
636 | unsigned long new_type; | ||
637 | int ret; | ||
638 | |||
639 | WARN_ON_ONCE(iomem_map_sanity_check(start, size)); | ||
640 | |||
641 | ret = reserve_memtype(start, end, req_type, &new_type); | ||
642 | if (ret) | ||
643 | goto out_err; | ||
644 | |||
645 | if (!is_new_memtype_allowed(start, size, req_type, new_type)) | ||
646 | goto out_free; | ||
647 | |||
648 | if (kernel_map_sync_memtype(start, size, new_type) < 0) | ||
649 | goto out_free; | ||
650 | |||
651 | *type = new_type; | ||
652 | return 0; | ||
653 | |||
654 | out_free: | ||
655 | free_memtype(start, end); | ||
656 | ret = -EBUSY; | ||
657 | out_err: | ||
658 | return ret; | ||
659 | } | ||
660 | |||
661 | /** | ||
662 | * io_free_memtype - Release a memory type mapping for a region of memory | ||
663 | * @start: start (physical address) of the region | ||
664 | * @end: end (physical address) of the region | ||
665 | */ | ||
666 | void io_free_memtype(resource_size_t start, resource_size_t end) | ||
667 | { | ||
668 | free_memtype(start, end); | ||
669 | } | ||
670 | |||
499 | pgprot_t phys_mem_access_prot(struct file *file, unsigned long pfn, | 671 | pgprot_t phys_mem_access_prot(struct file *file, unsigned long pfn, |
500 | unsigned long size, pgprot_t vma_prot) | 672 | unsigned long size, pgprot_t vma_prot) |
501 | { | 673 | { |
@@ -577,7 +749,7 @@ int kernel_map_sync_memtype(u64 base, unsigned long size, unsigned long flags) | |||
577 | { | 749 | { |
578 | unsigned long id_sz; | 750 | unsigned long id_sz; |
579 | 751 | ||
580 | if (!pat_enabled || base >= __pa(high_memory)) | 752 | if (base >= __pa(high_memory)) |
581 | return 0; | 753 | return 0; |
582 | 754 | ||
583 | id_sz = (__pa(high_memory) < base + size) ? | 755 | id_sz = (__pa(high_memory) < base + size) ? |
@@ -612,18 +784,37 @@ static int reserve_pfn_range(u64 paddr, unsigned long size, pgprot_t *vma_prot, | |||
612 | is_ram = pat_pagerange_is_ram(paddr, paddr + size); | 784 | is_ram = pat_pagerange_is_ram(paddr, paddr + size); |
613 | 785 | ||
614 | /* | 786 | /* |
615 | * reserve_pfn_range() doesn't support RAM pages. Maintain the current | 787 | * reserve_pfn_range() for RAM pages. We do not refcount to keep |
616 | * behavior with RAM pages by returning success. | 788 | * track of number of mappings of RAM pages. We can assert that |
789 | * the type requested matches the type of first page in the range. | ||
617 | */ | 790 | */ |
618 | if (is_ram != 0) | 791 | if (is_ram) { |
792 | if (!pat_enabled) | ||
793 | return 0; | ||
794 | |||
795 | flags = lookup_memtype(paddr); | ||
796 | if (want_flags != flags) { | ||
797 | printk(KERN_WARNING | ||
798 | "%s:%d map pfn RAM range req %s for %Lx-%Lx, got %s\n", | ||
799 | current->comm, current->pid, | ||
800 | cattr_name(want_flags), | ||
801 | (unsigned long long)paddr, | ||
802 | (unsigned long long)(paddr + size), | ||
803 | cattr_name(flags)); | ||
804 | *vma_prot = __pgprot((pgprot_val(*vma_prot) & | ||
805 | (~_PAGE_CACHE_MASK)) | | ||
806 | flags); | ||
807 | } | ||
619 | return 0; | 808 | return 0; |
809 | } | ||
620 | 810 | ||
621 | ret = reserve_memtype(paddr, paddr + size, want_flags, &flags); | 811 | ret = reserve_memtype(paddr, paddr + size, want_flags, &flags); |
622 | if (ret) | 812 | if (ret) |
623 | return ret; | 813 | return ret; |
624 | 814 | ||
625 | if (flags != want_flags) { | 815 | if (flags != want_flags) { |
626 | if (strict_prot || !is_new_memtype_allowed(want_flags, flags)) { | 816 | if (strict_prot || |
817 | !is_new_memtype_allowed(paddr, size, want_flags, flags)) { | ||
627 | free_memtype(paddr, paddr + size); | 818 | free_memtype(paddr, paddr + size); |
628 | printk(KERN_ERR "%s:%d map pfn expected mapping type %s" | 819 | printk(KERN_ERR "%s:%d map pfn expected mapping type %s" |
629 | " for %Lx-%Lx, got %s\n", | 820 | " for %Lx-%Lx, got %s\n", |
@@ -677,14 +868,6 @@ int track_pfn_vma_copy(struct vm_area_struct *vma) | |||
677 | unsigned long vma_size = vma->vm_end - vma->vm_start; | 868 | unsigned long vma_size = vma->vm_end - vma->vm_start; |
678 | pgprot_t pgprot; | 869 | pgprot_t pgprot; |
679 | 870 | ||
680 | if (!pat_enabled) | ||
681 | return 0; | ||
682 | |||
683 | /* | ||
684 | * For now, only handle remap_pfn_range() vmas where | ||
685 | * is_linear_pfn_mapping() == TRUE. Handling of | ||
686 | * vm_insert_pfn() is TBD. | ||
687 | */ | ||
688 | if (is_linear_pfn_mapping(vma)) { | 871 | if (is_linear_pfn_mapping(vma)) { |
689 | /* | 872 | /* |
690 | * reserve the whole chunk covered by vma. We need the | 873 | * reserve the whole chunk covered by vma. We need the |
@@ -712,23 +895,24 @@ int track_pfn_vma_copy(struct vm_area_struct *vma) | |||
712 | int track_pfn_vma_new(struct vm_area_struct *vma, pgprot_t *prot, | 895 | int track_pfn_vma_new(struct vm_area_struct *vma, pgprot_t *prot, |
713 | unsigned long pfn, unsigned long size) | 896 | unsigned long pfn, unsigned long size) |
714 | { | 897 | { |
898 | unsigned long flags; | ||
715 | resource_size_t paddr; | 899 | resource_size_t paddr; |
716 | unsigned long vma_size = vma->vm_end - vma->vm_start; | 900 | unsigned long vma_size = vma->vm_end - vma->vm_start; |
717 | 901 | ||
718 | if (!pat_enabled) | ||
719 | return 0; | ||
720 | |||
721 | /* | ||
722 | * For now, only handle remap_pfn_range() vmas where | ||
723 | * is_linear_pfn_mapping() == TRUE. Handling of | ||
724 | * vm_insert_pfn() is TBD. | ||
725 | */ | ||
726 | if (is_linear_pfn_mapping(vma)) { | 902 | if (is_linear_pfn_mapping(vma)) { |
727 | /* reserve the whole chunk starting from vm_pgoff */ | 903 | /* reserve the whole chunk starting from vm_pgoff */ |
728 | paddr = (resource_size_t)vma->vm_pgoff << PAGE_SHIFT; | 904 | paddr = (resource_size_t)vma->vm_pgoff << PAGE_SHIFT; |
729 | return reserve_pfn_range(paddr, vma_size, prot, 0); | 905 | return reserve_pfn_range(paddr, vma_size, prot, 0); |
730 | } | 906 | } |
731 | 907 | ||
908 | if (!pat_enabled) | ||
909 | return 0; | ||
910 | |||
911 | /* for vm_insert_pfn and friends, we set prot based on lookup */ | ||
912 | flags = lookup_memtype(pfn << PAGE_SHIFT); | ||
913 | *prot = __pgprot((pgprot_val(vma->vm_page_prot) & (~_PAGE_CACHE_MASK)) | | ||
914 | flags); | ||
915 | |||
732 | return 0; | 916 | return 0; |
733 | } | 917 | } |
734 | 918 | ||
@@ -743,14 +927,6 @@ void untrack_pfn_vma(struct vm_area_struct *vma, unsigned long pfn, | |||
743 | resource_size_t paddr; | 927 | resource_size_t paddr; |
744 | unsigned long vma_size = vma->vm_end - vma->vm_start; | 928 | unsigned long vma_size = vma->vm_end - vma->vm_start; |
745 | 929 | ||
746 | if (!pat_enabled) | ||
747 | return; | ||
748 | |||
749 | /* | ||
750 | * For now, only handle remap_pfn_range() vmas where | ||
751 | * is_linear_pfn_mapping() == TRUE. Handling of | ||
752 | * vm_insert_pfn() is TBD. | ||
753 | */ | ||
754 | if (is_linear_pfn_mapping(vma)) { | 930 | if (is_linear_pfn_mapping(vma)) { |
755 | /* free the whole chunk starting from vm_pgoff */ | 931 | /* free the whole chunk starting from vm_pgoff */ |
756 | paddr = (resource_size_t)vma->vm_pgoff << PAGE_SHIFT; | 932 | paddr = (resource_size_t)vma->vm_pgoff << PAGE_SHIFT; |
@@ -826,7 +1002,7 @@ static int memtype_seq_show(struct seq_file *seq, void *v) | |||
826 | return 0; | 1002 | return 0; |
827 | } | 1003 | } |
828 | 1004 | ||
829 | static struct seq_operations memtype_seq_ops = { | 1005 | static const struct seq_operations memtype_seq_ops = { |
830 | .start = memtype_seq_start, | 1006 | .start = memtype_seq_start, |
831 | .next = memtype_seq_next, | 1007 | .next = memtype_seq_next, |
832 | .stop = memtype_seq_stop, | 1008 | .stop = memtype_seq_stop, |
diff --git a/arch/x86/mm/physaddr.c b/arch/x86/mm/physaddr.c new file mode 100644 index 000000000000..d2e2735327b4 --- /dev/null +++ b/arch/x86/mm/physaddr.c | |||
@@ -0,0 +1,70 @@ | |||
1 | #include <linux/mmdebug.h> | ||
2 | #include <linux/module.h> | ||
3 | #include <linux/mm.h> | ||
4 | |||
5 | #include <asm/page.h> | ||
6 | |||
7 | #include "physaddr.h" | ||
8 | |||
9 | #ifdef CONFIG_X86_64 | ||
10 | |||
11 | unsigned long __phys_addr(unsigned long x) | ||
12 | { | ||
13 | if (x >= __START_KERNEL_map) { | ||
14 | x -= __START_KERNEL_map; | ||
15 | VIRTUAL_BUG_ON(x >= KERNEL_IMAGE_SIZE); | ||
16 | x += phys_base; | ||
17 | } else { | ||
18 | VIRTUAL_BUG_ON(x < PAGE_OFFSET); | ||
19 | x -= PAGE_OFFSET; | ||
20 | VIRTUAL_BUG_ON(!phys_addr_valid(x)); | ||
21 | } | ||
22 | return x; | ||
23 | } | ||
24 | EXPORT_SYMBOL(__phys_addr); | ||
25 | |||
26 | bool __virt_addr_valid(unsigned long x) | ||
27 | { | ||
28 | if (x >= __START_KERNEL_map) { | ||
29 | x -= __START_KERNEL_map; | ||
30 | if (x >= KERNEL_IMAGE_SIZE) | ||
31 | return false; | ||
32 | x += phys_base; | ||
33 | } else { | ||
34 | if (x < PAGE_OFFSET) | ||
35 | return false; | ||
36 | x -= PAGE_OFFSET; | ||
37 | if (!phys_addr_valid(x)) | ||
38 | return false; | ||
39 | } | ||
40 | |||
41 | return pfn_valid(x >> PAGE_SHIFT); | ||
42 | } | ||
43 | EXPORT_SYMBOL(__virt_addr_valid); | ||
44 | |||
45 | #else | ||
46 | |||
47 | #ifdef CONFIG_DEBUG_VIRTUAL | ||
48 | unsigned long __phys_addr(unsigned long x) | ||
49 | { | ||
50 | /* VMALLOC_* aren't constants */ | ||
51 | VIRTUAL_BUG_ON(x < PAGE_OFFSET); | ||
52 | VIRTUAL_BUG_ON(__vmalloc_start_set && is_vmalloc_addr((void *) x)); | ||
53 | return x - PAGE_OFFSET; | ||
54 | } | ||
55 | EXPORT_SYMBOL(__phys_addr); | ||
56 | #endif | ||
57 | |||
58 | bool __virt_addr_valid(unsigned long x) | ||
59 | { | ||
60 | if (x < PAGE_OFFSET) | ||
61 | return false; | ||
62 | if (__vmalloc_start_set && is_vmalloc_addr((void *) x)) | ||
63 | return false; | ||
64 | if (x >= FIXADDR_START) | ||
65 | return false; | ||
66 | return pfn_valid((x - PAGE_OFFSET) >> PAGE_SHIFT); | ||
67 | } | ||
68 | EXPORT_SYMBOL(__virt_addr_valid); | ||
69 | |||
70 | #endif /* CONFIG_X86_64 */ | ||
diff --git a/arch/x86/mm/physaddr.h b/arch/x86/mm/physaddr.h new file mode 100644 index 000000000000..a3cd5a0c97b3 --- /dev/null +++ b/arch/x86/mm/physaddr.h | |||
@@ -0,0 +1,10 @@ | |||
1 | #include <asm/processor.h> | ||
2 | |||
3 | static inline int phys_addr_valid(resource_size_t addr) | ||
4 | { | ||
5 | #ifdef CONFIG_PHYS_ADDR_T_64BIT | ||
6 | return !(addr >> boot_cpu_data.x86_phys_bits); | ||
7 | #else | ||
8 | return 1; | ||
9 | #endif | ||
10 | } | ||
diff --git a/arch/x86/mm/srat_32.c b/arch/x86/mm/srat_32.c index 29a0e37114f8..6f8aa33031c7 100644 --- a/arch/x86/mm/srat_32.c +++ b/arch/x86/mm/srat_32.c | |||
@@ -215,7 +215,7 @@ int __init get_memcfg_from_srat(void) | |||
215 | goto out_fail; | 215 | goto out_fail; |
216 | 216 | ||
217 | if (num_memory_chunks == 0) { | 217 | if (num_memory_chunks == 0) { |
218 | printk(KERN_WARNING | 218 | printk(KERN_DEBUG |
219 | "could not find any ACPI SRAT memory areas.\n"); | 219 | "could not find any ACPI SRAT memory areas.\n"); |
220 | goto out_fail; | 220 | goto out_fail; |
221 | } | 221 | } |
@@ -277,7 +277,7 @@ int __init get_memcfg_from_srat(void) | |||
277 | } | 277 | } |
278 | return 1; | 278 | return 1; |
279 | out_fail: | 279 | out_fail: |
280 | printk(KERN_ERR "failed to get NUMA memory information from SRAT" | 280 | printk(KERN_DEBUG "failed to get NUMA memory information from SRAT" |
281 | " table\n"); | 281 | " table\n"); |
282 | return 0; | 282 | return 0; |
283 | } | 283 | } |
diff --git a/arch/x86/mm/tlb.c b/arch/x86/mm/tlb.c index 821e97017e95..c814e144a3f0 100644 --- a/arch/x86/mm/tlb.c +++ b/arch/x86/mm/tlb.c | |||
@@ -183,18 +183,17 @@ static void flush_tlb_others_ipi(const struct cpumask *cpumask, | |||
183 | 183 | ||
184 | f->flush_mm = mm; | 184 | f->flush_mm = mm; |
185 | f->flush_va = va; | 185 | f->flush_va = va; |
186 | cpumask_andnot(to_cpumask(f->flush_cpumask), | 186 | if (cpumask_andnot(to_cpumask(f->flush_cpumask), cpumask, cpumask_of(smp_processor_id()))) { |
187 | cpumask, cpumask_of(smp_processor_id())); | 187 | /* |
188 | 188 | * We have to send the IPI only to | |
189 | /* | 189 | * CPUs affected. |
190 | * We have to send the IPI only to | 190 | */ |
191 | * CPUs affected. | 191 | apic->send_IPI_mask(to_cpumask(f->flush_cpumask), |
192 | */ | 192 | INVALIDATE_TLB_VECTOR_START + sender); |
193 | apic->send_IPI_mask(to_cpumask(f->flush_cpumask), | ||
194 | INVALIDATE_TLB_VECTOR_START + sender); | ||
195 | 193 | ||
196 | while (!cpumask_empty(to_cpumask(f->flush_cpumask))) | 194 | while (!cpumask_empty(to_cpumask(f->flush_cpumask))) |
197 | cpu_relax(); | 195 | cpu_relax(); |
196 | } | ||
198 | 197 | ||
199 | f->flush_mm = NULL; | 198 | f->flush_mm = NULL; |
200 | f->flush_va = 0; | 199 | f->flush_va = 0; |