aboutsummaryrefslogtreecommitdiffstats
path: root/include
diff options
context:
space:
mode:
authorLinus Torvalds <torvalds@woody.linux-foundation.org>2007-10-22 22:23:34 -0400
committerLinus Torvalds <torvalds@woody.linux-foundation.org>2007-10-22 22:23:34 -0400
commit56d61a0e26c5a61c66d1ac259a59960295939da9 (patch)
treea23a30a966fe4220060682179294087cba1f9c57 /include
parent5f48b338cd28f4095697a174d7e3e72084aca893 (diff)
parent190a1d722a59725706daf832bc8a511ed62f249d (diff)
Merge branch 'for-linus' of git://git390.osdl.marist.edu/pub/scm/linux-2.6
* 'for-linus' of git://git390.osdl.marist.edu/pub/scm/linux-2.6: [S390] 4level-fixup cleanup [S390] Cleanup page table definitions. [S390] Introduce follow_table in uaccess_pt.c [S390] Remove unused user_seg from thread structure. [S390] tlb flush fix. [S390] kernel: Fix dump on panic for DASDs under LPAR. [S390] struct class_device -> struct device conversion. [S390] cio: Fix incomplete commit for uevent suppression. [S390] cio: Use to_channelpath() for device to channel path conversion. [S390] Add per-cpu idle time / idle count sysfs attributes. [S390] Update default configuration.
Diffstat (limited to 'include')
-rw-r--r--include/asm-s390/cpu.h25
-rw-r--r--include/asm-s390/mmu_context.h50
-rw-r--r--include/asm-s390/page.h4
-rw-r--r--include/asm-s390/pgalloc.h250
-rw-r--r--include/asm-s390/pgtable.h429
-rw-r--r--include/asm-s390/processor.h20
-rw-r--r--include/asm-s390/tlb.h129
-rw-r--r--include/asm-s390/tlbflush.h152
8 files changed, 541 insertions, 518 deletions
diff --git a/include/asm-s390/cpu.h b/include/asm-s390/cpu.h
new file mode 100644
index 000000000000..352dde194f3c
--- /dev/null
+++ b/include/asm-s390/cpu.h
@@ -0,0 +1,25 @@
1/*
2 * include/asm-s390/cpu.h
3 *
4 * Copyright IBM Corp. 2007
5 * Author(s): Heiko Carstens <heiko.carstens@de.ibm.com>
6 */
7
8#ifndef _ASM_S390_CPU_H_
9#define _ASM_S390_CPU_H_
10
11#include <linux/types.h>
12#include <linux/percpu.h>
13#include <linux/spinlock.h>
14
15struct s390_idle_data {
16 spinlock_t lock;
17 unsigned int in_idle;
18 unsigned long long idle_count;
19 unsigned long long idle_enter;
20 unsigned long long idle_time;
21};
22
23DECLARE_PER_CPU(struct s390_idle_data, s390_idle);
24
25#endif /* _ASM_S390_CPU_H_ */
diff --git a/include/asm-s390/mmu_context.h b/include/asm-s390/mmu_context.h
index 501cb9b06314..05b842126b99 100644
--- a/include/asm-s390/mmu_context.h
+++ b/include/asm-s390/mmu_context.h
@@ -21,45 +21,43 @@
21 21
22#ifndef __s390x__ 22#ifndef __s390x__
23#define LCTL_OPCODE "lctl" 23#define LCTL_OPCODE "lctl"
24#define PGTABLE_BITS (_SEGMENT_TABLE|USER_STD_MASK)
25#else 24#else
26#define LCTL_OPCODE "lctlg" 25#define LCTL_OPCODE "lctlg"
27#define PGTABLE_BITS (_REGION_TABLE|USER_STD_MASK)
28#endif 26#endif
29 27
30static inline void enter_lazy_tlb(struct mm_struct *mm, 28static inline void update_mm(struct mm_struct *mm, struct task_struct *tsk)
31 struct task_struct *tsk)
32{ 29{
30 pgd_t *pgd = mm->pgd;
31 unsigned long asce_bits;
32
33 /* Calculate asce bits from the first pgd table entry. */
34 asce_bits = _ASCE_TABLE_LENGTH | _ASCE_USER_BITS;
35#ifdef CONFIG_64BIT
36 asce_bits |= _ASCE_TYPE_REGION3;
37#endif
38 S390_lowcore.user_asce = asce_bits | __pa(pgd);
39 if (switch_amode) {
40 /* Load primary space page table origin. */
41 pgd_t *shadow_pgd = get_shadow_table(pgd) ? : pgd;
42 S390_lowcore.user_exec_asce = asce_bits | __pa(shadow_pgd);
43 asm volatile(LCTL_OPCODE" 1,1,%0\n"
44 : : "m" (S390_lowcore.user_exec_asce) );
45 } else
46 /* Load home space page table origin. */
47 asm volatile(LCTL_OPCODE" 13,13,%0"
48 : : "m" (S390_lowcore.user_asce) );
33} 49}
34 50
35static inline void switch_mm(struct mm_struct *prev, struct mm_struct *next, 51static inline void switch_mm(struct mm_struct *prev, struct mm_struct *next,
36 struct task_struct *tsk) 52 struct task_struct *tsk)
37{ 53{
38 pgd_t *shadow_pgd = get_shadow_pgd(next->pgd); 54 if (unlikely(prev == next))
39 55 return;
40 if (prev != next) {
41 S390_lowcore.user_asce = (__pa(next->pgd) & PAGE_MASK) |
42 PGTABLE_BITS;
43 if (shadow_pgd) {
44 /* Load primary/secondary space page table origin. */
45 S390_lowcore.user_exec_asce =
46 (__pa(shadow_pgd) & PAGE_MASK) | PGTABLE_BITS;
47 asm volatile(LCTL_OPCODE" 1,1,%0\n"
48 LCTL_OPCODE" 7,7,%1"
49 : : "m" (S390_lowcore.user_exec_asce),
50 "m" (S390_lowcore.user_asce) );
51 } else if (switch_amode) {
52 /* Load primary space page table origin. */
53 asm volatile(LCTL_OPCODE" 1,1,%0"
54 : : "m" (S390_lowcore.user_asce) );
55 } else
56 /* Load home space page table origin. */
57 asm volatile(LCTL_OPCODE" 13,13,%0"
58 : : "m" (S390_lowcore.user_asce) );
59 }
60 cpu_set(smp_processor_id(), next->cpu_vm_mask); 56 cpu_set(smp_processor_id(), next->cpu_vm_mask);
57 update_mm(next, tsk);
61} 58}
62 59
60#define enter_lazy_tlb(mm,tsk) do { } while (0)
63#define deactivate_mm(tsk,mm) do { } while (0) 61#define deactivate_mm(tsk,mm) do { } while (0)
64 62
65static inline void activate_mm(struct mm_struct *prev, 63static inline void activate_mm(struct mm_struct *prev,
diff --git a/include/asm-s390/page.h b/include/asm-s390/page.h
index ceec3826a67c..584d0ee3c7f6 100644
--- a/include/asm-s390/page.h
+++ b/include/asm-s390/page.h
@@ -82,6 +82,7 @@ typedef struct { unsigned long pte; } pte_t;
82#ifndef __s390x__ 82#ifndef __s390x__
83 83
84typedef struct { unsigned long pmd; } pmd_t; 84typedef struct { unsigned long pmd; } pmd_t;
85typedef struct { unsigned long pud; } pud_t;
85typedef struct { 86typedef struct {
86 unsigned long pgd0; 87 unsigned long pgd0;
87 unsigned long pgd1; 88 unsigned long pgd1;
@@ -90,6 +91,7 @@ typedef struct {
90 } pgd_t; 91 } pgd_t;
91 92
92#define pmd_val(x) ((x).pmd) 93#define pmd_val(x) ((x).pmd)
94#define pud_val(x) ((x).pud)
93#define pgd_val(x) ((x).pgd0) 95#define pgd_val(x) ((x).pgd0)
94 96
95#else /* __s390x__ */ 97#else /* __s390x__ */
@@ -98,10 +100,12 @@ typedef struct {
98 unsigned long pmd0; 100 unsigned long pmd0;
99 unsigned long pmd1; 101 unsigned long pmd1;
100 } pmd_t; 102 } pmd_t;
103typedef struct { unsigned long pud; } pud_t;
101typedef struct { unsigned long pgd; } pgd_t; 104typedef struct { unsigned long pgd; } pgd_t;
102 105
103#define pmd_val(x) ((x).pmd0) 106#define pmd_val(x) ((x).pmd0)
104#define pmd_val1(x) ((x).pmd1) 107#define pmd_val1(x) ((x).pmd1)
108#define pud_val(x) ((x).pud)
105#define pgd_val(x) ((x).pgd) 109#define pgd_val(x) ((x).pgd)
106 110
107#endif /* __s390x__ */ 111#endif /* __s390x__ */
diff --git a/include/asm-s390/pgalloc.h b/include/asm-s390/pgalloc.h
index e45d3c9a4b7e..709dd1740956 100644
--- a/include/asm-s390/pgalloc.h
+++ b/include/asm-s390/pgalloc.h
@@ -19,140 +19,115 @@
19 19
20#define check_pgt_cache() do {} while (0) 20#define check_pgt_cache() do {} while (0)
21 21
22/* 22unsigned long *crst_table_alloc(struct mm_struct *, int);
23 * Page allocation orders. 23void crst_table_free(unsigned long *);
24 */
25#ifndef __s390x__
26# define PTE_ALLOC_ORDER 0
27# define PMD_ALLOC_ORDER 0
28# define PGD_ALLOC_ORDER 1
29#else /* __s390x__ */
30# define PTE_ALLOC_ORDER 0
31# define PMD_ALLOC_ORDER 2
32# define PGD_ALLOC_ORDER 2
33#endif /* __s390x__ */
34 24
35/* 25unsigned long *page_table_alloc(int);
36 * Allocate and free page tables. The xxx_kernel() versions are 26void page_table_free(unsigned long *);
37 * used to allocate a kernel page table - this turns on ASN bits
38 * if any.
39 */
40 27
41static inline pgd_t *pgd_alloc(struct mm_struct *mm) 28static inline void clear_table(unsigned long *s, unsigned long val, size_t n)
42{ 29{
43 pgd_t *pgd = (pgd_t *) __get_free_pages(GFP_KERNEL, PGD_ALLOC_ORDER); 30 *s = val;
44 int i; 31 n = (n / 256) - 1;
45 32 asm volatile(
46 if (!pgd) 33#ifdef CONFIG_64BIT
47 return NULL; 34 " mvc 8(248,%0),0(%0)\n"
48 if (s390_noexec) {
49 pgd_t *shadow_pgd = (pgd_t *)
50 __get_free_pages(GFP_KERNEL, PGD_ALLOC_ORDER);
51 struct page *page = virt_to_page(pgd);
52
53 if (!shadow_pgd) {
54 free_pages((unsigned long) pgd, PGD_ALLOC_ORDER);
55 return NULL;
56 }
57 page->lru.next = (void *) shadow_pgd;
58 }
59 for (i = 0; i < PTRS_PER_PGD; i++)
60#ifndef __s390x__
61 pmd_clear(pmd_offset(pgd + i, i*PGDIR_SIZE));
62#else 35#else
63 pgd_clear(pgd + i); 36 " mvc 4(252,%0),0(%0)\n"
64#endif 37#endif
65 return pgd; 38 "0: mvc 256(256,%0),0(%0)\n"
39 " la %0,256(%0)\n"
40 " brct %1,0b\n"
41 : "+a" (s), "+d" (n));
66} 42}
67 43
68static inline void pgd_free(pgd_t *pgd) 44static inline void crst_table_init(unsigned long *crst, unsigned long entry)
69{ 45{
70 pgd_t *shadow_pgd = get_shadow_pgd(pgd); 46 clear_table(crst, entry, sizeof(unsigned long)*2048);
71 47 crst = get_shadow_table(crst);
72 if (shadow_pgd) 48 if (crst)
73 free_pages((unsigned long) shadow_pgd, PGD_ALLOC_ORDER); 49 clear_table(crst, entry, sizeof(unsigned long)*2048);
74 free_pages((unsigned long) pgd, PGD_ALLOC_ORDER);
75} 50}
76 51
77#ifndef __s390x__ 52#ifndef __s390x__
78/* 53
79 * page middle directory allocation/free routines. 54static inline unsigned long pgd_entry_type(struct mm_struct *mm)
80 * We use pmd cache only on s390x, so these are dummy routines. This
81 * code never triggers because the pgd will always be present.
82 */
83#define pmd_alloc_one(mm,address) ({ BUG(); ((pmd_t *)2); })
84#define pmd_free(x) do { } while (0)
85#define __pmd_free_tlb(tlb,x) do { } while (0)
86#define pgd_populate(mm, pmd, pte) BUG()
87#define pgd_populate_kernel(mm, pmd, pte) BUG()
88#else /* __s390x__ */
89static inline pmd_t * pmd_alloc_one(struct mm_struct *mm, unsigned long vmaddr)
90{ 55{
91 pmd_t *pmd = (pmd_t *) __get_free_pages(GFP_KERNEL, PMD_ALLOC_ORDER); 56 return _SEGMENT_ENTRY_EMPTY;
92 int i;
93
94 if (!pmd)
95 return NULL;
96 if (s390_noexec) {
97 pmd_t *shadow_pmd = (pmd_t *)
98 __get_free_pages(GFP_KERNEL, PMD_ALLOC_ORDER);
99 struct page *page = virt_to_page(pmd);
100
101 if (!shadow_pmd) {
102 free_pages((unsigned long) pmd, PMD_ALLOC_ORDER);
103 return NULL;
104 }
105 page->lru.next = (void *) shadow_pmd;
106 }
107 for (i=0; i < PTRS_PER_PMD; i++)
108 pmd_clear(pmd + i);
109 return pmd;
110} 57}
111 58
112static inline void pmd_free (pmd_t *pmd) 59#define pud_alloc_one(mm,address) ({ BUG(); ((pud_t *)2); })
60#define pud_free(x) do { } while (0)
61
62#define pmd_alloc_one(mm,address) ({ BUG(); ((pmd_t *)2); })
63#define pmd_free(x) do { } while (0)
64
65#define pgd_populate(mm, pgd, pud) BUG()
66#define pgd_populate_kernel(mm, pgd, pud) BUG()
67
68#define pud_populate(mm, pud, pmd) BUG()
69#define pud_populate_kernel(mm, pud, pmd) BUG()
70
71#else /* __s390x__ */
72
73static inline unsigned long pgd_entry_type(struct mm_struct *mm)
113{ 74{
114 pmd_t *shadow_pmd = get_shadow_pmd(pmd); 75 return _REGION3_ENTRY_EMPTY;
76}
77
78#define pud_alloc_one(mm,address) ({ BUG(); ((pud_t *)2); })
79#define pud_free(x) do { } while (0)
115 80
116 if (shadow_pmd) 81static inline pmd_t *pmd_alloc_one(struct mm_struct *mm, unsigned long vmaddr)
117 free_pages((unsigned long) shadow_pmd, PMD_ALLOC_ORDER); 82{
118 free_pages((unsigned long) pmd, PMD_ALLOC_ORDER); 83 unsigned long *crst = crst_table_alloc(mm, s390_noexec);
84 if (crst)
85 crst_table_init(crst, _SEGMENT_ENTRY_EMPTY);
86 return (pmd_t *) crst;
119} 87}
88#define pmd_free(pmd) crst_table_free((unsigned long *) pmd)
120 89
121#define __pmd_free_tlb(tlb,pmd) \ 90#define pgd_populate(mm, pgd, pud) BUG()
122 do { \ 91#define pgd_populate_kernel(mm, pgd, pud) BUG()
123 tlb_flush_mmu(tlb, 0, 0); \
124 pmd_free(pmd); \
125 } while (0)
126 92
127static inline void 93static inline void pud_populate_kernel(struct mm_struct *mm,
128pgd_populate_kernel(struct mm_struct *mm, pgd_t *pgd, pmd_t *pmd) 94 pud_t *pud, pmd_t *pmd)
129{ 95{
130 pgd_val(*pgd) = _PGD_ENTRY | __pa(pmd); 96 pud_val(*pud) = _REGION3_ENTRY | __pa(pmd);
131} 97}
132 98
133static inline void pgd_populate(struct mm_struct *mm, pgd_t *pgd, pmd_t *pmd) 99static inline void pud_populate(struct mm_struct *mm, pud_t *pud, pmd_t *pmd)
134{ 100{
135 pgd_t *shadow_pgd = get_shadow_pgd(pgd); 101 pud_t *shadow_pud = get_shadow_table(pud);
136 pmd_t *shadow_pmd = get_shadow_pmd(pmd); 102 pmd_t *shadow_pmd = get_shadow_table(pmd);
137 103
138 if (shadow_pgd && shadow_pmd) 104 if (shadow_pud && shadow_pmd)
139 pgd_populate_kernel(mm, shadow_pgd, shadow_pmd); 105 pud_populate_kernel(mm, shadow_pud, shadow_pmd);
140 pgd_populate_kernel(mm, pgd, pmd); 106 pud_populate_kernel(mm, pud, pmd);
141} 107}
142 108
143#endif /* __s390x__ */ 109#endif /* __s390x__ */
144 110
111static inline pgd_t *pgd_alloc(struct mm_struct *mm)
112{
113 unsigned long *crst = crst_table_alloc(mm, s390_noexec);
114 if (crst)
115 crst_table_init(crst, pgd_entry_type(mm));
116 return (pgd_t *) crst;
117}
118#define pgd_free(pgd) crst_table_free((unsigned long *) pgd)
119
145static inline void 120static inline void
146pmd_populate_kernel(struct mm_struct *mm, pmd_t *pmd, pte_t *pte) 121pmd_populate_kernel(struct mm_struct *mm, pmd_t *pmd, pte_t *pte)
147{ 122{
148#ifndef __s390x__ 123#ifndef __s390x__
149 pmd_val(pmd[0]) = _PAGE_TABLE + __pa(pte); 124 pmd_val(pmd[0]) = _SEGMENT_ENTRY + __pa(pte);
150 pmd_val(pmd[1]) = _PAGE_TABLE + __pa(pte+256); 125 pmd_val(pmd[1]) = _SEGMENT_ENTRY + __pa(pte+256);
151 pmd_val(pmd[2]) = _PAGE_TABLE + __pa(pte+512); 126 pmd_val(pmd[2]) = _SEGMENT_ENTRY + __pa(pte+512);
152 pmd_val(pmd[3]) = _PAGE_TABLE + __pa(pte+768); 127 pmd_val(pmd[3]) = _SEGMENT_ENTRY + __pa(pte+768);
153#else /* __s390x__ */ 128#else /* __s390x__ */
154 pmd_val(*pmd) = _PMD_ENTRY + __pa(pte); 129 pmd_val(*pmd) = _SEGMENT_ENTRY + __pa(pte);
155 pmd_val1(*pmd) = _PMD_ENTRY + __pa(pte+256); 130 pmd_val1(*pmd) = _SEGMENT_ENTRY + __pa(pte+256);
156#endif /* __s390x__ */ 131#endif /* __s390x__ */
157} 132}
158 133
@@ -160,7 +135,7 @@ static inline void
160pmd_populate(struct mm_struct *mm, pmd_t *pmd, struct page *page) 135pmd_populate(struct mm_struct *mm, pmd_t *pmd, struct page *page)
161{ 136{
162 pte_t *pte = (pte_t *)page_to_phys(page); 137 pte_t *pte = (pte_t *)page_to_phys(page);
163 pmd_t *shadow_pmd = get_shadow_pmd(pmd); 138 pmd_t *shadow_pmd = get_shadow_table(pmd);
164 pte_t *shadow_pte = get_shadow_pte(pte); 139 pte_t *shadow_pte = get_shadow_pte(pte);
165 140
166 pmd_populate_kernel(mm, pmd, pte); 141 pmd_populate_kernel(mm, pmd, pte);
@@ -171,67 +146,14 @@ pmd_populate(struct mm_struct *mm, pmd_t *pmd, struct page *page)
171/* 146/*
172 * page table entry allocation/free routines. 147 * page table entry allocation/free routines.
173 */ 148 */
174static inline pte_t * 149#define pte_alloc_one_kernel(mm, vmaddr) \
175pte_alloc_one_kernel(struct mm_struct *mm, unsigned long vmaddr) 150 ((pte_t *) page_table_alloc(s390_noexec))
176{ 151#define pte_alloc_one(mm, vmaddr) \
177 pte_t *pte = (pte_t *) __get_free_page(GFP_KERNEL|__GFP_REPEAT); 152 virt_to_page(page_table_alloc(s390_noexec))
178 int i; 153
179 154#define pte_free_kernel(pte) \
180 if (!pte) 155 page_table_free((unsigned long *) pte)
181 return NULL; 156#define pte_free(pte) \
182 if (s390_noexec) { 157 page_table_free((unsigned long *) page_to_phys((struct page *) pte))
183 pte_t *shadow_pte = (pte_t *)
184 __get_free_page(GFP_KERNEL|__GFP_REPEAT);
185 struct page *page = virt_to_page(pte);
186
187 if (!shadow_pte) {
188 free_page((unsigned long) pte);
189 return NULL;
190 }
191 page->lru.next = (void *) shadow_pte;
192 }
193 for (i=0; i < PTRS_PER_PTE; i++) {
194 pte_clear(mm, vmaddr, pte + i);
195 vmaddr += PAGE_SIZE;
196 }
197 return pte;
198}
199
200static inline struct page *
201pte_alloc_one(struct mm_struct *mm, unsigned long vmaddr)
202{
203 pte_t *pte = pte_alloc_one_kernel(mm, vmaddr);
204 if (pte)
205 return virt_to_page(pte);
206 return NULL;
207}
208
209static inline void pte_free_kernel(pte_t *pte)
210{
211 pte_t *shadow_pte = get_shadow_pte(pte);
212
213 if (shadow_pte)
214 free_page((unsigned long) shadow_pte);
215 free_page((unsigned long) pte);
216}
217
218static inline void pte_free(struct page *pte)
219{
220 struct page *shadow_page = get_shadow_page(pte);
221
222 if (shadow_page)
223 __free_page(shadow_page);
224 __free_page(pte);
225}
226
227#define __pte_free_tlb(tlb, pte) \
228({ \
229 struct mmu_gather *__tlb = (tlb); \
230 struct page *__pte = (pte); \
231 struct page *shadow_page = get_shadow_page(__pte); \
232 if (shadow_page) \
233 tlb_remove_page(__tlb, shadow_page); \
234 tlb_remove_page(__tlb, __pte); \
235})
236 158
237#endif /* _S390_PGALLOC_H */ 159#endif /* _S390_PGALLOC_H */
diff --git a/include/asm-s390/pgtable.h b/include/asm-s390/pgtable.h
index 39bb5192dc31..f2cc25b74adf 100644
--- a/include/asm-s390/pgtable.h
+++ b/include/asm-s390/pgtable.h
@@ -13,8 +13,6 @@
13#ifndef _ASM_S390_PGTABLE_H 13#ifndef _ASM_S390_PGTABLE_H
14#define _ASM_S390_PGTABLE_H 14#define _ASM_S390_PGTABLE_H
15 15
16#include <asm-generic/4level-fixup.h>
17
18/* 16/*
19 * The Linux memory management assumes a three-level page table setup. For 17 * The Linux memory management assumes a three-level page table setup. For
20 * s390 31 bit we "fold" the mid level into the top-level page table, so 18 * s390 31 bit we "fold" the mid level into the top-level page table, so
@@ -35,9 +33,6 @@
35#include <asm/bug.h> 33#include <asm/bug.h>
36#include <asm/processor.h> 34#include <asm/processor.h>
37 35
38struct vm_area_struct; /* forward declaration (include/linux/mm.h) */
39struct mm_struct;
40
41extern pgd_t swapper_pg_dir[] __attribute__ ((aligned (4096))); 36extern pgd_t swapper_pg_dir[] __attribute__ ((aligned (4096)));
42extern void paging_init(void); 37extern void paging_init(void);
43extern void vmem_map_init(void); 38extern void vmem_map_init(void);
@@ -63,14 +58,18 @@ extern char empty_zero_page[PAGE_SIZE];
63 */ 58 */
64#ifndef __s390x__ 59#ifndef __s390x__
65# define PMD_SHIFT 22 60# define PMD_SHIFT 22
61# define PUD_SHIFT 22
66# define PGDIR_SHIFT 22 62# define PGDIR_SHIFT 22
67#else /* __s390x__ */ 63#else /* __s390x__ */
68# define PMD_SHIFT 21 64# define PMD_SHIFT 21
65# define PUD_SHIFT 31
69# define PGDIR_SHIFT 31 66# define PGDIR_SHIFT 31
70#endif /* __s390x__ */ 67#endif /* __s390x__ */
71 68
72#define PMD_SIZE (1UL << PMD_SHIFT) 69#define PMD_SIZE (1UL << PMD_SHIFT)
73#define PMD_MASK (~(PMD_SIZE-1)) 70#define PMD_MASK (~(PMD_SIZE-1))
71#define PUD_SIZE (1UL << PUD_SHIFT)
72#define PUD_MASK (~(PUD_SIZE-1))
74#define PGDIR_SIZE (1UL << PGDIR_SHIFT) 73#define PGDIR_SIZE (1UL << PGDIR_SHIFT)
75#define PGDIR_MASK (~(PGDIR_SIZE-1)) 74#define PGDIR_MASK (~(PGDIR_SIZE-1))
76 75
@@ -83,10 +82,12 @@ extern char empty_zero_page[PAGE_SIZE];
83#ifndef __s390x__ 82#ifndef __s390x__
84# define PTRS_PER_PTE 1024 83# define PTRS_PER_PTE 1024
85# define PTRS_PER_PMD 1 84# define PTRS_PER_PMD 1
85# define PTRS_PER_PUD 1
86# define PTRS_PER_PGD 512 86# define PTRS_PER_PGD 512
87#else /* __s390x__ */ 87#else /* __s390x__ */
88# define PTRS_PER_PTE 512 88# define PTRS_PER_PTE 512
89# define PTRS_PER_PMD 1024 89# define PTRS_PER_PMD 1024
90# define PTRS_PER_PUD 1
90# define PTRS_PER_PGD 2048 91# define PTRS_PER_PGD 2048
91#endif /* __s390x__ */ 92#endif /* __s390x__ */
92 93
@@ -96,6 +97,8 @@ extern char empty_zero_page[PAGE_SIZE];
96 printk("%s:%d: bad pte %p.\n", __FILE__, __LINE__, (void *) pte_val(e)) 97 printk("%s:%d: bad pte %p.\n", __FILE__, __LINE__, (void *) pte_val(e))
97#define pmd_ERROR(e) \ 98#define pmd_ERROR(e) \
98 printk("%s:%d: bad pmd %p.\n", __FILE__, __LINE__, (void *) pmd_val(e)) 99 printk("%s:%d: bad pmd %p.\n", __FILE__, __LINE__, (void *) pmd_val(e))
100#define pud_ERROR(e) \
101 printk("%s:%d: bad pud %p.\n", __FILE__, __LINE__, (void *) pud_val(e))
99#define pgd_ERROR(e) \ 102#define pgd_ERROR(e) \
100 printk("%s:%d: bad pgd %p.\n", __FILE__, __LINE__, (void *) pgd_val(e)) 103 printk("%s:%d: bad pgd %p.\n", __FILE__, __LINE__, (void *) pgd_val(e))
101 104
@@ -195,7 +198,7 @@ extern unsigned long vmalloc_end;
195 * I Segment-Invalid Bit: Segment is not available for address-translation 198 * I Segment-Invalid Bit: Segment is not available for address-translation
196 * TT Type 01 199 * TT Type 01
197 * TF 200 * TF
198 * TL Table lenght 201 * TL Table length
199 * 202 *
200 * The 64 bit regiontable origin of S390 has following format: 203 * The 64 bit regiontable origin of S390 has following format:
201 * | region table origon | DTTL 204 * | region table origon | DTTL
@@ -221,6 +224,8 @@ extern unsigned long vmalloc_end;
221/* Hardware bits in the page table entry */ 224/* Hardware bits in the page table entry */
222#define _PAGE_RO 0x200 /* HW read-only bit */ 225#define _PAGE_RO 0x200 /* HW read-only bit */
223#define _PAGE_INVALID 0x400 /* HW invalid bit */ 226#define _PAGE_INVALID 0x400 /* HW invalid bit */
227
228/* Software bits in the page table entry */
224#define _PAGE_SWT 0x001 /* SW pte type bit t */ 229#define _PAGE_SWT 0x001 /* SW pte type bit t */
225#define _PAGE_SWX 0x002 /* SW pte type bit x */ 230#define _PAGE_SWX 0x002 /* SW pte type bit x */
226 231
@@ -264,60 +269,75 @@ extern unsigned long vmalloc_end;
264 269
265#ifndef __s390x__ 270#ifndef __s390x__
266 271
267/* Bits in the segment table entry */ 272/* Bits in the segment table address-space-control-element */
268#define _PAGE_TABLE_LEN 0xf /* only full page-tables */ 273#define _ASCE_SPACE_SWITCH 0x80000000UL /* space switch event */
269#define _PAGE_TABLE_COM 0x10 /* common page-table */ 274#define _ASCE_ORIGIN_MASK 0x7ffff000UL /* segment table origin */
270#define _PAGE_TABLE_INV 0x20 /* invalid page-table */ 275#define _ASCE_PRIVATE_SPACE 0x100 /* private space control */
271#define _SEG_PRESENT 0x001 /* Software (overlap with PTL) */ 276#define _ASCE_ALT_EVENT 0x80 /* storage alteration event control */
272 277#define _ASCE_TABLE_LENGTH 0x7f /* 128 x 64 entries = 8k */
273/* Bits int the storage key */
274#define _PAGE_CHANGED 0x02 /* HW changed bit */
275#define _PAGE_REFERENCED 0x04 /* HW referenced bit */
276
277#define _USER_SEG_TABLE_LEN 0x7f /* user-segment-table up to 2 GB */
278#define _KERNEL_SEG_TABLE_LEN 0x7f /* kernel-segment-table up to 2 GB */
279
280/*
281 * User and Kernel pagetables are identical
282 */
283#define _PAGE_TABLE _PAGE_TABLE_LEN
284#define _KERNPG_TABLE _PAGE_TABLE_LEN
285
286/*
287 * The Kernel segment-tables includes the User segment-table
288 */
289 278
290#define _SEGMENT_TABLE (_USER_SEG_TABLE_LEN|0x80000000|0x100) 279/* Bits in the segment table entry */
291#define _KERNSEG_TABLE _KERNEL_SEG_TABLE_LEN 280#define _SEGMENT_ENTRY_ORIGIN 0x7fffffc0UL /* page table origin */
281#define _SEGMENT_ENTRY_INV 0x20 /* invalid segment table entry */
282#define _SEGMENT_ENTRY_COMMON 0x10 /* common segment bit */
283#define _SEGMENT_ENTRY_PTL 0x0f /* page table length */
292 284
293#define USER_STD_MASK 0x00000080UL 285#define _SEGMENT_ENTRY (_SEGMENT_ENTRY_PTL)
286#define _SEGMENT_ENTRY_EMPTY (_SEGMENT_ENTRY_INV)
294 287
295#else /* __s390x__ */ 288#else /* __s390x__ */
296 289
290/* Bits in the segment/region table address-space-control-element */
291#define _ASCE_ORIGIN ~0xfffUL/* segment table origin */
292#define _ASCE_PRIVATE_SPACE 0x100 /* private space control */
293#define _ASCE_ALT_EVENT 0x80 /* storage alteration event control */
294#define _ASCE_SPACE_SWITCH 0x40 /* space switch event */
295#define _ASCE_REAL_SPACE 0x20 /* real space control */
296#define _ASCE_TYPE_MASK 0x0c /* asce table type mask */
297#define _ASCE_TYPE_REGION1 0x0c /* region first table type */
298#define _ASCE_TYPE_REGION2 0x08 /* region second table type */
299#define _ASCE_TYPE_REGION3 0x04 /* region third table type */
300#define _ASCE_TYPE_SEGMENT 0x00 /* segment table type */
301#define _ASCE_TABLE_LENGTH 0x03 /* region table length */
302
303/* Bits in the region table entry */
304#define _REGION_ENTRY_ORIGIN ~0xfffUL/* region/segment table origin */
305#define _REGION_ENTRY_INV 0x20 /* invalid region table entry */
306#define _REGION_ENTRY_TYPE_MASK 0x0c /* region/segment table type mask */
307#define _REGION_ENTRY_TYPE_R1 0x0c /* region first table type */
308#define _REGION_ENTRY_TYPE_R2 0x08 /* region second table type */
309#define _REGION_ENTRY_TYPE_R3 0x04 /* region third table type */
310#define _REGION_ENTRY_LENGTH 0x03 /* region third length */
311
312#define _REGION1_ENTRY (_REGION_ENTRY_TYPE_R1 | _REGION_ENTRY_LENGTH)
313#define _REGION1_ENTRY_EMPTY (_REGION_ENTRY_TYPE_R1 | _REGION_ENTRY_INV)
314#define _REGION2_ENTRY (_REGION_ENTRY_TYPE_R2 | _REGION_ENTRY_LENGTH)
315#define _REGION2_ENTRY_EMPTY (_REGION_ENTRY_TYPE_R2 | _REGION_ENTRY_INV)
316#define _REGION3_ENTRY (_REGION_ENTRY_TYPE_R3 | _REGION_ENTRY_LENGTH)
317#define _REGION3_ENTRY_EMPTY (_REGION_ENTRY_TYPE_R3 | _REGION_ENTRY_INV)
318
297/* Bits in the segment table entry */ 319/* Bits in the segment table entry */
298#define _PMD_ENTRY_INV 0x20 /* invalid segment table entry */ 320#define _SEGMENT_ENTRY_ORIGIN ~0x7ffUL/* segment table origin */
299#define _PMD_ENTRY 0x00 321#define _SEGMENT_ENTRY_RO 0x200 /* page protection bit */
322#define _SEGMENT_ENTRY_INV 0x20 /* invalid segment table entry */
323
324#define _SEGMENT_ENTRY (0)
325#define _SEGMENT_ENTRY_EMPTY (_SEGMENT_ENTRY_INV)
300 326
301/* Bits in the region third table entry */ 327#endif /* __s390x__ */
302#define _PGD_ENTRY_INV 0x20 /* invalid region table entry */
303#define _PGD_ENTRY 0x07
304 328
305/* 329/*
306 * User and kernel page directory 330 * A user page table pointer has the space-switch-event bit, the
331 * private-space-control bit and the storage-alteration-event-control
332 * bit set. A kernel page table pointer doesn't need them.
307 */ 333 */
308#define _REGION_THIRD 0x4 334#define _ASCE_USER_BITS (_ASCE_SPACE_SWITCH | _ASCE_PRIVATE_SPACE | \
309#define _REGION_THIRD_LEN 0x3 335 _ASCE_ALT_EVENT)
310#define _REGION_TABLE (_REGION_THIRD|_REGION_THIRD_LEN|0x40|0x100)
311#define _KERN_REGION_TABLE (_REGION_THIRD|_REGION_THIRD_LEN)
312
313#define USER_STD_MASK 0x0000000000000080UL
314 336
315/* Bits in the storage key */ 337/* Bits int the storage key */
316#define _PAGE_CHANGED 0x02 /* HW changed bit */ 338#define _PAGE_CHANGED 0x02 /* HW changed bit */
317#define _PAGE_REFERENCED 0x04 /* HW referenced bit */ 339#define _PAGE_REFERENCED 0x04 /* HW referenced bit */
318 340
319#endif /* __s390x__ */
320
321/* 341/*
322 * Page protection definitions. 342 * Page protection definitions.
323 */ 343 */
@@ -358,65 +378,38 @@ extern unsigned long vmalloc_end;
358#define __S111 PAGE_EX_RW 378#define __S111 PAGE_EX_RW
359 379
360#ifndef __s390x__ 380#ifndef __s390x__
361# define PMD_SHADOW_SHIFT 1 381# define PxD_SHADOW_SHIFT 1
362# define PGD_SHADOW_SHIFT 1
363#else /* __s390x__ */ 382#else /* __s390x__ */
364# define PMD_SHADOW_SHIFT 2 383# define PxD_SHADOW_SHIFT 2
365# define PGD_SHADOW_SHIFT 2
366#endif /* __s390x__ */ 384#endif /* __s390x__ */
367 385
368static inline struct page *get_shadow_page(struct page *page) 386static inline struct page *get_shadow_page(struct page *page)
369{ 387{
370 if (s390_noexec && !list_empty(&page->lru)) 388 if (s390_noexec && page->index)
371 return virt_to_page(page->lru.next); 389 return virt_to_page((void *)(addr_t) page->index);
372 return NULL;
373}
374
375static inline pte_t *get_shadow_pte(pte_t *ptep)
376{
377 unsigned long pteptr = (unsigned long) (ptep);
378
379 if (s390_noexec) {
380 unsigned long offset = pteptr & (PAGE_SIZE - 1);
381 void *addr = (void *) (pteptr ^ offset);
382 struct page *page = virt_to_page(addr);
383 if (!list_empty(&page->lru))
384 return (pte_t *) ((unsigned long) page->lru.next |
385 offset);
386 }
387 return NULL; 390 return NULL;
388} 391}
389 392
390static inline pmd_t *get_shadow_pmd(pmd_t *pmdp) 393static inline void *get_shadow_pte(void *table)
391{ 394{
392 unsigned long pmdptr = (unsigned long) (pmdp); 395 unsigned long addr, offset;
396 struct page *page;
393 397
394 if (s390_noexec) { 398 addr = (unsigned long) table;
395 unsigned long offset = pmdptr & 399 offset = addr & (PAGE_SIZE - 1);
396 ((PAGE_SIZE << PMD_SHADOW_SHIFT) - 1); 400 page = virt_to_page((void *)(addr ^ offset));
397 void *addr = (void *) (pmdptr ^ offset); 401 return (void *)(addr_t)(page->index ? (page->index | offset) : 0UL);
398 struct page *page = virt_to_page(addr);
399 if (!list_empty(&page->lru))
400 return (pmd_t *) ((unsigned long) page->lru.next |
401 offset);
402 }
403 return NULL;
404} 402}
405 403
406static inline pgd_t *get_shadow_pgd(pgd_t *pgdp) 404static inline void *get_shadow_table(void *table)
407{ 405{
408 unsigned long pgdptr = (unsigned long) (pgdp); 406 unsigned long addr, offset;
407 struct page *page;
409 408
410 if (s390_noexec) { 409 addr = (unsigned long) table;
411 unsigned long offset = pgdptr & 410 offset = addr & ((PAGE_SIZE << PxD_SHADOW_SHIFT) - 1);
412 ((PAGE_SIZE << PGD_SHADOW_SHIFT) - 1); 411 page = virt_to_page((void *)(addr ^ offset));
413 void *addr = (void *) (pgdptr ^ offset); 412 return (void *)(addr_t)(page->index ? (page->index | offset) : 0UL);
414 struct page *page = virt_to_page(addr);
415 if (!list_empty(&page->lru))
416 return (pgd_t *) ((unsigned long) page->lru.next |
417 offset);
418 }
419 return NULL;
420} 413}
421 414
422/* 415/*
@@ -424,7 +417,8 @@ static inline pgd_t *get_shadow_pgd(pgd_t *pgdp)
424 * within a page table are directly modified. Thus, the following 417 * within a page table are directly modified. Thus, the following
425 * hook is made available. 418 * hook is made available.
426 */ 419 */
427static inline void set_pte(pte_t *pteptr, pte_t pteval) 420static inline void set_pte_at(struct mm_struct *mm, unsigned long addr,
421 pte_t *pteptr, pte_t pteval)
428{ 422{
429 pte_t *shadow_pte = get_shadow_pte(pteptr); 423 pte_t *shadow_pte = get_shadow_pte(pteptr);
430 424
@@ -437,7 +431,6 @@ static inline void set_pte(pte_t *pteptr, pte_t pteval)
437 pte_val(*shadow_pte) = _PAGE_TYPE_EMPTY; 431 pte_val(*shadow_pte) = _PAGE_TYPE_EMPTY;
438 } 432 }
439} 433}
440#define set_pte_at(mm,addr,ptep,pteval) set_pte(ptep,pteval)
441 434
442/* 435/*
443 * pgd/pmd/pte query functions 436 * pgd/pmd/pte query functions
@@ -448,47 +441,50 @@ static inline int pgd_present(pgd_t pgd) { return 1; }
448static inline int pgd_none(pgd_t pgd) { return 0; } 441static inline int pgd_none(pgd_t pgd) { return 0; }
449static inline int pgd_bad(pgd_t pgd) { return 0; } 442static inline int pgd_bad(pgd_t pgd) { return 0; }
450 443
451static inline int pmd_present(pmd_t pmd) { return pmd_val(pmd) & _SEG_PRESENT; } 444static inline int pud_present(pud_t pud) { return 1; }
452static inline int pmd_none(pmd_t pmd) { return pmd_val(pmd) & _PAGE_TABLE_INV; } 445static inline int pud_none(pud_t pud) { return 0; }
453static inline int pmd_bad(pmd_t pmd) 446static inline int pud_bad(pud_t pud) { return 0; }
454{
455 return (pmd_val(pmd) & (~PAGE_MASK & ~_PAGE_TABLE_INV)) != _PAGE_TABLE;
456}
457 447
458#else /* __s390x__ */ 448#else /* __s390x__ */
459 449
460static inline int pgd_present(pgd_t pgd) 450static inline int pgd_present(pgd_t pgd) { return 1; }
451static inline int pgd_none(pgd_t pgd) { return 0; }
452static inline int pgd_bad(pgd_t pgd) { return 0; }
453
454static inline int pud_present(pud_t pud)
461{ 455{
462 return (pgd_val(pgd) & ~PAGE_MASK) == _PGD_ENTRY; 456 return pud_val(pud) & _REGION_ENTRY_ORIGIN;
463} 457}
464 458
465static inline int pgd_none(pgd_t pgd) 459static inline int pud_none(pud_t pud)
466{ 460{
467 return pgd_val(pgd) & _PGD_ENTRY_INV; 461 return pud_val(pud) & _REGION_ENTRY_INV;
468} 462}
469 463
470static inline int pgd_bad(pgd_t pgd) 464static inline int pud_bad(pud_t pud)
471{ 465{
472 return (pgd_val(pgd) & (~PAGE_MASK & ~_PGD_ENTRY_INV)) != _PGD_ENTRY; 466 unsigned long mask = ~_REGION_ENTRY_ORIGIN & ~_REGION_ENTRY_INV;
467 return (pud_val(pud) & mask) != _REGION3_ENTRY;
473} 468}
474 469
470#endif /* __s390x__ */
471
475static inline int pmd_present(pmd_t pmd) 472static inline int pmd_present(pmd_t pmd)
476{ 473{
477 return (pmd_val(pmd) & ~PAGE_MASK) == _PMD_ENTRY; 474 return pmd_val(pmd) & _SEGMENT_ENTRY_ORIGIN;
478} 475}
479 476
480static inline int pmd_none(pmd_t pmd) 477static inline int pmd_none(pmd_t pmd)
481{ 478{
482 return pmd_val(pmd) & _PMD_ENTRY_INV; 479 return pmd_val(pmd) & _SEGMENT_ENTRY_INV;
483} 480}
484 481
485static inline int pmd_bad(pmd_t pmd) 482static inline int pmd_bad(pmd_t pmd)
486{ 483{
487 return (pmd_val(pmd) & (~PAGE_MASK & ~_PMD_ENTRY_INV)) != _PMD_ENTRY; 484 unsigned long mask = ~_SEGMENT_ENTRY_ORIGIN & ~_SEGMENT_ENTRY_INV;
485 return (pmd_val(pmd) & mask) != _SEGMENT_ENTRY;
488} 486}
489 487
490#endif /* __s390x__ */
491
492static inline int pte_none(pte_t pte) 488static inline int pte_none(pte_t pte)
493{ 489{
494 return (pte_val(pte) & _PAGE_INVALID) && !(pte_val(pte) & _PAGE_SWT); 490 return (pte_val(pte) & _PAGE_INVALID) && !(pte_val(pte) & _PAGE_SWT);
@@ -508,7 +504,8 @@ static inline int pte_file(pte_t pte)
508 return (pte_val(pte) & mask) == _PAGE_TYPE_FILE; 504 return (pte_val(pte) & mask) == _PAGE_TYPE_FILE;
509} 505}
510 506
511#define pte_same(a,b) (pte_val(a) == pte_val(b)) 507#define __HAVE_ARCH_PTE_SAME
508#define pte_same(a,b) (pte_val(a) == pte_val(b))
512 509
513/* 510/*
514 * query functions pte_write/pte_dirty/pte_young only work if 511 * query functions pte_write/pte_dirty/pte_young only work if
@@ -543,58 +540,52 @@ static inline int pte_young(pte_t pte)
543 540
544#ifndef __s390x__ 541#ifndef __s390x__
545 542
546static inline void pgd_clear(pgd_t * pgdp) { } 543#define pgd_clear(pgd) do { } while (0)
544#define pud_clear(pud) do { } while (0)
547 545
548static inline void pmd_clear_kernel(pmd_t * pmdp) 546static inline void pmd_clear_kernel(pmd_t * pmdp)
549{ 547{
550 pmd_val(pmdp[0]) = _PAGE_TABLE_INV; 548 pmd_val(pmdp[0]) = _SEGMENT_ENTRY_EMPTY;
551 pmd_val(pmdp[1]) = _PAGE_TABLE_INV; 549 pmd_val(pmdp[1]) = _SEGMENT_ENTRY_EMPTY;
552 pmd_val(pmdp[2]) = _PAGE_TABLE_INV; 550 pmd_val(pmdp[2]) = _SEGMENT_ENTRY_EMPTY;
553 pmd_val(pmdp[3]) = _PAGE_TABLE_INV; 551 pmd_val(pmdp[3]) = _SEGMENT_ENTRY_EMPTY;
554}
555
556static inline void pmd_clear(pmd_t * pmdp)
557{
558 pmd_t *shadow_pmd = get_shadow_pmd(pmdp);
559
560 pmd_clear_kernel(pmdp);
561 if (shadow_pmd)
562 pmd_clear_kernel(shadow_pmd);
563} 552}
564 553
565#else /* __s390x__ */ 554#else /* __s390x__ */
566 555
567static inline void pgd_clear_kernel(pgd_t * pgdp) 556#define pgd_clear(pgd) do { } while (0)
557
558static inline void pud_clear_kernel(pud_t *pud)
568{ 559{
569 pgd_val(*pgdp) = _PGD_ENTRY_INV | _PGD_ENTRY; 560 pud_val(*pud) = _REGION3_ENTRY_EMPTY;
570} 561}
571 562
572static inline void pgd_clear(pgd_t * pgdp) 563static inline void pud_clear(pud_t * pud)
573{ 564{
574 pgd_t *shadow_pgd = get_shadow_pgd(pgdp); 565 pud_t *shadow = get_shadow_table(pud);
575 566
576 pgd_clear_kernel(pgdp); 567 pud_clear_kernel(pud);
577 if (shadow_pgd) 568 if (shadow)
578 pgd_clear_kernel(shadow_pgd); 569 pud_clear_kernel(shadow);
579} 570}
580 571
581static inline void pmd_clear_kernel(pmd_t * pmdp) 572static inline void pmd_clear_kernel(pmd_t * pmdp)
582{ 573{
583 pmd_val(*pmdp) = _PMD_ENTRY_INV | _PMD_ENTRY; 574 pmd_val(*pmdp) = _SEGMENT_ENTRY_EMPTY;
584 pmd_val1(*pmdp) = _PMD_ENTRY_INV | _PMD_ENTRY; 575 pmd_val1(*pmdp) = _SEGMENT_ENTRY_EMPTY;
585} 576}
586 577
578#endif /* __s390x__ */
579
587static inline void pmd_clear(pmd_t * pmdp) 580static inline void pmd_clear(pmd_t * pmdp)
588{ 581{
589 pmd_t *shadow_pmd = get_shadow_pmd(pmdp); 582 pmd_t *shadow_pmd = get_shadow_table(pmdp);
590 583
591 pmd_clear_kernel(pmdp); 584 pmd_clear_kernel(pmdp);
592 if (shadow_pmd) 585 if (shadow_pmd)
593 pmd_clear_kernel(shadow_pmd); 586 pmd_clear_kernel(shadow_pmd);
594} 587}
595 588
596#endif /* __s390x__ */
597
598static inline void pte_clear(struct mm_struct *mm, unsigned long addr, pte_t *ptep) 589static inline void pte_clear(struct mm_struct *mm, unsigned long addr, pte_t *ptep)
599{ 590{
600 pte_t *shadow_pte = get_shadow_pte(ptep); 591 pte_t *shadow_pte = get_shadow_pte(ptep);
@@ -663,24 +654,19 @@ static inline pte_t pte_mkyoung(pte_t pte)
663 return pte; 654 return pte;
664} 655}
665 656
666static inline int ptep_test_and_clear_young(struct vm_area_struct *vma, unsigned long addr, pte_t *ptep) 657#define __HAVE_ARCH_PTEP_TEST_AND_CLEAR_YOUNG
658static inline int ptep_test_and_clear_young(struct vm_area_struct *vma,
659 unsigned long addr, pte_t *ptep)
667{ 660{
668 return 0; 661 return 0;
669} 662}
670 663
671static inline int 664#define __HAVE_ARCH_PTEP_CLEAR_YOUNG_FLUSH
672ptep_clear_flush_young(struct vm_area_struct *vma, 665static inline int ptep_clear_flush_young(struct vm_area_struct *vma,
673 unsigned long address, pte_t *ptep) 666 unsigned long address, pte_t *ptep)
674{ 667{
675 /* No need to flush TLB; bits are in storage key */ 668 /* No need to flush TLB; bits are in storage key */
676 return ptep_test_and_clear_young(vma, address, ptep); 669 return 0;
677}
678
679static inline pte_t ptep_get_and_clear(struct mm_struct *mm, unsigned long addr, pte_t *ptep)
680{
681 pte_t pte = *ptep;
682 pte_clear(mm, addr, ptep);
683 return pte;
684} 670}
685 671
686static inline void __ptep_ipte(unsigned long address, pte_t *ptep) 672static inline void __ptep_ipte(unsigned long address, pte_t *ptep)
@@ -709,6 +695,32 @@ static inline void ptep_invalidate(unsigned long address, pte_t *ptep)
709 __ptep_ipte(address, ptep); 695 __ptep_ipte(address, ptep);
710} 696}
711 697
698/*
699 * This is hard to understand. ptep_get_and_clear and ptep_clear_flush
700 * both clear the TLB for the unmapped pte. The reason is that
701 * ptep_get_and_clear is used in common code (e.g. change_pte_range)
702 * to modify an active pte. The sequence is
703 * 1) ptep_get_and_clear
704 * 2) set_pte_at
705 * 3) flush_tlb_range
706 * On s390 the tlb needs to get flushed with the modification of the pte
707 * if the pte is active. The only way how this can be implemented is to
708 * have ptep_get_and_clear do the tlb flush. In exchange flush_tlb_range
709 * is a nop.
710 */
711#define __HAVE_ARCH_PTEP_GET_AND_CLEAR
712#define ptep_get_and_clear(__mm, __address, __ptep) \
713({ \
714 pte_t __pte = *(__ptep); \
715 if (atomic_read(&(__mm)->mm_users) > 1 || \
716 (__mm) != current->active_mm) \
717 ptep_invalidate(__address, __ptep); \
718 else \
719 pte_clear((__mm), (__address), (__ptep)); \
720 __pte; \
721})
722
723#define __HAVE_ARCH_PTEP_CLEAR_FLUSH
712static inline pte_t ptep_clear_flush(struct vm_area_struct *vma, 724static inline pte_t ptep_clear_flush(struct vm_area_struct *vma,
713 unsigned long address, pte_t *ptep) 725 unsigned long address, pte_t *ptep)
714{ 726{
@@ -717,12 +729,40 @@ static inline pte_t ptep_clear_flush(struct vm_area_struct *vma,
717 return pte; 729 return pte;
718} 730}
719 731
720static inline void ptep_set_wrprotect(struct mm_struct *mm, unsigned long addr, pte_t *ptep) 732/*
733 * The batched pte unmap code uses ptep_get_and_clear_full to clear the
734 * ptes. Here an optimization is possible. tlb_gather_mmu flushes all
735 * tlbs of an mm if it can guarantee that the ptes of the mm_struct
736 * cannot be accessed while the batched unmap is running. In this case
737 * full==1 and a simple pte_clear is enough. See tlb.h.
738 */
739#define __HAVE_ARCH_PTEP_GET_AND_CLEAR_FULL
740static inline pte_t ptep_get_and_clear_full(struct mm_struct *mm,
741 unsigned long addr,
742 pte_t *ptep, int full)
721{ 743{
722 pte_t old_pte = *ptep; 744 pte_t pte = *ptep;
723 set_pte_at(mm, addr, ptep, pte_wrprotect(old_pte)); 745
746 if (full)
747 pte_clear(mm, addr, ptep);
748 else
749 ptep_invalidate(addr, ptep);
750 return pte;
724} 751}
725 752
753#define __HAVE_ARCH_PTEP_SET_WRPROTECT
754#define ptep_set_wrprotect(__mm, __addr, __ptep) \
755({ \
756 pte_t __pte = *(__ptep); \
757 if (pte_write(__pte)) { \
758 if (atomic_read(&(__mm)->mm_users) > 1 || \
759 (__mm) != current->active_mm) \
760 ptep_invalidate(__addr, __ptep); \
761 set_pte_at(__mm, __addr, __ptep, pte_wrprotect(__pte)); \
762 } \
763})
764
765#define __HAVE_ARCH_PTEP_SET_ACCESS_FLAGS
726#define ptep_set_access_flags(__vma, __addr, __ptep, __entry, __dirty) \ 766#define ptep_set_access_flags(__vma, __addr, __ptep, __entry, __dirty) \
727({ \ 767({ \
728 int __changed = !pte_same(*(__ptep), __entry); \ 768 int __changed = !pte_same(*(__ptep), __entry); \
@@ -740,11 +780,13 @@ static inline void ptep_set_wrprotect(struct mm_struct *mm, unsigned long addr,
740 * should therefore only be called if it is not mapped in any 780 * should therefore only be called if it is not mapped in any
741 * address space. 781 * address space.
742 */ 782 */
783#define __HAVE_ARCH_PAGE_TEST_DIRTY
743static inline int page_test_dirty(struct page *page) 784static inline int page_test_dirty(struct page *page)
744{ 785{
745 return (page_get_storage_key(page_to_phys(page)) & _PAGE_CHANGED) != 0; 786 return (page_get_storage_key(page_to_phys(page)) & _PAGE_CHANGED) != 0;
746} 787}
747 788
789#define __HAVE_ARCH_PAGE_CLEAR_DIRTY
748static inline void page_clear_dirty(struct page *page) 790static inline void page_clear_dirty(struct page *page)
749{ 791{
750 page_set_storage_key(page_to_phys(page), PAGE_DEFAULT_KEY); 792 page_set_storage_key(page_to_phys(page), PAGE_DEFAULT_KEY);
@@ -753,6 +795,7 @@ static inline void page_clear_dirty(struct page *page)
753/* 795/*
754 * Test and clear referenced bit in storage key. 796 * Test and clear referenced bit in storage key.
755 */ 797 */
798#define __HAVE_ARCH_PAGE_TEST_AND_CLEAR_YOUNG
756static inline int page_test_and_clear_young(struct page *page) 799static inline int page_test_and_clear_young(struct page *page)
757{ 800{
758 unsigned long physpage = page_to_phys(page); 801 unsigned long physpage = page_to_phys(page);
@@ -784,63 +827,48 @@ static inline pte_t mk_pte(struct page *page, pgprot_t pgprot)
784 return mk_pte_phys(physpage, pgprot); 827 return mk_pte_phys(physpage, pgprot);
785} 828}
786 829
787static inline pte_t pfn_pte(unsigned long pfn, pgprot_t pgprot) 830#define pgd_index(address) (((address) >> PGDIR_SHIFT) & (PTRS_PER_PGD-1))
788{ 831#define pud_index(address) (((address) >> PUD_SHIFT) & (PTRS_PER_PUD-1))
789 unsigned long physpage = __pa((pfn) << PAGE_SHIFT); 832#define pmd_index(address) (((address) >> PMD_SHIFT) & (PTRS_PER_PMD-1))
790 833#define pte_index(address) (((address) >> PAGE_SHIFT) & (PTRS_PER_PTE-1))
791 return mk_pte_phys(physpage, pgprot);
792}
793
794#ifdef __s390x__
795
796static inline pmd_t pfn_pmd(unsigned long pfn, pgprot_t pgprot)
797{
798 unsigned long physpage = __pa((pfn) << PAGE_SHIFT);
799
800 return __pmd(physpage + pgprot_val(pgprot));
801}
802
803#endif /* __s390x__ */
804
805#define pte_pfn(x) (pte_val(x) >> PAGE_SHIFT)
806#define pte_page(x) pfn_to_page(pte_pfn(x))
807 834
808#define pmd_page_vaddr(pmd) (pmd_val(pmd) & PAGE_MASK) 835#define pgd_offset(mm, address) ((mm)->pgd + pgd_index(address))
836#define pgd_offset_k(address) pgd_offset(&init_mm, address)
809 837
810#define pmd_page(pmd) pfn_to_page(pmd_val(pmd) >> PAGE_SHIFT) 838#ifndef __s390x__
811 839
812#define pgd_page_vaddr(pgd) (pgd_val(pgd) & PAGE_MASK) 840#define pmd_deref(pmd) (pmd_val(pmd) & _SEGMENT_ENTRY_ORIGIN)
841#define pud_deref(pmd) ({ BUG(); 0UL; })
842#define pgd_deref(pmd) ({ BUG(); 0UL; })
813 843
814#define pgd_page(pgd) pfn_to_page(pgd_val(pgd) >> PAGE_SHIFT) 844#define pud_offset(pgd, address) ((pud_t *) pgd)
845#define pmd_offset(pud, address) ((pmd_t *) pud + pmd_index(address))
815 846
816/* to find an entry in a page-table-directory */ 847#else /* __s390x__ */
817#define pgd_index(address) (((address) >> PGDIR_SHIFT) & (PTRS_PER_PGD-1))
818#define pgd_offset(mm, address) ((mm)->pgd+pgd_index(address))
819 848
820/* to find an entry in a kernel page-table-directory */ 849#define pmd_deref(pmd) (pmd_val(pmd) & _SEGMENT_ENTRY_ORIGIN)
821#define pgd_offset_k(address) pgd_offset(&init_mm, address) 850#define pud_deref(pud) (pud_val(pud) & _REGION_ENTRY_ORIGIN)
851#define pgd_deref(pgd) ({ BUG(); 0UL; })
822 852
823#ifndef __s390x__ 853#define pud_offset(pgd, address) ((pud_t *) pgd)
824 854
825/* Find an entry in the second-level page table.. */ 855static inline pmd_t *pmd_offset(pud_t *pud, unsigned long address)
826static inline pmd_t * pmd_offset(pgd_t * dir, unsigned long address)
827{ 856{
828 return (pmd_t *) dir; 857 pmd_t *pmd = (pmd_t *) pud_deref(*pud);
858 return pmd + pmd_index(address);
829} 859}
830 860
831#else /* __s390x__ */ 861#endif /* __s390x__ */
832 862
833/* Find an entry in the second-level page table.. */ 863#define pfn_pte(pfn,pgprot) mk_pte_phys(__pa((pfn) << PAGE_SHIFT),(pgprot))
834#define pmd_index(address) (((address) >> PMD_SHIFT) & (PTRS_PER_PMD-1)) 864#define pte_pfn(x) (pte_val(x) >> PAGE_SHIFT)
835#define pmd_offset(dir,addr) \ 865#define pte_page(x) pfn_to_page(pte_pfn(x))
836 ((pmd_t *) pgd_page_vaddr(*(dir)) + pmd_index(addr))
837 866
838#endif /* __s390x__ */ 867#define pmd_page(pmd) pfn_to_page(pmd_val(pmd) >> PAGE_SHIFT)
839 868
840/* Find an entry in the third-level page table.. */ 869/* Find an entry in the lowest level page table.. */
841#define pte_index(address) (((address) >> PAGE_SHIFT) & (PTRS_PER_PTE-1)) 870#define pte_offset(pmd, addr) ((pte_t *) pmd_deref(*(pmd)) + pte_index(addr))
842#define pte_offset_kernel(pmd, address) \ 871#define pte_offset_kernel(pmd, address) pte_offset(pmd,address)
843 ((pte_t *) pmd_page_vaddr(*(pmd)) + pte_index(address))
844#define pte_offset_map(pmd, address) pte_offset_kernel(pmd, address) 872#define pte_offset_map(pmd, address) pte_offset_kernel(pmd, address)
845#define pte_offset_map_nested(pmd, address) pte_offset_kernel(pmd, address) 873#define pte_offset_map_nested(pmd, address) pte_offset_kernel(pmd, address)
846#define pte_unmap(pte) do { } while (0) 874#define pte_unmap(pte) do { } while (0)
@@ -930,17 +958,6 @@ extern int remove_shared_memory(unsigned long start, unsigned long size);
930#define __HAVE_ARCH_MEMMAP_INIT 958#define __HAVE_ARCH_MEMMAP_INIT
931extern void memmap_init(unsigned long, int, unsigned long, unsigned long); 959extern void memmap_init(unsigned long, int, unsigned long, unsigned long);
932 960
933#define __HAVE_ARCH_PTEP_SET_ACCESS_FLAGS
934#define __HAVE_ARCH_PTEP_TEST_AND_CLEAR_YOUNG
935#define __HAVE_ARCH_PTEP_CLEAR_YOUNG_FLUSH
936#define __HAVE_ARCH_PTEP_GET_AND_CLEAR
937#define __HAVE_ARCH_PTEP_CLEAR_FLUSH
938#define __HAVE_ARCH_PTEP_SET_WRPROTECT
939#define __HAVE_ARCH_PTE_SAME
940#define __HAVE_ARCH_PAGE_TEST_DIRTY
941#define __HAVE_ARCH_PAGE_CLEAR_DIRTY
942#define __HAVE_ARCH_PAGE_TEST_AND_CLEAR_YOUNG
943#include <asm-generic/pgtable.h> 961#include <asm-generic/pgtable.h>
944 962
945#endif /* _S390_PAGE_H */ 963#endif /* _S390_PAGE_H */
946
diff --git a/include/asm-s390/processor.h b/include/asm-s390/processor.h
index 3b972d4c6b29..21d40a19355e 100644
--- a/include/asm-s390/processor.h
+++ b/include/asm-s390/processor.h
@@ -93,7 +93,6 @@ struct thread_struct {
93 s390_fp_regs fp_regs; 93 s390_fp_regs fp_regs;
94 unsigned int acrs[NUM_ACRS]; 94 unsigned int acrs[NUM_ACRS];
95 unsigned long ksp; /* kernel stack pointer */ 95 unsigned long ksp; /* kernel stack pointer */
96 unsigned long user_seg; /* HSTD */
97 mm_segment_t mm_segment; 96 mm_segment_t mm_segment;
98 unsigned long prot_addr; /* address of protection-excep. */ 97 unsigned long prot_addr; /* address of protection-excep. */
99 unsigned int error_code; /* error-code of last prog-excep. */ 98 unsigned int error_code; /* error-code of last prog-excep. */
@@ -128,22 +127,9 @@ struct stack_frame {
128 127
129#define ARCH_MIN_TASKALIGN 8 128#define ARCH_MIN_TASKALIGN 8
130 129
131#ifndef __s390x__ 130#define INIT_THREAD { \
132# define __SWAPPER_PG_DIR __pa(&swapper_pg_dir[0]) + _SEGMENT_TABLE 131 .ksp = sizeof(init_stack) + (unsigned long) &init_stack, \
133#else /* __s390x__ */ 132}
134# define __SWAPPER_PG_DIR __pa(&swapper_pg_dir[0]) + _REGION_TABLE
135#endif /* __s390x__ */
136
137#define INIT_THREAD {{0,{{0},{0},{0},{0},{0},{0},{0},{0},{0},{0}, \
138 {0},{0},{0},{0},{0},{0}}}, \
139 {0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0}, \
140 sizeof(init_stack) + (unsigned long) &init_stack, \
141 __SWAPPER_PG_DIR, \
142 {0}, \
143 0,0,0, \
144 (per_struct) {{{{0,}}},0,0,0,0,{{0,}}}, \
145 0, 0 \
146}
147 133
148/* 134/*
149 * Do necessary setup to start up a new thread. 135 * Do necessary setup to start up a new thread.
diff --git a/include/asm-s390/tlb.h b/include/asm-s390/tlb.h
index 51bd957b85bd..618693cfc10f 100644
--- a/include/asm-s390/tlb.h
+++ b/include/asm-s390/tlb.h
@@ -2,19 +2,130 @@
2#define _S390_TLB_H 2#define _S390_TLB_H
3 3
4/* 4/*
5 * s390 doesn't need any special per-pte or 5 * TLB flushing on s390 is complicated. The following requirement
6 * per-vma handling.. 6 * from the principles of operation is the most arduous:
7 *
8 * "A valid table entry must not be changed while it is attached
9 * to any CPU and may be used for translation by that CPU except to
10 * (1) invalidate the entry by using INVALIDATE PAGE TABLE ENTRY,
11 * or INVALIDATE DAT TABLE ENTRY, (2) alter bits 56-63 of a page
12 * table entry, or (3) make a change by means of a COMPARE AND SWAP
13 * AND PURGE instruction that purges the TLB."
14 *
15 * The modification of a pte of an active mm struct therefore is
16 * a two step process: i) invalidate the pte, ii) store the new pte.
17 * This is true for the page protection bit as well.
18 * The only possible optimization is to flush at the beginning of
19 * a tlb_gather_mmu cycle if the mm_struct is currently not in use.
20 *
21 * Pages used for the page tables is a different story. FIXME: more
7 */ 22 */
8#define tlb_start_vma(tlb, vma) do { } while (0) 23
9#define tlb_end_vma(tlb, vma) do { } while (0) 24#include <linux/mm.h>
10#define __tlb_remove_tlb_entry(tlb, ptep, address) do { } while (0) 25#include <linux/swap.h>
26#include <asm/processor.h>
27#include <asm/pgalloc.h>
28#include <asm/smp.h>
29#include <asm/tlbflush.h>
30
31#ifndef CONFIG_SMP
32#define TLB_NR_PTRS 1
33#else
34#define TLB_NR_PTRS 508
35#endif
36
37struct mmu_gather {
38 struct mm_struct *mm;
39 unsigned int fullmm;
40 unsigned int nr_ptes;
41 unsigned int nr_pmds;
42 void *array[TLB_NR_PTRS];
43};
44
45DECLARE_PER_CPU(struct mmu_gather, mmu_gathers);
46
47static inline struct mmu_gather *tlb_gather_mmu(struct mm_struct *mm,
48 unsigned int full_mm_flush)
49{
50 struct mmu_gather *tlb = &get_cpu_var(mmu_gathers);
51
52 tlb->mm = mm;
53 tlb->fullmm = full_mm_flush || (num_online_cpus() == 1) ||
54 (atomic_read(&mm->mm_users) <= 1 && mm == current->active_mm);
55 tlb->nr_ptes = 0;
56 tlb->nr_pmds = TLB_NR_PTRS;
57 if (tlb->fullmm)
58 __tlb_flush_mm(mm);
59 return tlb;
60}
61
62static inline void tlb_flush_mmu(struct mmu_gather *tlb,
63 unsigned long start, unsigned long end)
64{
65 if (!tlb->fullmm && (tlb->nr_ptes > 0 || tlb->nr_pmds < TLB_NR_PTRS))
66 __tlb_flush_mm(tlb->mm);
67 while (tlb->nr_ptes > 0)
68 pte_free(tlb->array[--tlb->nr_ptes]);
69 while (tlb->nr_pmds < TLB_NR_PTRS)
70 pmd_free((pmd_t *) tlb->array[tlb->nr_pmds++]);
71}
72
73static inline void tlb_finish_mmu(struct mmu_gather *tlb,
74 unsigned long start, unsigned long end)
75{
76 tlb_flush_mmu(tlb, start, end);
77
78 /* keep the page table cache within bounds */
79 check_pgt_cache();
80
81 put_cpu_var(mmu_gathers);
82}
11 83
12/* 84/*
13 * .. because we flush the whole mm when it 85 * Release the page cache reference for a pte removed by
14 * fills up. 86 * tlb_ptep_clear_flush. In both flush modes the tlb fo a page cache page
87 * has already been freed, so just do free_page_and_swap_cache.
15 */ 88 */
16#define tlb_flush(tlb) flush_tlb_mm((tlb)->mm) 89static inline void tlb_remove_page(struct mmu_gather *tlb, struct page *page)
90{
91 free_page_and_swap_cache(page);
92}
17 93
18#include <asm-generic/tlb.h> 94/*
95 * pte_free_tlb frees a pte table and clears the CRSTE for the
96 * page table from the tlb.
97 */
98static inline void pte_free_tlb(struct mmu_gather *tlb, struct page *page)
99{
100 if (!tlb->fullmm) {
101 tlb->array[tlb->nr_ptes++] = page;
102 if (tlb->nr_ptes >= tlb->nr_pmds)
103 tlb_flush_mmu(tlb, 0, 0);
104 } else
105 pte_free(page);
106}
19 107
108/*
109 * pmd_free_tlb frees a pmd table and clears the CRSTE for the
110 * segment table entry from the tlb.
111 */
112static inline void pmd_free_tlb(struct mmu_gather *tlb, pmd_t *pmd)
113{
114#ifdef __s390x__
115 if (!tlb->fullmm) {
116 tlb->array[--tlb->nr_pmds] = (struct page *) pmd;
117 if (tlb->nr_ptes >= tlb->nr_pmds)
118 tlb_flush_mmu(tlb, 0, 0);
119 } else
120 pmd_free(pmd);
20#endif 121#endif
122}
123
124#define pud_free_tlb(tlb, pud) do { } while (0)
125
126#define tlb_start_vma(tlb, vma) do { } while (0)
127#define tlb_end_vma(tlb, vma) do { } while (0)
128#define tlb_remove_tlb_entry(tlb, ptep, addr) do { } while (0)
129#define tlb_migrate_finish(mm) do { } while (0)
130
131#endif /* _S390_TLB_H */
diff --git a/include/asm-s390/tlbflush.h b/include/asm-s390/tlbflush.h
index 6de2632a3e4f..a69bd2490d52 100644
--- a/include/asm-s390/tlbflush.h
+++ b/include/asm-s390/tlbflush.h
@@ -6,68 +6,19 @@
6#include <asm/pgalloc.h> 6#include <asm/pgalloc.h>
7 7
8/* 8/*
9 * TLB flushing: 9 * Flush all tlb entries on the local cpu.
10 *
11 * - flush_tlb() flushes the current mm struct TLBs
12 * - flush_tlb_all() flushes all processes TLBs
13 * - flush_tlb_mm(mm) flushes the specified mm context TLB's
14 * - flush_tlb_page(vma, vmaddr) flushes one page
15 * - flush_tlb_range(vma, start, end) flushes a range of pages
16 * - flush_tlb_kernel_range(start, end) flushes a range of kernel pages
17 */
18
19/*
20 * S/390 has three ways of flushing TLBs
21 * 'ptlb' does a flush of the local processor
22 * 'csp' flushes the TLBs on all PUs of a SMP
23 * 'ipte' invalidates a pte in a page table and flushes that out of
24 * the TLBs of all PUs of a SMP
25 */
26
27#define local_flush_tlb() \
28do { asm volatile("ptlb": : :"memory"); } while (0)
29
30#ifndef CONFIG_SMP
31
32/*
33 * We always need to flush, since s390 does not flush tlb
34 * on each context switch
35 */ 10 */
36 11static inline void __tlb_flush_local(void)
37static inline void flush_tlb(void)
38{ 12{
39 local_flush_tlb(); 13 asm volatile("ptlb" : : : "memory");
40} 14}
41static inline void flush_tlb_all(void)
42{
43 local_flush_tlb();
44}
45static inline void flush_tlb_mm(struct mm_struct *mm)
46{
47 local_flush_tlb();
48}
49static inline void flush_tlb_page(struct vm_area_struct *vma,
50 unsigned long addr)
51{
52 local_flush_tlb();
53}
54static inline void flush_tlb_range(struct vm_area_struct *vma,
55 unsigned long start, unsigned long end)
56{
57 local_flush_tlb();
58}
59
60#define flush_tlb_kernel_range(start, end) \
61 local_flush_tlb();
62
63#else
64 15
65#include <asm/smp.h> 16/*
66 17 * Flush all tlb entries on all cpus.
67extern void smp_ptlb_all(void); 18 */
68 19static inline void __tlb_flush_global(void)
69static inline void global_flush_tlb(void)
70{ 20{
21 extern void smp_ptlb_all(void);
71 register unsigned long reg2 asm("2"); 22 register unsigned long reg2 asm("2");
72 register unsigned long reg3 asm("3"); 23 register unsigned long reg3 asm("3");
73 register unsigned long reg4 asm("4"); 24 register unsigned long reg4 asm("4");
@@ -89,66 +40,75 @@ static inline void global_flush_tlb(void)
89} 40}
90 41
91/* 42/*
92 * We only have to do global flush of tlb if process run since last 43 * Flush all tlb entries of a page table on all cpus.
93 * flush on any other pu than current.
94 * If we have threads (mm->count > 1) we always do a global flush,
95 * since the process runs on more than one processor at the same time.
96 */ 44 */
45static inline void __tlb_flush_idte(pgd_t *pgd)
46{
47 asm volatile(
48 " .insn rrf,0xb98e0000,0,%0,%1,0"
49 : : "a" (2048), "a" (__pa(pgd) & PAGE_MASK) : "cc" );
50}
97 51
98static inline void __flush_tlb_mm(struct mm_struct * mm) 52static inline void __tlb_flush_mm(struct mm_struct * mm)
99{ 53{
100 cpumask_t local_cpumask; 54 cpumask_t local_cpumask;
101 55
102 if (unlikely(cpus_empty(mm->cpu_vm_mask))) 56 if (unlikely(cpus_empty(mm->cpu_vm_mask)))
103 return; 57 return;
58 /*
59 * If the machine has IDTE we prefer to do a per mm flush
60 * on all cpus instead of doing a local flush if the mm
61 * only ran on the local cpu.
62 */
104 if (MACHINE_HAS_IDTE) { 63 if (MACHINE_HAS_IDTE) {
105 pgd_t *shadow_pgd = get_shadow_pgd(mm->pgd); 64 pgd_t *shadow_pgd = get_shadow_table(mm->pgd);
106 65
107 if (shadow_pgd) { 66 if (shadow_pgd)
108 asm volatile( 67 __tlb_flush_idte(shadow_pgd);
109 " .insn rrf,0xb98e0000,0,%0,%1,0" 68 __tlb_flush_idte(mm->pgd);
110 : : "a" (2048),
111 "a" (__pa(shadow_pgd) & PAGE_MASK) : "cc" );
112 }
113 asm volatile(
114 " .insn rrf,0xb98e0000,0,%0,%1,0"
115 : : "a" (2048), "a" (__pa(mm->pgd)&PAGE_MASK) : "cc");
116 return; 69 return;
117 } 70 }
118 preempt_disable(); 71 preempt_disable();
72 /*
73 * If the process only ran on the local cpu, do a local flush.
74 */
119 local_cpumask = cpumask_of_cpu(smp_processor_id()); 75 local_cpumask = cpumask_of_cpu(smp_processor_id());
120 if (cpus_equal(mm->cpu_vm_mask, local_cpumask)) 76 if (cpus_equal(mm->cpu_vm_mask, local_cpumask))
121 local_flush_tlb(); 77 __tlb_flush_local();
122 else 78 else
123 global_flush_tlb(); 79 __tlb_flush_global();
124 preempt_enable(); 80 preempt_enable();
125} 81}
126 82
127static inline void flush_tlb(void) 83static inline void __tlb_flush_mm_cond(struct mm_struct * mm)
128{
129 __flush_tlb_mm(current->mm);
130}
131static inline void flush_tlb_all(void)
132{
133 global_flush_tlb();
134}
135static inline void flush_tlb_mm(struct mm_struct *mm)
136{
137 __flush_tlb_mm(mm);
138}
139static inline void flush_tlb_page(struct vm_area_struct *vma,
140 unsigned long addr)
141{
142 __flush_tlb_mm(vma->vm_mm);
143}
144static inline void flush_tlb_range(struct vm_area_struct *vma,
145 unsigned long start, unsigned long end)
146{ 84{
147 __flush_tlb_mm(vma->vm_mm); 85 if (atomic_read(&mm->mm_users) <= 1 && mm == current->active_mm)
86 __tlb_flush_mm(mm);
148} 87}
149 88
150#define flush_tlb_kernel_range(start, end) global_flush_tlb() 89/*
90 * TLB flushing:
91 * flush_tlb() - flushes the current mm struct TLBs
92 * flush_tlb_all() - flushes all processes TLBs
93 * flush_tlb_mm(mm) - flushes the specified mm context TLB's
94 * flush_tlb_page(vma, vmaddr) - flushes one page
95 * flush_tlb_range(vma, start, end) - flushes a range of pages
96 * flush_tlb_kernel_range(start, end) - flushes a range of kernel pages
97 */
151 98
152#endif 99/*
100 * flush_tlb_mm goes together with ptep_set_wrprotect for the
101 * copy_page_range operation and flush_tlb_range is related to
102 * ptep_get_and_clear for change_protection. ptep_set_wrprotect and
103 * ptep_get_and_clear do not flush the TLBs directly if the mm has
104 * only one user. At the end of the update the flush_tlb_mm and
105 * flush_tlb_range functions need to do the flush.
106 */
107#define flush_tlb() do { } while (0)
108#define flush_tlb_all() do { } while (0)
109#define flush_tlb_mm(mm) __tlb_flush_mm_cond(mm)
110#define flush_tlb_page(vma, addr) do { } while (0)
111#define flush_tlb_range(vma, start, end) __tlb_flush_mm_cond(mm)
112#define flush_tlb_kernel_range(start, end) __tlb_flush_mm(&init_mm)
153 113
154#endif /* _S390_TLBFLUSH_H */ 114#endif /* _S390_TLBFLUSH_H */
generated by cgit v1.2.2 (git 2.25.0) at 2024-11-23 06:08:34 -0500