x86: optimize page faults like all other achitectures and kill notifier cruft
[linux-2.6] / include / asm-sh / system.h
1 #ifndef __ASM_SH_SYSTEM_H
2 #define __ASM_SH_SYSTEM_H
3
4 /*
5  * Copyright (C) 1999, 2000  Niibe Yutaka  &  Kaz Kojima
6  * Copyright (C) 2002 Paul Mundt
7  */
8
9 #include <linux/irqflags.h>
10 #include <linux/compiler.h>
11 #include <linux/linkage.h>
12 #include <asm/types.h>
13 #include <asm/ptrace.h>
14
15 struct task_struct *__switch_to(struct task_struct *prev,
16                                 struct task_struct *next);
17
18 /*
19  *      switch_to() should switch tasks to task nr n, first
20  */
21
22 #define switch_to(prev, next, last) do {                                \
23  struct task_struct *__last;                                            \
24  register unsigned long *__ts1 __asm__ ("r1") = &prev->thread.sp;       \
25  register unsigned long *__ts2 __asm__ ("r2") = &prev->thread.pc;       \
26  register unsigned long *__ts4 __asm__ ("r4") = (unsigned long *)prev;  \
27  register unsigned long *__ts5 __asm__ ("r5") = (unsigned long *)next;  \
28  register unsigned long *__ts6 __asm__ ("r6") = &next->thread.sp;       \
29  register unsigned long __ts7 __asm__ ("r7") = next->thread.pc;         \
30  __asm__ __volatile__ (".balign 4\n\t"                                  \
31                        "stc.l   gbr, @-r15\n\t"                         \
32                        "sts.l   pr, @-r15\n\t"                          \
33                        "mov.l   r8, @-r15\n\t"                          \
34                        "mov.l   r9, @-r15\n\t"                          \
35                        "mov.l   r10, @-r15\n\t"                         \
36                        "mov.l   r11, @-r15\n\t"                         \
37                        "mov.l   r12, @-r15\n\t"                         \
38                        "mov.l   r13, @-r15\n\t"                         \
39                        "mov.l   r14, @-r15\n\t"                         \
40                        "mov.l   r15, @r1        ! save SP\n\t"          \
41                        "mov.l   @r6, r15        ! change to new stack\n\t" \
42                        "mova    1f, %0\n\t"                             \
43                        "mov.l   %0, @r2         ! save PC\n\t"          \
44                        "mov.l   2f, %0\n\t"                             \
45                        "jmp     @%0             ! call __switch_to\n\t" \
46                        " lds    r7, pr          !  with return to new PC\n\t" \
47                        ".balign 4\n"                                    \
48                        "2:\n\t"                                         \
49                        ".long   __switch_to\n"                          \
50                        "1:\n\t"                                         \
51                        "mov.l   @r15+, r14\n\t"                         \
52                        "mov.l   @r15+, r13\n\t"                         \
53                        "mov.l   @r15+, r12\n\t"                         \
54                        "mov.l   @r15+, r11\n\t"                         \
55                        "mov.l   @r15+, r10\n\t"                         \
56                        "mov.l   @r15+, r9\n\t"                          \
57                        "mov.l   @r15+, r8\n\t"                          \
58                        "lds.l   @r15+, pr\n\t"                          \
59                        "ldc.l   @r15+, gbr\n\t"                         \
60                        : "=z" (__last)                                  \
61                        : "r" (__ts1), "r" (__ts2), "r" (__ts4),         \
62                          "r" (__ts5), "r" (__ts6), "r" (__ts7)          \
63                        : "r3", "t");                                    \
64         last = __last;                                                  \
65 } while (0)
66
67 #ifdef CONFIG_CPU_SH4A
68 #define __icbi()                        \
69 {                                       \
70         unsigned long __addr;           \
71         __addr = 0xa8000000;            \
72         __asm__ __volatile__(           \
73                 "icbi   %0\n\t"         \
74                 : /* no output */       \
75                 : "m" (__m(__addr)));   \
76 }
77 #endif
78
79 /*
80  * A brief note on ctrl_barrier(), the control register write barrier.
81  *
82  * Legacy SH cores typically require a sequence of 8 nops after
83  * modification of a control register in order for the changes to take
84  * effect. On newer cores (like the sh4a and sh5) this is accomplished
85  * with icbi.
86  *
87  * Also note that on sh4a in the icbi case we can forego a synco for the
88  * write barrier, as it's not necessary for control registers.
89  *
90  * Historically we have only done this type of barrier for the MMUCR, but
91  * it's also necessary for the CCR, so we make it generic here instead.
92  */
93 #ifdef CONFIG_CPU_SH4A
94 #define mb()            __asm__ __volatile__ ("synco": : :"memory")
95 #define rmb()           mb()
96 #define wmb()           __asm__ __volatile__ ("synco": : :"memory")
97 #define ctrl_barrier()  __icbi()
98 #define read_barrier_depends()  do { } while(0)
99 #else
100 #define mb()            __asm__ __volatile__ ("": : :"memory")
101 #define rmb()           mb()
102 #define wmb()           __asm__ __volatile__ ("": : :"memory")
103 #define ctrl_barrier()  __asm__ __volatile__ ("nop;nop;nop;nop;nop;nop;nop;nop")
104 #define read_barrier_depends()  do { } while(0)
105 #endif
106
107 #ifdef CONFIG_SMP
108 #define smp_mb()        mb()
109 #define smp_rmb()       rmb()
110 #define smp_wmb()       wmb()
111 #define smp_read_barrier_depends()      read_barrier_depends()
112 #else
113 #define smp_mb()        barrier()
114 #define smp_rmb()       barrier()
115 #define smp_wmb()       barrier()
116 #define smp_read_barrier_depends()      do { } while(0)
117 #endif
118
119 #define set_mb(var, value) do { (void)xchg(&var, value); } while (0)
120
121 /*
122  * Jump to P2 area.
123  * When handling TLB or caches, we need to do it from P2 area.
124  */
125 #define jump_to_P2()                    \
126 do {                                    \
127         unsigned long __dummy;          \
128         __asm__ __volatile__(           \
129                 "mov.l  1f, %0\n\t"     \
130                 "or     %1, %0\n\t"     \
131                 "jmp    @%0\n\t"        \
132                 " nop\n\t"              \
133                 ".balign 4\n"           \
134                 "1:     .long 2f\n"     \
135                 "2:"                    \
136                 : "=&r" (__dummy)       \
137                 : "r" (0x20000000));    \
138 } while (0)
139
140 /*
141  * Back to P1 area.
142  */
143 #define back_to_P1()                                    \
144 do {                                                    \
145         unsigned long __dummy;                          \
146         ctrl_barrier();                                 \
147         __asm__ __volatile__(                           \
148                 "mov.l  1f, %0\n\t"                     \
149                 "jmp    @%0\n\t"                        \
150                 " nop\n\t"                              \
151                 ".balign 4\n"                           \
152                 "1:     .long 2f\n"                     \
153                 "2:"                                    \
154                 : "=&r" (__dummy));                     \
155 } while (0)
156
157 static inline unsigned long xchg_u32(volatile u32 *m, unsigned long val)
158 {
159         unsigned long flags, retval;
160
161         local_irq_save(flags);
162         retval = *m;
163         *m = val;
164         local_irq_restore(flags);
165         return retval;
166 }
167
168 static inline unsigned long xchg_u8(volatile u8 *m, unsigned long val)
169 {
170         unsigned long flags, retval;
171
172         local_irq_save(flags);
173         retval = *m;
174         *m = val & 0xff;
175         local_irq_restore(flags);
176         return retval;
177 }
178
179 extern void __xchg_called_with_bad_pointer(void);
180
181 #define __xchg(ptr, x, size)                            \
182 ({                                                      \
183         unsigned long __xchg__res;                      \
184         volatile void *__xchg_ptr = (ptr);              \
185         switch (size) {                                 \
186         case 4:                                         \
187                 __xchg__res = xchg_u32(__xchg_ptr, x);  \
188                 break;                                  \
189         case 1:                                         \
190                 __xchg__res = xchg_u8(__xchg_ptr, x);   \
191                 break;                                  \
192         default:                                        \
193                 __xchg_called_with_bad_pointer();       \
194                 __xchg__res = x;                        \
195                 break;                                  \
196         }                                               \
197                                                         \
198         __xchg__res;                                    \
199 })
200
201 #define xchg(ptr,x)     \
202         ((__typeof__(*(ptr)))__xchg((ptr),(unsigned long)(x), sizeof(*(ptr))))
203
204 static inline unsigned long __cmpxchg_u32(volatile int * m, unsigned long old,
205         unsigned long new)
206 {
207         __u32 retval;
208         unsigned long flags;
209
210         local_irq_save(flags);
211         retval = *m;
212         if (retval == old)
213                 *m = new;
214         local_irq_restore(flags);       /* implies memory barrier  */
215         return retval;
216 }
217
218 /* This function doesn't exist, so you'll get a linker error
219  * if something tries to do an invalid cmpxchg(). */
220 extern void __cmpxchg_called_with_bad_pointer(void);
221
222 #define __HAVE_ARCH_CMPXCHG 1
223
224 static inline unsigned long __cmpxchg(volatile void * ptr, unsigned long old,
225                 unsigned long new, int size)
226 {
227         switch (size) {
228         case 4:
229                 return __cmpxchg_u32(ptr, old, new);
230         }
231         __cmpxchg_called_with_bad_pointer();
232         return old;
233 }
234
235 #define cmpxchg(ptr,o,n)                                                 \
236   ({                                                                     \
237      __typeof__(*(ptr)) _o_ = (o);                                       \
238      __typeof__(*(ptr)) _n_ = (n);                                       \
239      (__typeof__(*(ptr))) __cmpxchg((ptr), (unsigned long)_o_,           \
240                                     (unsigned long)_n_, sizeof(*(ptr))); \
241   })
242
243 extern void die(const char *str, struct pt_regs *regs, long err) __attribute__ ((noreturn));
244
245 extern void *set_exception_table_vec(unsigned int vec, void *handler);
246
247 static inline void *set_exception_table_evt(unsigned int evt, void *handler)
248 {
249         return set_exception_table_vec(evt >> 5, handler);
250 }
251
252 /*
253  * SH-2A has both 16 and 32-bit opcodes, do lame encoding checks.
254  */
255 #ifdef CONFIG_CPU_SH2A
256 extern unsigned int instruction_size(unsigned int insn);
257 #else
258 #define instruction_size(insn)  (2)
259 #endif
260
261 /* XXX
262  * disable hlt during certain critical i/o operations
263  */
264 #define HAVE_DISABLE_HLT
265 void disable_hlt(void);
266 void enable_hlt(void);
267
268 void default_idle(void);
269 void per_cpu_trap_init(void);
270
271 asmlinkage void break_point_trap(void);
272 asmlinkage void debug_trap_handler(unsigned long r4, unsigned long r5,
273                                    unsigned long r6, unsigned long r7,
274                                    struct pt_regs __regs);
275 asmlinkage void bug_trap_handler(unsigned long r4, unsigned long r5,
276                                  unsigned long r6, unsigned long r7,
277                                  struct pt_regs __regs);
278
279 #define arch_align_stack(x) (x)
280
281 #endif