Merge branch 'stable/bug.fixes' of git://git.kernel.org/pub/scm/linux/kernel/git...
[pandora-kernel.git] / arch / sparc / include / asm / system_64.h
1 #ifndef __SPARC64_SYSTEM_H
2 #define __SPARC64_SYSTEM_H
3
4 #include <asm/ptrace.h>
5 #include <asm/processor.h>
6 #include <asm/visasm.h>
7
8 #ifndef __ASSEMBLY__
9
10 #include <linux/irqflags.h>
11 #include <asm-generic/cmpxchg-local.h>
12
13 /*
14  * Sparc (general) CPU types
15  */
16 enum sparc_cpu {
17   sun4        = 0x00,
18   sun4c       = 0x01,
19   sun4m       = 0x02,
20   sun4d       = 0x03,
21   sun4e       = 0x04,
22   sun4u       = 0x05, /* V8 ploos ploos */
23   sun_unknown = 0x06,
24   ap1000      = 0x07, /* almost a sun4m */
25 };
26
27 #define sparc_cpu_model sun4u
28
29 /* This cannot ever be a sun4c :) That's just history. */
30 #define ARCH_SUN4C 0
31
32 extern char reboot_command[];
33
34 /* These are here in an effort to more fully work around Spitfire Errata
35  * #51.  Essentially, if a memory barrier occurs soon after a mispredicted
36  * branch, the chip can stop executing instructions until a trap occurs.
37  * Therefore, if interrupts are disabled, the chip can hang forever.
38  *
39  * It used to be believed that the memory barrier had to be right in the
40  * delay slot, but a case has been traced recently wherein the memory barrier
41  * was one instruction after the branch delay slot and the chip still hung.
42  * The offending sequence was the following in sym_wakeup_done() of the
43  * sym53c8xx_2 driver:
44  *
45  *      call    sym_ccb_from_dsa, 0
46  *       movge  %icc, 0, %l0
47  *      brz,pn  %o0, .LL1303
48  *       mov    %o0, %l2
49  *      membar  #LoadLoad
50  *
51  * The branch has to be mispredicted for the bug to occur.  Therefore, we put
52  * the memory barrier explicitly into a "branch always, predicted taken"
53  * delay slot to avoid the problem case.
54  */
55 #define membar_safe(type) \
56 do {    __asm__ __volatile__("ba,pt     %%xcc, 1f\n\t" \
57                              " membar   " type "\n" \
58                              "1:\n" \
59                              : : : "memory"); \
60 } while (0)
61
62 /* The kernel always executes in TSO memory model these days,
63  * and furthermore most sparc64 chips implement more stringent
64  * memory ordering than required by the specifications.
65  */
66 #define mb()    membar_safe("#StoreLoad")
67 #define rmb()   __asm__ __volatile__("":::"memory")
68 #define wmb()   __asm__ __volatile__("":::"memory")
69
70 #endif
71
72 #define nop()           __asm__ __volatile__ ("nop")
73
74 #define read_barrier_depends()          do { } while(0)
75 #define set_mb(__var, __value) \
76         do { __var = __value; membar_safe("#StoreLoad"); } while(0)
77
78 #ifdef CONFIG_SMP
79 #define smp_mb()        mb()
80 #define smp_rmb()       rmb()
81 #define smp_wmb()       wmb()
82 #else
83 #define smp_mb()        __asm__ __volatile__("":::"memory")
84 #define smp_rmb()       __asm__ __volatile__("":::"memory")
85 #define smp_wmb()       __asm__ __volatile__("":::"memory")
86 #endif
87
88 #define smp_read_barrier_depends()      do { } while(0)
89
90 #define flushi(addr)    __asm__ __volatile__ ("flush %0" : : "r" (addr) : "memory")
91
92 #define flushw_all()    __asm__ __volatile__("flushw")
93
94 /* Performance counter register access. */
95 #define read_pcr(__p)  __asm__ __volatile__("rd %%pcr, %0" : "=r" (__p))
96 #define write_pcr(__p) __asm__ __volatile__("wr %0, 0x0, %%pcr" : : "r" (__p))
97 #define read_pic(__p)  __asm__ __volatile__("rd %%pic, %0" : "=r" (__p))
98
99 /* Blackbird errata workaround.  See commentary in
100  * arch/sparc64/kernel/smp.c:smp_percpu_timer_interrupt()
101  * for more information.
102  */
103 #define write_pic(__p)                                          \
104         __asm__ __volatile__("ba,pt     %%xcc, 99f\n\t"         \
105                              " nop\n\t"                         \
106                              ".align    64\n"                   \
107                           "99:wr        %0, 0x0, %%pic\n\t"     \
108                              "rd        %%pic, %%g0" : : "r" (__p))
109 #define reset_pic()     write_pic(0)
110
111 #ifndef __ASSEMBLY__
112
113 extern void sun_do_break(void);
114 extern int stop_a_enabled;
115 extern int scons_pwroff;
116
117 extern void fault_in_user_windows(void);
118 extern void synchronize_user_stack(void);
119
120 extern void __flushw_user(void);
121 #define flushw_user() __flushw_user()
122
123 #define flush_user_windows flushw_user
124 #define flush_register_windows flushw_all
125
126 /* Don't hold the runqueue lock over context switch */
127 #define __ARCH_WANT_UNLOCKED_CTXSW
128 #define prepare_arch_switch(next)               \
129 do {                                            \
130         flushw_all();                           \
131 } while (0)
132
133         /* See what happens when you design the chip correctly?
134          *
135          * We tell gcc we clobber all non-fixed-usage registers except
136          * for l0/l1.  It will use one for 'next' and the other to hold
137          * the output value of 'last'.  'next' is not referenced again
138          * past the invocation of switch_to in the scheduler, so we need
139          * not preserve it's value.  Hairy, but it lets us remove 2 loads
140          * and 2 stores in this critical code path.  -DaveM
141          */
142 #define switch_to(prev, next, last)                                     \
143 do {    flush_tlb_pending();                                            \
144         save_and_clear_fpu();                                           \
145         /* If you are tempted to conditionalize the following */        \
146         /* so that ASI is only written if it changes, think again. */   \
147         __asm__ __volatile__("wr %%g0, %0, %%asi"                       \
148         : : "r" (__thread_flag_byte_ptr(task_thread_info(next))[TI_FLAG_BYTE_CURRENT_DS]));\
149         trap_block[current_thread_info()->cpu].thread =                 \
150                 task_thread_info(next);                                 \
151         __asm__ __volatile__(                                           \
152         "mov    %%g4, %%g7\n\t"                                         \
153         "stx    %%i6, [%%sp + 2047 + 0x70]\n\t"                         \
154         "stx    %%i7, [%%sp + 2047 + 0x78]\n\t"                         \
155         "rdpr   %%wstate, %%o5\n\t"                                     \
156         "stx    %%o6, [%%g6 + %6]\n\t"                                  \
157         "stb    %%o5, [%%g6 + %5]\n\t"                                  \
158         "rdpr   %%cwp, %%o5\n\t"                                        \
159         "stb    %%o5, [%%g6 + %8]\n\t"                                  \
160         "wrpr   %%g0, 15, %%pil\n\t"                                    \
161         "mov    %4, %%g6\n\t"                                           \
162         "ldub   [%4 + %8], %%g1\n\t"                                    \
163         "wrpr   %%g1, %%cwp\n\t"                                        \
164         "ldx    [%%g6 + %6], %%o6\n\t"                                  \
165         "ldub   [%%g6 + %5], %%o5\n\t"                                  \
166         "ldub   [%%g6 + %7], %%o7\n\t"                                  \
167         "wrpr   %%o5, 0x0, %%wstate\n\t"                                \
168         "ldx    [%%sp + 2047 + 0x70], %%i6\n\t"                         \
169         "ldx    [%%sp + 2047 + 0x78], %%i7\n\t"                         \
170         "ldx    [%%g6 + %9], %%g4\n\t"                                  \
171         "wrpr   %%g0, 14, %%pil\n\t"                                    \
172         "brz,pt %%o7, switch_to_pc\n\t"                                 \
173         " mov   %%g7, %0\n\t"                                           \
174         "sethi  %%hi(ret_from_syscall), %%g1\n\t"                       \
175         "jmpl   %%g1 + %%lo(ret_from_syscall), %%g0\n\t"                \
176         " nop\n\t"                                                      \
177         ".globl switch_to_pc\n\t"                                       \
178         "switch_to_pc:\n\t"                                             \
179         : "=&r" (last), "=r" (current), "=r" (current_thread_info_reg), \
180           "=r" (__local_per_cpu_offset)                                 \
181         : "0" (task_thread_info(next)),                                 \
182           "i" (TI_WSTATE), "i" (TI_KSP), "i" (TI_NEW_CHILD),            \
183           "i" (TI_CWP), "i" (TI_TASK)                                   \
184         : "cc",                                                         \
185                 "g1", "g2", "g3",                   "g7",               \
186                 "l1", "l2", "l3", "l4", "l5", "l6", "l7",               \
187           "i0", "i1", "i2", "i3", "i4", "i5",                           \
188           "o0", "o1", "o2", "o3", "o4", "o5",       "o7");              \
189 } while(0)
190
191 static inline unsigned long xchg32(__volatile__ unsigned int *m, unsigned int val)
192 {
193         unsigned long tmp1, tmp2;
194
195         __asm__ __volatile__(
196 "       mov             %0, %1\n"
197 "1:     lduw            [%4], %2\n"
198 "       cas             [%4], %2, %0\n"
199 "       cmp             %2, %0\n"
200 "       bne,a,pn        %%icc, 1b\n"
201 "        mov            %1, %0\n"
202         : "=&r" (val), "=&r" (tmp1), "=&r" (tmp2)
203         : "0" (val), "r" (m)
204         : "cc", "memory");
205         return val;
206 }
207
208 static inline unsigned long xchg64(__volatile__ unsigned long *m, unsigned long val)
209 {
210         unsigned long tmp1, tmp2;
211
212         __asm__ __volatile__(
213 "       mov             %0, %1\n"
214 "1:     ldx             [%4], %2\n"
215 "       casx            [%4], %2, %0\n"
216 "       cmp             %2, %0\n"
217 "       bne,a,pn        %%xcc, 1b\n"
218 "        mov            %1, %0\n"
219         : "=&r" (val), "=&r" (tmp1), "=&r" (tmp2)
220         : "0" (val), "r" (m)
221         : "cc", "memory");
222         return val;
223 }
224
225 #define xchg(ptr,x) ((__typeof__(*(ptr)))__xchg((unsigned long)(x),(ptr),sizeof(*(ptr))))
226
227 extern void __xchg_called_with_bad_pointer(void);
228
229 static inline unsigned long __xchg(unsigned long x, __volatile__ void * ptr,
230                                        int size)
231 {
232         switch (size) {
233         case 4:
234                 return xchg32(ptr, x);
235         case 8:
236                 return xchg64(ptr, x);
237         }
238         __xchg_called_with_bad_pointer();
239         return x;
240 }
241
242 extern void die_if_kernel(char *str, struct pt_regs *regs) __attribute__ ((noreturn));
243
244 /*
245  * Atomic compare and exchange.  Compare OLD with MEM, if identical,
246  * store NEW in MEM.  Return the initial value in MEM.  Success is
247  * indicated by comparing RETURN with OLD.
248  */
249
250 #define __HAVE_ARCH_CMPXCHG 1
251
252 static inline unsigned long
253 __cmpxchg_u32(volatile int *m, int old, int new)
254 {
255         __asm__ __volatile__("cas [%2], %3, %0"
256                              : "=&r" (new)
257                              : "0" (new), "r" (m), "r" (old)
258                              : "memory");
259
260         return new;
261 }
262
263 static inline unsigned long
264 __cmpxchg_u64(volatile long *m, unsigned long old, unsigned long new)
265 {
266         __asm__ __volatile__("casx [%2], %3, %0"
267                              : "=&r" (new)
268                              : "0" (new), "r" (m), "r" (old)
269                              : "memory");
270
271         return new;
272 }
273
274 /* This function doesn't exist, so you'll get a linker error
275    if something tries to do an invalid cmpxchg().  */
276 extern void __cmpxchg_called_with_bad_pointer(void);
277
278 static inline unsigned long
279 __cmpxchg(volatile void *ptr, unsigned long old, unsigned long new, int size)
280 {
281         switch (size) {
282                 case 4:
283                         return __cmpxchg_u32(ptr, old, new);
284                 case 8:
285                         return __cmpxchg_u64(ptr, old, new);
286         }
287         __cmpxchg_called_with_bad_pointer();
288         return old;
289 }
290
291 #define cmpxchg(ptr,o,n)                                                 \
292   ({                                                                     \
293      __typeof__(*(ptr)) _o_ = (o);                                       \
294      __typeof__(*(ptr)) _n_ = (n);                                       \
295      (__typeof__(*(ptr))) __cmpxchg((ptr), (unsigned long)_o_,           \
296                                     (unsigned long)_n_, sizeof(*(ptr))); \
297   })
298
299 /*
300  * cmpxchg_local and cmpxchg64_local are atomic wrt current CPU. Always make
301  * them available.
302  */
303
304 static inline unsigned long __cmpxchg_local(volatile void *ptr,
305                                       unsigned long old,
306                                       unsigned long new, int size)
307 {
308         switch (size) {
309         case 4:
310         case 8: return __cmpxchg(ptr, old, new, size);
311         default:
312                 return __cmpxchg_local_generic(ptr, old, new, size);
313         }
314
315         return old;
316 }
317
318 #define cmpxchg_local(ptr, o, n)                                        \
319         ((__typeof__(*(ptr)))__cmpxchg_local((ptr), (unsigned long)(o), \
320                         (unsigned long)(n), sizeof(*(ptr))))
321 #define cmpxchg64_local(ptr, o, n)                                      \
322   ({                                                                    \
323         BUILD_BUG_ON(sizeof(*(ptr)) != 8);                              \
324         cmpxchg_local((ptr), (o), (n));                                 \
325   })
326
327 #endif /* !(__ASSEMBLY__) */
328
329 #define arch_align_stack(x) (x)
330
331 #endif /* !(__SPARC64_SYSTEM_H) */