1 #ifndef _ASM_M32R_SYSTEM_H
2 #define _ASM_M32R_SYSTEM_H
3 
4 /*
5  * This file is subject to the terms and conditions of the GNU General Public
6  * License.  See the file "COPYING" in the main directory of this archive
7  * for more details.
8  *
9  * Copyright (C) 2001  Hiroyuki Kondo, Hirokazu Takata, and Hitoshi Yamamoto
10  * Copyright (C) 2004, 2006  Hirokazu Takata <takata at linux-m32r.org>
11  */
12 
13 #include <linux/compiler.h>
14 #include <linux/irqflags.h>
15 #include <asm/assembler.h>
16 
17 #ifdef __KERNEL__
18 
19 /*
20  * switch_to(prev, next) should switch from task `prev' to `next'
21  * `prev' will never be the same as `next'.
22  *
23  * `next' and `prev' should be struct task_struct, but it isn't always defined
24  */
25 
26 #if defined(CONFIG_FRAME_POINTER) || \
27 	!defined(CONFIG_SCHED_OMIT_FRAME_POINTER)
28 #define M32R_PUSH_FP "	push fp\n"
29 #define M32R_POP_FP  "	pop  fp\n"
30 #else
31 #define M32R_PUSH_FP ""
32 #define M32R_POP_FP  ""
33 #endif
34 
35 #define switch_to(prev, next, last)  do { \
36 	__asm__ __volatile__ ( \
37 		"	seth	lr, #high(1f)				\n" \
38 		"	or3	lr, lr, #low(1f)			\n" \
39 		"	st	lr, @%4  ; store old LR			\n" \
40 		"	ld	lr, @%5  ; load new LR			\n" \
41 			M32R_PUSH_FP \
42 		"	st	sp, @%2  ; store old SP			\n" \
43 		"	ld	sp, @%3  ; load new SP			\n" \
44 		"	push	%1  ; store `prev' on new stack		\n" \
45 		"	jmp	lr					\n" \
46 		"	.fillinsn					\n" \
47 		"1:							\n" \
48 		"	pop	%0  ; restore `__last' from new stack	\n" \
49 			M32R_POP_FP \
50 		: "=r" (last) \
51 		: "0" (prev), \
52 		  "r" (&(prev->thread.sp)), "r" (&(next->thread.sp)), \
53 		  "r" (&(prev->thread.lr)), "r" (&(next->thread.lr)) \
54 		: "memory", "lr" \
55 	); \
56 } while(0)
57 
58 #define nop()	__asm__ __volatile__ ("nop" : : )
59 
60 #define xchg(ptr, x)							\
61 	((__typeof__(*(ptr)))__xchg((unsigned long)(x), (ptr), sizeof(*(ptr))))
62 #define xchg_local(ptr, x)						\
63 	((__typeof__(*(ptr)))__xchg_local((unsigned long)(x), (ptr),	\
64 			sizeof(*(ptr))))
65 
66 extern void  __xchg_called_with_bad_pointer(void);
67 
68 #ifdef CONFIG_CHIP_M32700_TS1
69 #define DCACHE_CLEAR(reg0, reg1, addr)				\
70 	"seth	"reg1", #high(dcache_dummy);		\n\t"	\
71 	"or3	"reg1", "reg1", #low(dcache_dummy);	\n\t"	\
72 	"lock	"reg0", @"reg1";			\n\t"	\
73 	"add3	"reg0", "addr", #0x1000;		\n\t"	\
74 	"ld	"reg0", @"reg0";			\n\t"	\
75 	"add3	"reg0", "addr", #0x2000;		\n\t"	\
76 	"ld	"reg0", @"reg0";			\n\t"	\
77 	"unlock	"reg0", @"reg1";			\n\t"
78 	/* FIXME: This workaround code cannot handle kernel modules
79 	 * correctly under SMP environment.
80 	 */
81 #else	/* CONFIG_CHIP_M32700_TS1 */
82 #define DCACHE_CLEAR(reg0, reg1, addr)
83 #endif	/* CONFIG_CHIP_M32700_TS1 */
84 
85 static __always_inline unsigned long
__xchg(unsigned long x,volatile void * ptr,int size)86 __xchg(unsigned long x, volatile void *ptr, int size)
87 {
88 	unsigned long flags;
89 	unsigned long tmp = 0;
90 
91 	local_irq_save(flags);
92 
93 	switch (size) {
94 #ifndef CONFIG_SMP
95 	case 1:
96 		__asm__ __volatile__ (
97 			"ldb	%0, @%2 \n\t"
98 			"stb	%1, @%2 \n\t"
99 			: "=&r" (tmp) : "r" (x), "r" (ptr) : "memory");
100 		break;
101 	case 2:
102 		__asm__ __volatile__ (
103 			"ldh	%0, @%2 \n\t"
104 			"sth	%1, @%2 \n\t"
105 			: "=&r" (tmp) : "r" (x), "r" (ptr) : "memory");
106 		break;
107 	case 4:
108 		__asm__ __volatile__ (
109 			"ld	%0, @%2 \n\t"
110 			"st	%1, @%2 \n\t"
111 			: "=&r" (tmp) : "r" (x), "r" (ptr) : "memory");
112 		break;
113 #else  /* CONFIG_SMP */
114 	case 4:
115 		__asm__ __volatile__ (
116 			DCACHE_CLEAR("%0", "r4", "%2")
117 			"lock	%0, @%2;	\n\t"
118 			"unlock	%1, @%2;	\n\t"
119 			: "=&r" (tmp) : "r" (x), "r" (ptr)
120 			: "memory"
121 #ifdef CONFIG_CHIP_M32700_TS1
122 			, "r4"
123 #endif	/* CONFIG_CHIP_M32700_TS1 */
124 		);
125 		break;
126 #endif  /* CONFIG_SMP */
127 	default:
128 		__xchg_called_with_bad_pointer();
129 	}
130 
131 	local_irq_restore(flags);
132 
133 	return (tmp);
134 }
135 
136 static __always_inline unsigned long
__xchg_local(unsigned long x,volatile void * ptr,int size)137 __xchg_local(unsigned long x, volatile void *ptr, int size)
138 {
139 	unsigned long flags;
140 	unsigned long tmp = 0;
141 
142 	local_irq_save(flags);
143 
144 	switch (size) {
145 	case 1:
146 		__asm__ __volatile__ (
147 			"ldb	%0, @%2 \n\t"
148 			"stb	%1, @%2 \n\t"
149 			: "=&r" (tmp) : "r" (x), "r" (ptr) : "memory");
150 		break;
151 	case 2:
152 		__asm__ __volatile__ (
153 			"ldh	%0, @%2 \n\t"
154 			"sth	%1, @%2 \n\t"
155 			: "=&r" (tmp) : "r" (x), "r" (ptr) : "memory");
156 		break;
157 	case 4:
158 		__asm__ __volatile__ (
159 			"ld	%0, @%2 \n\t"
160 			"st	%1, @%2 \n\t"
161 			: "=&r" (tmp) : "r" (x), "r" (ptr) : "memory");
162 		break;
163 	default:
164 		__xchg_called_with_bad_pointer();
165 	}
166 
167 	local_irq_restore(flags);
168 
169 	return (tmp);
170 }
171 
172 #define __HAVE_ARCH_CMPXCHG	1
173 
174 static inline unsigned long
__cmpxchg_u32(volatile unsigned int * p,unsigned int old,unsigned int new)175 __cmpxchg_u32(volatile unsigned int *p, unsigned int old, unsigned int new)
176 {
177 	unsigned long flags;
178 	unsigned int retval;
179 
180 	local_irq_save(flags);
181 	__asm__ __volatile__ (
182 			DCACHE_CLEAR("%0", "r4", "%1")
183 			M32R_LOCK" %0, @%1;	\n"
184 		"	bne	%0, %2, 1f;	\n"
185 			M32R_UNLOCK" %3, @%1;	\n"
186 		"	bra	2f;		\n"
187                 "       .fillinsn		\n"
188 		"1:"
189 			M32R_UNLOCK" %0, @%1;	\n"
190                 "       .fillinsn		\n"
191 		"2:"
192 			: "=&r" (retval)
193 			: "r" (p), "r" (old), "r" (new)
194 			: "cbit", "memory"
195 #ifdef CONFIG_CHIP_M32700_TS1
196 			, "r4"
197 #endif  /* CONFIG_CHIP_M32700_TS1 */
198 		);
199 	local_irq_restore(flags);
200 
201 	return retval;
202 }
203 
204 static inline unsigned long
__cmpxchg_local_u32(volatile unsigned int * p,unsigned int old,unsigned int new)205 __cmpxchg_local_u32(volatile unsigned int *p, unsigned int old,
206 			unsigned int new)
207 {
208 	unsigned long flags;
209 	unsigned int retval;
210 
211 	local_irq_save(flags);
212 	__asm__ __volatile__ (
213 			DCACHE_CLEAR("%0", "r4", "%1")
214 			"ld %0, @%1;		\n"
215 		"	bne	%0, %2, 1f;	\n"
216 			"st %3, @%1;		\n"
217 		"	bra	2f;		\n"
218 		"       .fillinsn		\n"
219 		"1:"
220 			"st %0, @%1;		\n"
221 		"       .fillinsn		\n"
222 		"2:"
223 			: "=&r" (retval)
224 			: "r" (p), "r" (old), "r" (new)
225 			: "cbit", "memory"
226 #ifdef CONFIG_CHIP_M32700_TS1
227 			, "r4"
228 #endif  /* CONFIG_CHIP_M32700_TS1 */
229 		);
230 	local_irq_restore(flags);
231 
232 	return retval;
233 }
234 
235 /* This function doesn't exist, so you'll get a linker error
236    if something tries to do an invalid cmpxchg().  */
237 extern void __cmpxchg_called_with_bad_pointer(void);
238 
239 static inline unsigned long
__cmpxchg(volatile void * ptr,unsigned long old,unsigned long new,int size)240 __cmpxchg(volatile void *ptr, unsigned long old, unsigned long new, int size)
241 {
242 	switch (size) {
243 	case 4:
244 		return __cmpxchg_u32(ptr, old, new);
245 #if 0	/* we don't have __cmpxchg_u64 */
246 	case 8:
247 		return __cmpxchg_u64(ptr, old, new);
248 #endif /* 0 */
249 	}
250 	__cmpxchg_called_with_bad_pointer();
251 	return old;
252 }
253 
254 #define cmpxchg(ptr, o, n)						 \
255 	((__typeof__(*(ptr))) __cmpxchg((ptr), (unsigned long)(o),	 \
256 			(unsigned long)(n), sizeof(*(ptr))))
257 
258 #include <asm-generic/cmpxchg-local.h>
259 
__cmpxchg_local(volatile void * ptr,unsigned long old,unsigned long new,int size)260 static inline unsigned long __cmpxchg_local(volatile void *ptr,
261 				      unsigned long old,
262 				      unsigned long new, int size)
263 {
264 	switch (size) {
265 	case 4:
266 		return __cmpxchg_local_u32(ptr, old, new);
267 	default:
268 		return __cmpxchg_local_generic(ptr, old, new, size);
269 	}
270 
271 	return old;
272 }
273 
274 /*
275  * cmpxchg_local and cmpxchg64_local are atomic wrt current CPU. Always make
276  * them available.
277  */
278 #define cmpxchg_local(ptr, o, n)				  	    \
279 	((__typeof__(*(ptr)))__cmpxchg_local((ptr), (unsigned long)(o),	    \
280 			(unsigned long)(n), sizeof(*(ptr))))
281 #define cmpxchg64_local(ptr, o, n) __cmpxchg64_local_generic((ptr), (o), (n))
282 
283 #endif  /* __KERNEL__ */
284 
285 /*
286  * Memory barrier.
287  *
288  * mb() prevents loads and stores being reordered across this point.
289  * rmb() prevents loads being reordered across this point.
290  * wmb() prevents stores being reordered across this point.
291  */
292 #define mb()   barrier()
293 #define rmb()  mb()
294 #define wmb()  mb()
295 
296 /**
297  * read_barrier_depends - Flush all pending reads that subsequents reads
298  * depend on.
299  *
300  * No data-dependent reads from memory-like regions are ever reordered
301  * over this barrier.  All reads preceding this primitive are guaranteed
302  * to access memory (but not necessarily other CPUs' caches) before any
303  * reads following this primitive that depend on the data return by
304  * any of the preceding reads.  This primitive is much lighter weight than
305  * rmb() on most CPUs, and is never heavier weight than is
306  * rmb().
307  *
308  * These ordering constraints are respected by both the local CPU
309  * and the compiler.
310  *
311  * Ordering is not guaranteed by anything other than these primitives,
312  * not even by data dependencies.  See the documentation for
313  * memory_barrier() for examples and URLs to more information.
314  *
315  * For example, the following code would force ordering (the initial
316  * value of "a" is zero, "b" is one, and "p" is "&a"):
317  *
318  * <programlisting>
319  *      CPU 0                           CPU 1
320  *
321  *      b = 2;
322  *      memory_barrier();
323  *      p = &b;                         q = p;
324  *                                      read_barrier_depends();
325  *                                      d = *q;
326  * </programlisting>
327  *
328  *
329  * because the read of "*q" depends on the read of "p" and these
330  * two reads are separated by a read_barrier_depends().  However,
331  * the following code, with the same initial values for "a" and "b":
332  *
333  * <programlisting>
334  *      CPU 0                           CPU 1
335  *
336  *      a = 2;
337  *      memory_barrier();
338  *      b = 3;                          y = b;
339  *                                      read_barrier_depends();
340  *                                      x = a;
341  * </programlisting>
342  *
343  * does not enforce ordering, since there is no data dependency between
344  * the read of "a" and the read of "b".  Therefore, on some CPUs, such
345  * as Alpha, "y" could be set to 3 and "x" to 0.  Use rmb()
346  * in cases like this where there are no data dependencies.
347  **/
348 
349 #define read_barrier_depends()	do { } while (0)
350 
351 #ifdef CONFIG_SMP
352 #define smp_mb()	mb()
353 #define smp_rmb()	rmb()
354 #define smp_wmb()	wmb()
355 #define smp_read_barrier_depends()	read_barrier_depends()
356 #define set_mb(var, value) do { (void) xchg(&var, value); } while (0)
357 #else
358 #define smp_mb()	barrier()
359 #define smp_rmb()	barrier()
360 #define smp_wmb()	barrier()
361 #define smp_read_barrier_depends()	do { } while (0)
362 #define set_mb(var, value) do { var = value; barrier(); } while (0)
363 #endif
364 
365 #define arch_align_stack(x) (x)
366 
367 #endif /* _ASM_M32R_SYSTEM_H */
368