1/*
2 * This file is subject to the terms and conditions of the GNU General Public
3 * License.  See the file "COPYING" in the main directory of this archive
4 * for more details.
5 *
6 * Inline assembly cache operations.
7 *
8 * Copyright (C) 1996 David S. Miller (davem@davemloft.net)
9 * Copyright (C) 1997 - 2002 Ralf Baechle (ralf@gnu.org)
10 * Copyright (C) 2004 Ralf Baechle (ralf@linux-mips.org)
11 */
12#ifndef _ASM_R4KCACHE_H
13#define _ASM_R4KCACHE_H
14
15#include <asm/asm.h>
16#include <asm/cacheops.h>
17#include <asm/cpu-features.h>
18#include <asm/mipsmtregs.h>
19
20/*
21 * This macro return a properly sign-extended address suitable as base address
22 * for indexed cache operations.  Two issues here:
23 *
24 *  - The MIPS32 and MIPS64 specs permit an implementation to directly derive
25 *    the index bits from the virtual address.  This breaks with tradition
26 *    set by the R4000.  To keep unpleasant surprises from happening we pick
27 *    an address in KSEG0 / CKSEG0.
28 *  - We need a properly sign extended address for 64-bit code.  To get away
29 *    without ifdefs we let the compiler do it by a type cast.
30 */
31#define INDEX_BASE	CKSEG0
32
33#define cache_op(op,addr)						\
34	__asm__ __volatile__(						\
35	"	.set	push					\n"	\
36	"	.set	noreorder				\n"	\
37	"	.set	mips3\n\t				\n"	\
38	"	cache	%0, %1					\n"	\
39	"	.set	pop					\n"	\
40	:								\
41	: "i" (op), "R" (*(unsigned char *)(addr)))
42
43#ifdef CONFIG_MIPS_MT
44/*
45 * Temporary hacks for SMTC debug. Optionally force single-threaded
46 * execution during I-cache flushes.
47 */
48
49#define PROTECT_CACHE_FLUSHES 1
50
51#ifdef PROTECT_CACHE_FLUSHES
52
53extern int mt_protiflush;
54extern int mt_protdflush;
55extern void mt_cflush_lockdown(void);
56extern void mt_cflush_release(void);
57
58#define BEGIN_MT_IPROT \
59	unsigned long flags = 0;			\
60	unsigned long mtflags = 0;			\
61	if(mt_protiflush) {				\
62		local_irq_save(flags);			\
63		ehb();					\
64		mtflags = dvpe();			\
65		mt_cflush_lockdown();			\
66	}
67
68#define END_MT_IPROT \
69	if(mt_protiflush) {				\
70		mt_cflush_release();			\
71		evpe(mtflags);				\
72		local_irq_restore(flags);		\
73	}
74
75#define BEGIN_MT_DPROT \
76	unsigned long flags = 0;			\
77	unsigned long mtflags = 0;			\
78	if(mt_protdflush) {				\
79		local_irq_save(flags);			\
80		ehb();					\
81		mtflags = dvpe();			\
82		mt_cflush_lockdown();			\
83	}
84
85#define END_MT_DPROT \
86	if(mt_protdflush) {				\
87		mt_cflush_release();			\
88		evpe(mtflags);				\
89		local_irq_restore(flags);		\
90	}
91
92#else
93
94#define BEGIN_MT_IPROT
95#define BEGIN_MT_DPROT
96#define END_MT_IPROT
97#define END_MT_DPROT
98
99#endif /* PROTECT_CACHE_FLUSHES */
100
101#define __iflush_prologue						\
102	unsigned long redundance;					\
103	extern int mt_n_iflushes;					\
104	BEGIN_MT_IPROT							\
105	for (redundance = 0; redundance < mt_n_iflushes; redundance++) {
106
107#define __iflush_epilogue						\
108	END_MT_IPROT							\
109	}
110
111#define __dflush_prologue						\
112	unsigned long redundance;					\
113	extern int mt_n_dflushes;					\
114	BEGIN_MT_DPROT							\
115	for (redundance = 0; redundance < mt_n_dflushes; redundance++) {
116
117#define __dflush_epilogue \
118	END_MT_DPROT	 \
119	}
120
121#define __inv_dflush_prologue __dflush_prologue
122#define __inv_dflush_epilogue __dflush_epilogue
123#define __sflush_prologue {
124#define __sflush_epilogue }
125#define __inv_sflush_prologue __sflush_prologue
126#define __inv_sflush_epilogue __sflush_epilogue
127
128#else /* CONFIG_MIPS_MT */
129
130#define __iflush_prologue {
131#define __iflush_epilogue }
132#define __dflush_prologue {
133#define __dflush_epilogue }
134#define __inv_dflush_prologue {
135#define __inv_dflush_epilogue }
136#define __sflush_prologue {
137#define __sflush_epilogue }
138#define __inv_sflush_prologue {
139#define __inv_sflush_epilogue }
140
141#endif /* CONFIG_MIPS_MT */
142
143static inline void flush_icache_line_indexed(unsigned long addr)
144{
145	__iflush_prologue
146	cache_op(Index_Invalidate_I, addr);
147	__iflush_epilogue
148}
149
150static inline void flush_dcache_line_indexed(unsigned long addr)
151{
152	__dflush_prologue
153	cache_op(Index_Writeback_Inv_D, addr);
154	__dflush_epilogue
155}
156
157static inline void flush_scache_line_indexed(unsigned long addr)
158{
159	cache_op(Index_Writeback_Inv_SD, addr);
160}
161
162static inline void flush_icache_line(unsigned long addr)
163{
164	__iflush_prologue
165	cache_op(Hit_Invalidate_I, addr);
166	__iflush_epilogue
167}
168
169static inline void flush_dcache_line(unsigned long addr)
170{
171	__dflush_prologue
172	cache_op(Hit_Writeback_Inv_D, addr);
173	__dflush_epilogue
174}
175
176static inline void invalidate_dcache_line(unsigned long addr)
177{
178	__dflush_prologue
179	cache_op(Hit_Invalidate_D, addr);
180	__dflush_epilogue
181}
182
183static inline void invalidate_scache_line(unsigned long addr)
184{
185	cache_op(Hit_Invalidate_SD, addr);
186}
187
188static inline void flush_scache_line(unsigned long addr)
189{
190	cache_op(Hit_Writeback_Inv_SD, addr);
191}
192
193#define protected_cache_op(op,addr)				\
194	__asm__ __volatile__(					\
195	"	.set	push			\n"		\
196	"	.set	noreorder		\n"		\
197	"	.set	mips3			\n"		\
198	"1:	cache	%0, (%1)		\n"		\
199	"2:	.set	pop			\n"		\
200	"	.section __ex_table,\"a\"	\n"		\
201	"	"STR(PTR)" 1b, 2b		\n"		\
202	"	.previous"					\
203	:							\
204	: "i" (op), "r" (addr))
205
206/*
207 * The next two are for badland addresses like signal trampolines.
208 */
209static inline void protected_flush_icache_line(unsigned long addr)
210{
211	protected_cache_op(Hit_Invalidate_I, addr);
212}
213
214/*
215 * R10000 / R12000 hazard - these processors don't support the Hit_Writeback_D
216 * cacheop so we use Hit_Writeback_Inv_D which is supported by all R4000-style
217 * caches.  We're talking about one cacheline unnecessarily getting invalidated
218 * here so the penalty isn't overly hard.
219 */
220static inline void protected_writeback_dcache_line(unsigned long addr)
221{
222	protected_cache_op(Hit_Writeback_Inv_D, addr);
223}
224
225static inline void protected_writeback_scache_line(unsigned long addr)
226{
227	protected_cache_op(Hit_Writeback_Inv_SD, addr);
228}
229
230/*
231 * This one is RM7000-specific
232 */
233static inline void invalidate_tcache_page(unsigned long addr)
234{
235	cache_op(Page_Invalidate_T, addr);
236}
237
238#define cache16_unroll32(base,op)					\
239	__asm__ __volatile__(						\
240	"	.set push					\n"	\
241	"	.set noreorder					\n"	\
242	"	.set mips3					\n"	\
243	"	cache %1, 0x000(%0); cache %1, 0x010(%0)	\n"	\
244	"	cache %1, 0x020(%0); cache %1, 0x030(%0)	\n"	\
245	"	cache %1, 0x040(%0); cache %1, 0x050(%0)	\n"	\
246	"	cache %1, 0x060(%0); cache %1, 0x070(%0)	\n"	\
247	"	cache %1, 0x080(%0); cache %1, 0x090(%0)	\n"	\
248	"	cache %1, 0x0a0(%0); cache %1, 0x0b0(%0)	\n"	\
249	"	cache %1, 0x0c0(%0); cache %1, 0x0d0(%0)	\n"	\
250	"	cache %1, 0x0e0(%0); cache %1, 0x0f0(%0)	\n"	\
251	"	cache %1, 0x100(%0); cache %1, 0x110(%0)	\n"	\
252	"	cache %1, 0x120(%0); cache %1, 0x130(%0)	\n"	\
253	"	cache %1, 0x140(%0); cache %1, 0x150(%0)	\n"	\
254	"	cache %1, 0x160(%0); cache %1, 0x170(%0)	\n"	\
255	"	cache %1, 0x180(%0); cache %1, 0x190(%0)	\n"	\
256	"	cache %1, 0x1a0(%0); cache %1, 0x1b0(%0)	\n"	\
257	"	cache %1, 0x1c0(%0); cache %1, 0x1d0(%0)	\n"	\
258	"	cache %1, 0x1e0(%0); cache %1, 0x1f0(%0)	\n"	\
259	"	.set pop					\n"	\
260		:							\
261		: "r" (base),						\
262		  "i" (op));
263
264#define cache32_unroll32(base,op)					\
265	__asm__ __volatile__(						\
266	"	.set push					\n"	\
267	"	.set noreorder					\n"	\
268	"	.set mips3					\n"	\
269	"	cache %1, 0x000(%0); cache %1, 0x020(%0)	\n"	\
270	"	cache %1, 0x040(%0); cache %1, 0x060(%0)	\n"	\
271	"	cache %1, 0x080(%0); cache %1, 0x0a0(%0)	\n"	\
272	"	cache %1, 0x0c0(%0); cache %1, 0x0e0(%0)	\n"	\
273	"	cache %1, 0x100(%0); cache %1, 0x120(%0)	\n"	\
274	"	cache %1, 0x140(%0); cache %1, 0x160(%0)	\n"	\
275	"	cache %1, 0x180(%0); cache %1, 0x1a0(%0)	\n"	\
276	"	cache %1, 0x1c0(%0); cache %1, 0x1e0(%0)	\n"	\
277	"	cache %1, 0x200(%0); cache %1, 0x220(%0)	\n"	\
278	"	cache %1, 0x240(%0); cache %1, 0x260(%0)	\n"	\
279	"	cache %1, 0x280(%0); cache %1, 0x2a0(%0)	\n"	\
280	"	cache %1, 0x2c0(%0); cache %1, 0x2e0(%0)	\n"	\
281	"	cache %1, 0x300(%0); cache %1, 0x320(%0)	\n"	\
282	"	cache %1, 0x340(%0); cache %1, 0x360(%0)	\n"	\
283	"	cache %1, 0x380(%0); cache %1, 0x3a0(%0)	\n"	\
284	"	cache %1, 0x3c0(%0); cache %1, 0x3e0(%0)	\n"	\
285	"	.set pop					\n"	\
286		:							\
287		: "r" (base),						\
288		  "i" (op));
289
290#define cache64_unroll32(base,op)					\
291	__asm__ __volatile__(						\
292	"	.set push					\n"	\
293	"	.set noreorder					\n"	\
294	"	.set mips3					\n"	\
295	"	cache %1, 0x000(%0); cache %1, 0x040(%0)	\n"	\
296	"	cache %1, 0x080(%0); cache %1, 0x0c0(%0)	\n"	\
297	"	cache %1, 0x100(%0); cache %1, 0x140(%0)	\n"	\
298	"	cache %1, 0x180(%0); cache %1, 0x1c0(%0)	\n"	\
299	"	cache %1, 0x200(%0); cache %1, 0x240(%0)	\n"	\
300	"	cache %1, 0x280(%0); cache %1, 0x2c0(%0)	\n"	\
301	"	cache %1, 0x300(%0); cache %1, 0x340(%0)	\n"	\
302	"	cache %1, 0x380(%0); cache %1, 0x3c0(%0)	\n"	\
303	"	cache %1, 0x400(%0); cache %1, 0x440(%0)	\n"	\
304	"	cache %1, 0x480(%0); cache %1, 0x4c0(%0)	\n"	\
305	"	cache %1, 0x500(%0); cache %1, 0x540(%0)	\n"	\
306	"	cache %1, 0x580(%0); cache %1, 0x5c0(%0)	\n"	\
307	"	cache %1, 0x600(%0); cache %1, 0x640(%0)	\n"	\
308	"	cache %1, 0x680(%0); cache %1, 0x6c0(%0)	\n"	\
309	"	cache %1, 0x700(%0); cache %1, 0x740(%0)	\n"	\
310	"	cache %1, 0x780(%0); cache %1, 0x7c0(%0)	\n"	\
311	"	.set pop					\n"	\
312		:							\
313		: "r" (base),						\
314		  "i" (op));
315
316#define cache128_unroll32(base,op)					\
317	__asm__ __volatile__(						\
318	"	.set push					\n"	\
319	"	.set noreorder					\n"	\
320	"	.set mips3					\n"	\
321	"	cache %1, 0x000(%0); cache %1, 0x080(%0)	\n"	\
322	"	cache %1, 0x100(%0); cache %1, 0x180(%0)	\n"	\
323	"	cache %1, 0x200(%0); cache %1, 0x280(%0)	\n"	\
324	"	cache %1, 0x300(%0); cache %1, 0x380(%0)	\n"	\
325	"	cache %1, 0x400(%0); cache %1, 0x480(%0)	\n"	\
326	"	cache %1, 0x500(%0); cache %1, 0x580(%0)	\n"	\
327	"	cache %1, 0x600(%0); cache %1, 0x680(%0)	\n"	\
328	"	cache %1, 0x700(%0); cache %1, 0x780(%0)	\n"	\
329	"	cache %1, 0x800(%0); cache %1, 0x880(%0)	\n"	\
330	"	cache %1, 0x900(%0); cache %1, 0x980(%0)	\n"	\
331	"	cache %1, 0xa00(%0); cache %1, 0xa80(%0)	\n"	\
332	"	cache %1, 0xb00(%0); cache %1, 0xb80(%0)	\n"	\
333	"	cache %1, 0xc00(%0); cache %1, 0xc80(%0)	\n"	\
334	"	cache %1, 0xd00(%0); cache %1, 0xd80(%0)	\n"	\
335	"	cache %1, 0xe00(%0); cache %1, 0xe80(%0)	\n"	\
336	"	cache %1, 0xf00(%0); cache %1, 0xf80(%0)	\n"	\
337	"	.set pop					\n"	\
338		:							\
339		: "r" (base),						\
340		  "i" (op));
341
342/* build blast_xxx, blast_xxx_page, blast_xxx_page_indexed */
343#define __BUILD_BLAST_CACHE(pfx, desc, indexop, hitop, lsize) \
344static inline void blast_##pfx##cache##lsize(void)			\
345{									\
346	unsigned long start = INDEX_BASE;				\
347	unsigned long end = start + current_cpu_data.desc.waysize;	\
348	unsigned long ws_inc = 1UL << current_cpu_data.desc.waybit;	\
349	unsigned long ws_end = current_cpu_data.desc.ways <<		\
350	                       current_cpu_data.desc.waybit;		\
351	unsigned long ws, addr;						\
352									\
353	__##pfx##flush_prologue						\
354									\
355	for (ws = 0; ws < ws_end; ws += ws_inc)				\
356		for (addr = start; addr < end; addr += lsize * 32)	\
357			cache##lsize##_unroll32(addr|ws, indexop);	\
358									\
359	__##pfx##flush_epilogue						\
360}									\
361									\
362static inline void blast_##pfx##cache##lsize##_page(unsigned long page)	\
363{									\
364	unsigned long start = page;					\
365	unsigned long end = page + PAGE_SIZE;				\
366									\
367	__##pfx##flush_prologue						\
368									\
369	do {								\
370		cache##lsize##_unroll32(start, hitop);			\
371		start += lsize * 32;					\
372	} while (start < end);						\
373									\
374	__##pfx##flush_epilogue						\
375}									\
376									\
377static inline void blast_##pfx##cache##lsize##_page_indexed(unsigned long page) \
378{									\
379	unsigned long indexmask = current_cpu_data.desc.waysize - 1;	\
380	unsigned long start = INDEX_BASE + (page & indexmask);		\
381	unsigned long end = start + PAGE_SIZE;				\
382	unsigned long ws_inc = 1UL << current_cpu_data.desc.waybit;	\
383	unsigned long ws_end = current_cpu_data.desc.ways <<		\
384	                       current_cpu_data.desc.waybit;		\
385	unsigned long ws, addr;						\
386									\
387	__##pfx##flush_prologue						\
388									\
389	for (ws = 0; ws < ws_end; ws += ws_inc)				\
390		for (addr = start; addr < end; addr += lsize * 32)	\
391			cache##lsize##_unroll32(addr|ws, indexop);	\
392									\
393	__##pfx##flush_epilogue						\
394}
395
396__BUILD_BLAST_CACHE(d, dcache, Index_Writeback_Inv_D, Hit_Writeback_Inv_D, 16)
397__BUILD_BLAST_CACHE(i, icache, Index_Invalidate_I, Hit_Invalidate_I, 16)
398__BUILD_BLAST_CACHE(s, scache, Index_Writeback_Inv_SD, Hit_Writeback_Inv_SD, 16)
399__BUILD_BLAST_CACHE(d, dcache, Index_Writeback_Inv_D, Hit_Writeback_Inv_D, 32)
400__BUILD_BLAST_CACHE(i, icache, Index_Invalidate_I, Hit_Invalidate_I, 32)
401__BUILD_BLAST_CACHE(s, scache, Index_Writeback_Inv_SD, Hit_Writeback_Inv_SD, 32)
402__BUILD_BLAST_CACHE(d, dcache, Index_Writeback_Inv_D, Hit_Writeback_Inv_D, 64)
403__BUILD_BLAST_CACHE(i, icache, Index_Invalidate_I, Hit_Invalidate_I, 64)
404__BUILD_BLAST_CACHE(s, scache, Index_Writeback_Inv_SD, Hit_Writeback_Inv_SD, 64)
405__BUILD_BLAST_CACHE(s, scache, Index_Writeback_Inv_SD, Hit_Writeback_Inv_SD, 128)
406
407__BUILD_BLAST_CACHE(inv_d, dcache, Index_Writeback_Inv_D, Hit_Invalidate_D, 16)
408__BUILD_BLAST_CACHE(inv_d, dcache, Index_Writeback_Inv_D, Hit_Invalidate_D, 32)
409__BUILD_BLAST_CACHE(inv_s, scache, Index_Writeback_Inv_SD, Hit_Invalidate_SD, 16)
410__BUILD_BLAST_CACHE(inv_s, scache, Index_Writeback_Inv_SD, Hit_Invalidate_SD, 32)
411__BUILD_BLAST_CACHE(inv_s, scache, Index_Writeback_Inv_SD, Hit_Invalidate_SD, 64)
412__BUILD_BLAST_CACHE(inv_s, scache, Index_Writeback_Inv_SD, Hit_Invalidate_SD, 128)
413
414/* build blast_xxx_range, protected_blast_xxx_range */
415#define __BUILD_BLAST_CACHE_RANGE(pfx, desc, hitop, prot) \
416static inline void prot##blast_##pfx##cache##_range(unsigned long start, \
417						    unsigned long end)	\
418{									\
419	unsigned long lsize = cpu_##desc##_line_size();			\
420	unsigned long addr = start & ~(lsize - 1);			\
421	unsigned long aend = (end - 1) & ~(lsize - 1);			\
422									\
423	__##pfx##flush_prologue						\
424									\
425	while (1) {							\
426		prot##cache_op(hitop, addr);				\
427		if (addr == aend)					\
428			break;						\
429		addr += lsize;						\
430	}								\
431									\
432	__##pfx##flush_epilogue						\
433}
434
435__BUILD_BLAST_CACHE_RANGE(d, dcache, Hit_Writeback_Inv_D, protected_)
436__BUILD_BLAST_CACHE_RANGE(s, scache, Hit_Writeback_Inv_SD, protected_)
437__BUILD_BLAST_CACHE_RANGE(i, icache, Hit_Invalidate_I, protected_)
438__BUILD_BLAST_CACHE_RANGE(d, dcache, Hit_Writeback_Inv_D, )
439__BUILD_BLAST_CACHE_RANGE(s, scache, Hit_Writeback_Inv_SD, )
440/* blast_inv_dcache_range */
441__BUILD_BLAST_CACHE_RANGE(inv_d, dcache, Hit_Invalidate_D, )
442__BUILD_BLAST_CACHE_RANGE(inv_s, scache, Hit_Invalidate_SD, )
443
444#endif /* _ASM_R4KCACHE_H */
445