more fixes for the crazy cpu cache issues on brcm47xx-2.6 (mainly affects 4704) ...
[openwrt.git] / target / linux / brcm47xx-2.6 / patches-2.6.22 / 150-cpu_fixes.patch
1 Index: linux-2.6.22-rc6/arch/mips/kernel/genex.S
2 ===================================================================
3 --- linux-2.6.22-rc6.orig/arch/mips/kernel/genex.S      2007-07-04 01:52:47.812492000 +0200
4 +++ linux-2.6.22-rc6/arch/mips/kernel/genex.S   2007-07-04 01:53:01.585352750 +0200
5 @@ -51,6 +51,10 @@
6  NESTED(except_vec3_generic, 0, sp)
7         .set    push
8         .set    noat
9 +#ifdef CONFIG_BCM947XX
10 +       nop
11 +       nop
12 +#endif
13  #if R5432_CP0_INTERRUPT_WAR
14         mfc0    k0, CP0_INDEX
15  #endif
16 Index: linux-2.6.22-rc6/arch/mips/mm/c-r4k.c
17 ===================================================================
18 --- linux-2.6.22-rc6.orig/arch/mips/mm/c-r4k.c  2007-07-04 01:53:01.545350250 +0200
19 +++ linux-2.6.22-rc6/arch/mips/mm/c-r4k.c       2007-07-04 02:17:11.435962750 +0200
20 @@ -29,6 +29,9 @@
21  #include <asm/cacheflush.h> /* for run_uncached() */
22  
23  
24 +/* For enabling BCM4710 cache workarounds */
25 +int bcm4710 = 0;
26 +
27  /*
28   * Special Variant of smp_call_function for use by cache functions:
29   *
30 @@ -93,6 +96,9 @@
31  {
32         unsigned long  dc_lsize = cpu_dcache_line_size();
33  
34 +       if (bcm4710)
35 +               r4k_blast_dcache_page = blast_dcache_page;
36 +       else
37         if (dc_lsize == 0)
38                 r4k_blast_dcache_page = (void *)cache_noop;
39         else if (dc_lsize == 16)
40 @@ -107,6 +113,9 @@
41  {
42         unsigned long dc_lsize = cpu_dcache_line_size();
43  
44 +       if (bcm4710)
45 +               r4k_blast_dcache_page_indexed = blast_dcache_page_indexed;
46 +       else
47         if (dc_lsize == 0)
48                 r4k_blast_dcache_page_indexed = (void *)cache_noop;
49         else if (dc_lsize == 16)
50 @@ -121,6 +130,9 @@
51  {
52         unsigned long dc_lsize = cpu_dcache_line_size();
53  
54 +       if (bcm4710)
55 +               r4k_blast_dcache = blast_dcache;
56 +       else
57         if (dc_lsize == 0)
58                 r4k_blast_dcache = (void *)cache_noop;
59         else if (dc_lsize == 16)
60 @@ -538,6 +550,9 @@
61                 r4k_blast_icache();
62         else
63                 protected_blast_icache_range(start, end);
64 +
65 +       if (bcm4710)
66 +               r4k_flush_cache_all();
67  }
68  
69  static void r4k_flush_icache_range(unsigned long start, unsigned long end)
70 @@ -618,6 +633,8 @@
71         unsigned long addr = (unsigned long) arg;
72  
73         R4600_HIT_CACHEOP_WAR_IMPL;
74 +       BCM4710_PROTECTED_FILL_TLB(addr);
75 +       BCM4710_PROTECTED_FILL_TLB(addr + 4);
76         if (dc_lsize)
77                 protected_writeback_dcache_line(addr & ~(dc_lsize - 1));
78         if (!cpu_icache_snoops_remote_store && scache_size)
79 @@ -1144,6 +1161,17 @@
80          * silly idea of putting something else there ...
81          */
82         switch (current_cpu_data.cputype) {
83 +       case CPU_BCM3302:
84 +               {
85 +                       u32 cm;
86 +                       cm = read_c0_diag();
87 +                       /* Enable icache */
88 +                       cm |= (1 << 31);
89 +                       /* Enable dcache */
90 +                       cm |= (1 << 30);
91 +                       write_c0_diag(cm);
92 +               }
93 +               break;
94         case CPU_R4000PC:
95         case CPU_R4000SC:
96         case CPU_R4000MC:
97 @@ -1174,6 +1202,15 @@
98         /* Default cache error handler for R4000 and R5000 family */
99         set_uncached_handler (0x100, &except_vec2_generic, 0x80);
100  
101 +       /* Check if special workarounds are required */
102 +#ifdef CONFIG_BCM947XX
103 +       if (current_cpu_data.cputype == CPU_BCM4710 && (current_cpu_data.processor_id & 0xff) == 0) {
104 +               printk("Enabling BCM4710A0 cache workarounds.\n");
105 +               bcm4710 = 1;
106 +       } else
107 +#endif
108 +               bcm4710 = 0;
109 +
110         probe_pcache();
111         setup_scache();
112  
113 @@ -1219,5 +1256,13 @@
114         build_clear_page();
115         build_copy_page();
116         local_r4k___flush_cache_all(NULL);
117 +#ifdef CONFIG_BCM947XX
118 +       {
119 +               static void (*_coherency_setup)(void);
120 +               _coherency_setup = (void (*)(void)) KSEG1ADDR(coherency_setup);
121 +               _coherency_setup();
122 +       }
123 +#else
124         coherency_setup();
125 +#endif
126  }
127 Index: linux-2.6.22-rc6/arch/mips/mm/tlbex.c
128 ===================================================================
129 --- linux-2.6.22-rc6.orig/arch/mips/mm/tlbex.c  2007-07-04 01:53:01.193328250 +0200
130 +++ linux-2.6.22-rc6/arch/mips/mm/tlbex.c       2007-07-04 02:17:26.112880000 +0200
131 @@ -1229,6 +1229,10 @@
132  #endif
133  }
134  
135 +#ifdef CONFIG_BCM947XX
136 +extern int bcm4710;
137 +#endif
138 +
139  static void __init build_r4000_tlb_refill_handler(void)
140  {
141         u32 *p = tlb_handler;
142 @@ -1243,6 +1247,11 @@
143         memset(relocs, 0, sizeof(relocs));
144         memset(final_handler, 0, sizeof(final_handler));
145  
146 +#ifdef CONFIG_BCM947XX
147 +       if (current_cpu_data.cputype == CPU_BCM3302)
148 +               i_nop(&p);
149 +#endif
150 +
151         /*
152          * create the plain linear handler
153          */
154 Index: linux-2.6.22-rc6/include/asm-mips/r4kcache.h
155 ===================================================================
156 --- linux-2.6.22-rc6.orig/include/asm-mips/r4kcache.h   2007-07-04 01:52:47.840493750 +0200
157 +++ linux-2.6.22-rc6/include/asm-mips/r4kcache.h        2007-07-04 01:53:01.673358250 +0200
158 @@ -17,6 +17,20 @@
159  #include <asm/cpu-features.h>
160  #include <asm/mipsmtregs.h>
161  
162 +#ifdef CONFIG_BCM947XX
163 +#include <asm/paccess.h>
164 +#include <linux/ssb/ssb.h>
165 +#define BCM4710_DUMMY_RREG() ((void) *((u8 *) KSEG1ADDR(SSB_ENUM_BASE + SSB_IMSTATE)))
166 +
167 +#define BCM4710_FILL_TLB(addr) (*(volatile unsigned long *)(addr))
168 +#define BCM4710_PROTECTED_FILL_TLB(addr) ({ unsigned long x; get_dbe(x, (volatile unsigned long *)(addr)); })
169 +#else
170 +#define BCM4710_DUMMY_RREG()
171 +
172 +#define BCM4710_FILL_TLB(addr)
173 +#define BCM4710_PROTECTED_FILL_TLB(addr)
174 +#endif
175 +
176  /*
177   * This macro return a properly sign-extended address suitable as base address
178   * for indexed cache operations.  Two issues here:
179 @@ -150,6 +164,7 @@
180  static inline void flush_dcache_line_indexed(unsigned long addr)
181  {
182         __dflush_prologue
183 +       BCM4710_DUMMY_RREG();
184         cache_op(Index_Writeback_Inv_D, addr);
185         __dflush_epilogue
186  }
187 @@ -169,6 +184,7 @@
188  static inline void flush_dcache_line(unsigned long addr)
189  {
190         __dflush_prologue
191 +       BCM4710_DUMMY_RREG();
192         cache_op(Hit_Writeback_Inv_D, addr);
193         __dflush_epilogue
194  }
195 @@ -176,6 +192,7 @@
196  static inline void invalidate_dcache_line(unsigned long addr)
197  {
198         __dflush_prologue
199 +       BCM4710_DUMMY_RREG();
200         cache_op(Hit_Invalidate_D, addr);
201         __dflush_epilogue
202  }
203 @@ -208,6 +225,7 @@
204   */
205  static inline void protected_flush_icache_line(unsigned long addr)
206  {
207 +       BCM4710_DUMMY_RREG();
208         protected_cache_op(Hit_Invalidate_I, addr);
209  }
210  
211 @@ -219,6 +237,7 @@
212   */
213  static inline void protected_writeback_dcache_line(unsigned long addr)
214  {
215 +       BCM4710_DUMMY_RREG();
216         protected_cache_op(Hit_Writeback_Inv_D, addr);
217  }
218  
219 @@ -339,8 +358,52 @@
220                 : "r" (base),                                           \
221                   "i" (op));
222  
223 +static inline void blast_dcache(void)
224 +{
225 +       unsigned long start = KSEG0;
226 +       unsigned long dcache_size = current_cpu_data.dcache.waysize * current_cpu_data.dcache.ways;
227 +       unsigned long end = (start + dcache_size);
228 +
229 +       do {
230 +               BCM4710_DUMMY_RREG();
231 +               cache_op(Index_Writeback_Inv_D, start);
232 +               start += current_cpu_data.dcache.linesz;
233 +       } while(start < end);
234 +}
235 +
236 +static inline void blast_dcache_page(unsigned long page)
237 +{
238 +       unsigned long start = page;
239 +       unsigned long end = start + PAGE_SIZE;
240 +
241 +       BCM4710_FILL_TLB(start);
242 +       do {
243 +               BCM4710_DUMMY_RREG();
244 +               cache_op(Hit_Writeback_Inv_D, start);
245 +               start += current_cpu_data.dcache.linesz;
246 +       } while(start < end);
247 +}
248 +
249 +static inline void blast_dcache_page_indexed(unsigned long page)
250 +{
251 +       unsigned long start = page;
252 +       unsigned long end = start + PAGE_SIZE;
253 +       unsigned long ws_inc = 1UL << current_cpu_data.dcache.waybit;
254 +       unsigned long ws_end = current_cpu_data.dcache.ways <<
255 +                              current_cpu_data.dcache.waybit;
256 +       unsigned long ws, addr;
257 +       for (ws = 0; ws < ws_end; ws += ws_inc) {
258 +               start = page + ws;
259 +               for (addr = start; addr < end; addr += current_cpu_data.dcache.linesz) {
260 +                       BCM4710_DUMMY_RREG();
261 +                       cache_op(Index_Writeback_Inv_D, addr);
262 +               }
263 +       }
264 +}
265 +
266 +
267  /* build blast_xxx, blast_xxx_page, blast_xxx_page_indexed */
268 -#define __BUILD_BLAST_CACHE(pfx, desc, indexop, hitop, lsize) \
269 +#define __BUILD_BLAST_CACHE(pfx, desc, indexop, hitop, lsize, war) \
270  static inline void blast_##pfx##cache##lsize(void)                     \
271  {                                                                      \
272         unsigned long start = INDEX_BASE;                               \
273 @@ -352,6 +415,7 @@
274                                                                         \
275         __##pfx##flush_prologue                                         \
276                                                                         \
277 +       war                                                             \
278         for (ws = 0; ws < ws_end; ws += ws_inc)                         \
279                 for (addr = start; addr < end; addr += lsize * 32)      \
280                         cache##lsize##_unroll32(addr|ws,indexop);       \
281 @@ -366,6 +430,7 @@
282                                                                         \
283         __##pfx##flush_prologue                                         \
284                                                                         \
285 +       war                                                             \
286         do {                                                            \
287                 cache##lsize##_unroll32(start,hitop);                   \
288                 start += lsize * 32;                                    \
289 @@ -384,6 +449,8 @@
290                                current_cpu_data.desc.waybit;            \
291         unsigned long ws, addr;                                         \
292                                                                         \
293 +       war                                                             \
294 +                                                                       \
295         __##pfx##flush_prologue                                         \
296                                                                         \
297         for (ws = 0; ws < ws_end; ws += ws_inc)                         \
298 @@ -393,28 +460,30 @@
299         __##pfx##flush_epilogue                                         \
300  }
301  
302 -__BUILD_BLAST_CACHE(d, dcache, Index_Writeback_Inv_D, Hit_Writeback_Inv_D, 16)
303 -__BUILD_BLAST_CACHE(i, icache, Index_Invalidate_I, Hit_Invalidate_I, 16)
304 -__BUILD_BLAST_CACHE(s, scache, Index_Writeback_Inv_SD, Hit_Writeback_Inv_SD, 16)
305 -__BUILD_BLAST_CACHE(d, dcache, Index_Writeback_Inv_D, Hit_Writeback_Inv_D, 32)
306 -__BUILD_BLAST_CACHE(i, icache, Index_Invalidate_I, Hit_Invalidate_I, 32)
307 -__BUILD_BLAST_CACHE(s, scache, Index_Writeback_Inv_SD, Hit_Writeback_Inv_SD, 32)
308 -__BUILD_BLAST_CACHE(i, icache, Index_Invalidate_I, Hit_Invalidate_I, 64)
309 -__BUILD_BLAST_CACHE(s, scache, Index_Writeback_Inv_SD, Hit_Writeback_Inv_SD, 64)
310 -__BUILD_BLAST_CACHE(s, scache, Index_Writeback_Inv_SD, Hit_Writeback_Inv_SD, 128)
311 +__BUILD_BLAST_CACHE(d, dcache, Index_Writeback_Inv_D, Hit_Writeback_Inv_D, 16, )
312 +__BUILD_BLAST_CACHE(i, icache, Index_Invalidate_I, Hit_Invalidate_I, 16, BCM4710_FILL_TLB(start);)
313 +__BUILD_BLAST_CACHE(s, scache, Index_Writeback_Inv_SD, Hit_Writeback_Inv_SD, 16, )
314 +__BUILD_BLAST_CACHE(d, dcache, Index_Writeback_Inv_D, Hit_Writeback_Inv_D, 32, )
315 +__BUILD_BLAST_CACHE(i, icache, Index_Invalidate_I, Hit_Invalidate_I, 32, BCM4710_FILL_TLB(start);)
316 +__BUILD_BLAST_CACHE(s, scache, Index_Writeback_Inv_SD, Hit_Writeback_Inv_SD, 32, )
317 +__BUILD_BLAST_CACHE(i, icache, Index_Invalidate_I, Hit_Invalidate_I, 64, BCM4710_FILL_TLB(start);)
318 +__BUILD_BLAST_CACHE(s, scache, Index_Writeback_Inv_SD, Hit_Writeback_Inv_SD, 64, )
319 +__BUILD_BLAST_CACHE(s, scache, Index_Writeback_Inv_SD, Hit_Writeback_Inv_SD, 128, )
320  
321  /* build blast_xxx_range, protected_blast_xxx_range */
322 -#define __BUILD_BLAST_CACHE_RANGE(pfx, desc, hitop, prot) \
323 +#define __BUILD_BLAST_CACHE_RANGE(pfx, desc, hitop, prot, war, war2) \
324  static inline void prot##blast_##pfx##cache##_range(unsigned long start, \
325                                                     unsigned long end)  \
326  {                                                                      \
327         unsigned long lsize = cpu_##desc##_line_size();                 \
328         unsigned long addr = start & ~(lsize - 1);                      \
329         unsigned long aend = (end - 1) & ~(lsize - 1);                  \
330 +       war                                                             \
331                                                                         \
332         __##pfx##flush_prologue                                         \
333                                                                         \
334         while (1) {                                                     \
335 +               war2                                            \
336                 prot##cache_op(hitop, addr);                            \
337                 if (addr == aend)                                       \
338                         break;                                          \
339 @@ -424,13 +493,13 @@
340         __##pfx##flush_epilogue                                         \
341  }
342  
343 -__BUILD_BLAST_CACHE_RANGE(d, dcache, Hit_Writeback_Inv_D, protected_)
344 -__BUILD_BLAST_CACHE_RANGE(s, scache, Hit_Writeback_Inv_SD, protected_)
345 -__BUILD_BLAST_CACHE_RANGE(i, icache, Hit_Invalidate_I, protected_)
346 -__BUILD_BLAST_CACHE_RANGE(d, dcache, Hit_Writeback_Inv_D, )
347 -__BUILD_BLAST_CACHE_RANGE(s, scache, Hit_Writeback_Inv_SD, )
348 +__BUILD_BLAST_CACHE_RANGE(d, dcache, Hit_Writeback_Inv_D, protected_, BCM4710_PROTECTED_FILL_TLB(addr); BCM4710_PROTECTED_FILL_TLB(aend);, BCM4710_DUMMY_RREG();)
349 +__BUILD_BLAST_CACHE_RANGE(s, scache, Hit_Writeback_Inv_SD, protected_,, )
350 +__BUILD_BLAST_CACHE_RANGE(i, icache, Hit_Invalidate_I, protected_,, )
351 +__BUILD_BLAST_CACHE_RANGE(d, dcache, Hit_Writeback_Inv_D,, BCM4710_FILL_TLB(addr); BCM4710_FILL_TLB(aend);, BCM4710_DUMMY_RREG();)
352 +__BUILD_BLAST_CACHE_RANGE(s, scache, Hit_Writeback_Inv_SD,,, )
353  /* blast_inv_dcache_range */
354 -__BUILD_BLAST_CACHE_RANGE(inv_d, dcache, Hit_Invalidate_D, )
355 -__BUILD_BLAST_CACHE_RANGE(inv_s, scache, Hit_Invalidate_SD, )
356 +__BUILD_BLAST_CACHE_RANGE(inv_d, dcache, Hit_Invalidate_D,,,BCM4710_DUMMY_RREG();)
357 +__BUILD_BLAST_CACHE_RANGE(inv_s, scache, Hit_Invalidate_SD,,, )
358  
359  #endif /* _ASM_R4KCACHE_H */
360 Index: linux-2.6.22-rc6/include/asm-mips/stackframe.h
361 ===================================================================
362 --- linux-2.6.22-rc6.orig/include/asm-mips/stackframe.h 2007-07-04 01:52:47.852494500 +0200
363 +++ linux-2.6.22-rc6/include/asm-mips/stackframe.h      2007-07-04 01:53:01.697359750 +0200
364 @@ -350,6 +350,10 @@
365                 .macro  RESTORE_SP_AND_RET
366                 LONG_L  sp, PT_R29(sp)
367                 .set    mips3
368 +#ifdef CONFIG_BCM947XX
369 +               nop
370 +               nop
371 +#endif
372                 eret
373                 .set    mips0
374                 .endm