1 | /* |
---|
2 | * mmuAsm.S |
---|
3 | * |
---|
4 | * $Id$ |
---|
5 | * |
---|
6 | * Copyright (C) 1999 Eric Valette (valette@crf.canon.fr) |
---|
7 | * |
---|
8 | * This file contains the low-level support for various MMU |
---|
9 | * features. |
---|
10 | * |
---|
11 | * The license and distribution terms for this file may be |
---|
12 | * found in found in the file LICENSE in this distribution or at |
---|
13 | * http://www.rtems.com/license/LICENSE. |
---|
14 | * |
---|
15 | * T. Straumann - 11/2001: added support for 7400 (no AltiVec yet) |
---|
16 | */ |
---|
17 | |
---|
18 | #include <asm.h> |
---|
19 | #include <rtems/score/cpu.h> |
---|
20 | #include <libcpu/io.h> |
---|
21 | |
---|
22 | /* Unfortunately, the CPU types defined in cpu.h are |
---|
23 | * an 'enum' type and hence not available :-( |
---|
24 | */ |
---|
25 | #define PPC_601 0x1 |
---|
26 | #define PPC_603 0x3 |
---|
27 | #define PPC_604 0x4 |
---|
28 | #define PPC_603e 0x6 |
---|
29 | #define PPC_603ev 0x7 |
---|
30 | #define PPC_750 0x8 |
---|
31 | #define PPC_604e 0x9 |
---|
32 | #define PPC_604r 0xA |
---|
33 | #define PPC_7400 0xC |
---|
34 | #define PPC_620 0x16 |
---|
35 | #define PPC_860 0x50 |
---|
36 | #define PPC_821 PPC_860 |
---|
37 | #define PPC_8260 0x81 |
---|
38 | |
---|
39 | /* ALTIVEC instructions (not recognized by off-the shelf gcc yet) */ |
---|
40 | #define DSSALL .long 0x7e00066c /* DSSALL altivec instruction opcode */ |
---|
41 | |
---|
42 | /* A couple of defines to make the code more readable */ |
---|
43 | #define CACHE_LINE_SIZE 32 |
---|
44 | |
---|
45 | #ifndef MSSCR0 |
---|
46 | #define MSSCR0 1014 |
---|
47 | #else |
---|
48 | #warning MSSCR0 seems to be known, update __FILE__ |
---|
49 | #endif |
---|
50 | |
---|
51 | #define DL1HWF (1<<(31-8)) |
---|
52 | #define L2HWF (1<<(31-20)) |
---|
53 | |
---|
54 | |
---|
55 | |
---|
56 | /* |
---|
57 | * Each setdbat routine start by invalidating the DBAT as some |
---|
58 | * proc (604e) request the valid bit set to 0 before accepting |
---|
59 | * to write in BAT |
---|
60 | */ |
---|
61 | |
---|
62 | .globl asm_setdbat0 |
---|
63 | .type asm_setdbat0,@function |
---|
64 | asm_setdbat0: |
---|
65 | li r0,0 |
---|
66 | sync |
---|
67 | isync |
---|
68 | mtspr DBAT0U,r0 |
---|
69 | mtspr DBAT0L,r0 |
---|
70 | sync |
---|
71 | isync |
---|
72 | mtspr DBAT0L, r4 |
---|
73 | mtspr DBAT0U, r3 |
---|
74 | sync |
---|
75 | isync |
---|
76 | blr |
---|
77 | |
---|
78 | .globl asm_setdbat1 |
---|
79 | .type asm_setdbat1,@function |
---|
80 | asm_setdbat1: |
---|
81 | li r0,0 |
---|
82 | sync |
---|
83 | isync |
---|
84 | mtspr DBAT1U,r0 |
---|
85 | mtspr DBAT1L,r0 |
---|
86 | sync |
---|
87 | isync |
---|
88 | mtspr DBAT1L, r4 |
---|
89 | mtspr DBAT1U, r3 |
---|
90 | sync |
---|
91 | isync |
---|
92 | blr |
---|
93 | |
---|
94 | .globl asm_setdbat2 |
---|
95 | .type asm_setdbat2,@function |
---|
96 | asm_setdbat2: |
---|
97 | li r0,0 |
---|
98 | sync |
---|
99 | isync |
---|
100 | mtspr DBAT2U,r0 |
---|
101 | mtspr DBAT2L,r0 |
---|
102 | sync |
---|
103 | isync |
---|
104 | mtspr DBAT2L, r4 |
---|
105 | mtspr DBAT2U, r3 |
---|
106 | sync |
---|
107 | isync |
---|
108 | blr |
---|
109 | |
---|
110 | .globl asm_setdbat3 |
---|
111 | .type asm_setdbat3,@function |
---|
112 | asm_setdbat3: |
---|
113 | li r0,0 |
---|
114 | sync |
---|
115 | isync |
---|
116 | mtspr DBAT3U,r0 |
---|
117 | mtspr DBAT3L,r0 |
---|
118 | sync |
---|
119 | isync |
---|
120 | mtspr DBAT3L, r4 |
---|
121 | mtspr DBAT3U, r3 |
---|
122 | sync |
---|
123 | isync |
---|
124 | blr |
---|
125 | |
---|
126 | .globl L1_caches_enables |
---|
127 | .type L1_caches_enables, @function |
---|
128 | |
---|
129 | L1_caches_enables: |
---|
130 | /* |
---|
131 | * Enable caches and 604-specific features if necessary. |
---|
132 | */ |
---|
133 | mfspr r9,PVR |
---|
134 | rlwinm r9,r9,16,16,31 |
---|
135 | cmpi 0,r9,PPC_601 |
---|
136 | beq 4f /* not needed for 601 */ |
---|
137 | mfspr r11,HID0 |
---|
138 | andi. r0,r11,HID0_DCE |
---|
139 | ori r11,r11,HID0_ICE|HID0_DCE |
---|
140 | ori r8,r11,HID0_ICFI |
---|
141 | bne 3f /* don't invalidate the D-cache */ |
---|
142 | ori r8,r8,HID0_DCI /* unless it wasn't enabled */ |
---|
143 | 3: |
---|
144 | sync |
---|
145 | mtspr HID0,r8 /* enable and invalidate caches */ |
---|
146 | sync |
---|
147 | mtspr HID0,r11 /* enable caches */ |
---|
148 | sync |
---|
149 | isync |
---|
150 | cmpi 0,r9,PPC_604 /* check for 604 */ |
---|
151 | cmpi 1,r9,PPC_604e /* or 604e */ |
---|
152 | cmpi 2,r9,PPC_604r /* or mach5 */ |
---|
153 | cror 2,2,6 |
---|
154 | cror 2,2,10 |
---|
155 | cmpi 1,r9,PPC_750 /* or 750 */ |
---|
156 | cror 2,2,6 |
---|
157 | cmpi 1,r9,PPC_7400 /* or 7400 */ |
---|
158 | bne 3f |
---|
159 | ori r11,r11,HID0_BTIC /* enable branch tgt cache on 7400 */ |
---|
160 | 3: cror 2,2,6 |
---|
161 | bne 4f |
---|
162 | /* on 7400 SIED is actually SGE (store gathering enable) */ |
---|
163 | ori r11,r11,HID0_SIED|HID0_BHTE /* for 604[e], enable */ |
---|
164 | bne 2,5f |
---|
165 | ori r11,r11,HID0_BTCD |
---|
166 | 5: mtspr HID0,r11 /* superscalar exec & br history tbl */ |
---|
167 | 4: |
---|
168 | blr |
---|
169 | |
---|
170 | .globl get_L2CR |
---|
171 | .type get_L2CR, @function |
---|
172 | get_L2CR: |
---|
173 | /* Make sure this is a > 750 chip */ |
---|
174 | mfspr r3,PVR |
---|
175 | rlwinm r3,r3,16,16,31 |
---|
176 | cmplwi r3,PPC_750 /* it's a 750 */ |
---|
177 | beq 1f |
---|
178 | cmplwi r3,PPC_7400 /* it's a 7400 */ |
---|
179 | beq 1f |
---|
180 | li r3,0 |
---|
181 | blr |
---|
182 | |
---|
183 | 1: |
---|
184 | /* Return the L2CR contents */ |
---|
185 | mfspr r3,L2CR |
---|
186 | blr |
---|
187 | |
---|
188 | .globl set_L2CR |
---|
189 | .type set_L2CR, @function |
---|
190 | set_L2CR: |
---|
191 | /* Usage: |
---|
192 | * When setting the L2CR register, you must do a few special things. |
---|
193 | * If you are enabling the cache, you must perform a global invalidate. |
---|
194 | * If you are disabling the cache, you must flush the cache contents first. |
---|
195 | * This routine takes care of doing these things. When first |
---|
196 | * enabling the cache, make sure you pass in the L2CR you want, as well as |
---|
197 | * passing in the global invalidate bit set. A global invalidate will |
---|
198 | * only be performed if the L2I bit is set in applyThis. When enabling |
---|
199 | * the cache, you should also set the L2E bit in applyThis. If you |
---|
200 | * want to modify the L2CR contents after the cache has been enabled, |
---|
201 | * the recommended procedure is to first call __setL2CR(0) to disable |
---|
202 | * the cache and then call it again with the new values for L2CR. Examples: |
---|
203 | * |
---|
204 | * _setL2CR(0) - disables the cache |
---|
205 | * _setL2CR(0xb9A14000) - enables my G3 MCP750 card: |
---|
206 | * - L2E set to turn on the cache |
---|
207 | * - L2SIZ set to 1MB |
---|
208 | * - L2CLK set to %2 |
---|
209 | * - L2RAM set to pipelined syncronous late-write |
---|
210 | * - L2I set to perform a global invalidation |
---|
211 | * - L2OH set to 1 nS |
---|
212 | * |
---|
213 | * A similar call should work for your card. You need to know the correct |
---|
214 | * setting for your card and then place them in the fields I have outlined |
---|
215 | * above. Other fields support optional features, such as L2DO which caches |
---|
216 | * only data, or L2TS which causes cache pushes from the L1 cache to go to |
---|
217 | *the L2 cache instead of to main memory. |
---|
218 | */ |
---|
219 | |
---|
220 | /* Make sure this is a > 750 chip */ |
---|
221 | mfspr r0,PVR |
---|
222 | rlwinm r0,r0,16,16,31 |
---|
223 | cmplwi r0,PPC_750 |
---|
224 | beq thisIs750 |
---|
225 | cmplwi r0,PPC_7400 |
---|
226 | beq thisIs750 |
---|
227 | li r3,-1 |
---|
228 | blr |
---|
229 | |
---|
230 | thisIs750: |
---|
231 | /* Get the current enable bit of the L2CR into r4 */ |
---|
232 | mfspr r4,L2CR |
---|
233 | rlwinm r4,r4,0,0,0 |
---|
234 | |
---|
235 | /* See if we want to perform a global inval this time. */ |
---|
236 | rlwinm r6,r3,0,10,10 /* r6 contains the new invalidate bit */ |
---|
237 | rlwinm. r5,r3,0,0,0 /* r5 contains the new enable bit */ |
---|
238 | rlwinm r3,r3,0,11,9 /* Turn off the invalidate bit */ |
---|
239 | rlwinm r3,r3,0,1,31 /* Turn off the enable bit */ |
---|
240 | or r3,r3,r4 /* Keep the enable bit the same as it was for now. */ |
---|
241 | mfmsr r7 /* shut off interrupts around critical flush/invalidate sections */ |
---|
242 | rlwinm r4,r7,0,17,15 /* Turn off EE bit - an external exception while we are flushing |
---|
243 | the cache is fatal (comment this line and see!) */ |
---|
244 | mtmsr r4 |
---|
245 | bne dontDisableCache /* Only disable the cache if L2CRApply has the enable bit off */ |
---|
246 | |
---|
247 | cmplwi r0,PPC_7400 /* > 7400 ? */ |
---|
248 | bne disableCache /* use traditional method */ |
---|
249 | |
---|
250 | /* On the 7400, they recommend using the hardware flush feature */ |
---|
251 | DSSALL /* stop all data streams */ |
---|
252 | sync |
---|
253 | /* we wouldn't have to flush L1, but for sake of consistency with the other code we do it anyway */ |
---|
254 | mfspr r4, MSSCR0 |
---|
255 | oris r4, r4, DL1HWF@h |
---|
256 | mtspr MSSCR0, r4 |
---|
257 | sync |
---|
258 | /* L1 flushed */ |
---|
259 | mfspr r4, L2CR |
---|
260 | ori r4, r4, L2HWF |
---|
261 | mtspr L2CR, r4 |
---|
262 | sync |
---|
263 | /* L2 flushed */ |
---|
264 | b flushDone |
---|
265 | |
---|
266 | disableCache: |
---|
267 | /* Disable the cache. First, we turn off data relocation. */ |
---|
268 | rlwinm r4,r4,0,28,26 /* Turn off DR bit */ |
---|
269 | mtmsr r4 |
---|
270 | isync /* make sure memory accesses have completed */ |
---|
271 | |
---|
272 | /* |
---|
273 | Now, read the first 2MB of memory to put new data in the cache. |
---|
274 | (Actually we only need the size of the L2 cache plus |
---|
275 | the size of the L1 cache, but 2MB will cover everything just to be safe). |
---|
276 | */ |
---|
277 | lis r4,0x0001 |
---|
278 | mtctr r4 |
---|
279 | li r4,0 |
---|
280 | loadLoop: |
---|
281 | lwzx r0,r0,r4 |
---|
282 | addi r4,r4,CACHE_LINE_SIZE /* Go to start of next cache line */ |
---|
283 | bdnz loadLoop |
---|
284 | |
---|
285 | /* Now, flush the first 2MB of memory */ |
---|
286 | lis r4,0x0001 |
---|
287 | mtctr r4 |
---|
288 | li r4,0 |
---|
289 | sync |
---|
290 | flushLoop: |
---|
291 | dcbf r0,r4 |
---|
292 | addi r4,r4,CACHE_LINE_SIZE /* Go to start of next cache line */ |
---|
293 | bdnz flushLoop |
---|
294 | sync |
---|
295 | |
---|
296 | rlwinm r4,r7,0,17,15 /* still mask EE but reenable data relocation */ |
---|
297 | mtmsr r4 |
---|
298 | isync |
---|
299 | |
---|
300 | flushDone: |
---|
301 | |
---|
302 | /* Turn off the L2CR enable bit. */ |
---|
303 | rlwinm r3,r3,0,1,31 |
---|
304 | |
---|
305 | dontDisableCache: |
---|
306 | /* Set up the L2CR configuration bits */ |
---|
307 | sync |
---|
308 | mtspr L2CR,r3 |
---|
309 | sync |
---|
310 | cmplwi r6,0 |
---|
311 | beq noInval |
---|
312 | |
---|
313 | /* Perform a global invalidation */ |
---|
314 | oris r3,r3,0x0020 |
---|
315 | sync |
---|
316 | mtspr L2CR,r3 |
---|
317 | sync |
---|
318 | invalCompleteLoop: /* Wait for the invalidation to complete */ |
---|
319 | mfspr r3,L2CR |
---|
320 | rlwinm. r4,r3,0,31,31 |
---|
321 | bne invalCompleteLoop |
---|
322 | |
---|
323 | rlwinm r3,r3,0,11,9; /* Turn off the L2I bit */ |
---|
324 | sync |
---|
325 | mtspr L2CR,r3 |
---|
326 | sync |
---|
327 | |
---|
328 | noInval: |
---|
329 | /* re-enable interrupts, i.e. restore original MSR */ |
---|
330 | mtmsr r7 /* (no sync needed) */ |
---|
331 | /* See if we need to enable the cache */ |
---|
332 | cmplwi r5,0 |
---|
333 | beqlr |
---|
334 | |
---|
335 | enableCache: |
---|
336 | /* Enable the cache */ |
---|
337 | oris r3,r3,0x8000 |
---|
338 | mtspr L2CR,r3 |
---|
339 | sync |
---|
340 | blr |
---|