diff options
author | Linus Torvalds <torvalds@ppc970.osdl.org> | 2005-04-16 18:20:36 -0400 |
---|---|---|
committer | Linus Torvalds <torvalds@ppc970.osdl.org> | 2005-04-16 18:20:36 -0400 |
commit | 1da177e4c3f41524e886b7f1b8a0c1fc7321cac2 (patch) | |
tree | 0bba044c4ce775e45a88a51686b5d9f90697ea9d /arch/mips/mm/pg-sb1.c |
Linux-2.6.12-rc2v2.6.12-rc2
Initial git repository build. I'm not bothering with the full history,
even though we have it. We can create a separate "historical" git
archive of that later if we want to, and in the meantime it's about
3.2GB when imported into git - space that would just make the early
git days unnecessarily complicated, when we don't have a lot of good
infrastructure for it.
Let it rip!
Diffstat (limited to 'arch/mips/mm/pg-sb1.c')
-rw-r--r-- | arch/mips/mm/pg-sb1.c | 287 |
1 files changed, 287 insertions, 0 deletions
diff --git a/arch/mips/mm/pg-sb1.c b/arch/mips/mm/pg-sb1.c new file mode 100644 index 000000000000..59d131b5e536 --- /dev/null +++ b/arch/mips/mm/pg-sb1.c | |||
@@ -0,0 +1,287 @@ | |||
1 | /* | ||
2 | * Copyright (C) 1996 David S. Miller (dm@engr.sgi.com) | ||
3 | * Copyright (C) 1997, 2001 Ralf Baechle (ralf@gnu.org) | ||
4 | * Copyright (C) 2000 SiByte, Inc. | ||
5 | * Copyright (C) 2005 Thiemo Seufer | ||
6 | * | ||
7 | * Written by Justin Carlson of SiByte, Inc. | ||
8 | * and Kip Walker of Broadcom Corp. | ||
9 | * | ||
10 | * | ||
11 | * This program is free software; you can redistribute it and/or | ||
12 | * modify it under the terms of the GNU General Public License | ||
13 | * as published by the Free Software Foundation; either version 2 | ||
14 | * of the License, or (at your option) any later version. | ||
15 | * | ||
16 | * This program is distributed in the hope that it will be useful, | ||
17 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | ||
18 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | ||
19 | * GNU General Public License for more details. | ||
20 | * | ||
21 | * You should have received a copy of the GNU General Public License | ||
22 | * along with this program; if not, write to the Free Software | ||
23 | * Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA. | ||
24 | */ | ||
25 | #include <linux/config.h> | ||
26 | #include <linux/module.h> | ||
27 | #include <linux/sched.h> | ||
28 | #include <linux/smp.h> | ||
29 | |||
30 | #include <asm/io.h> | ||
31 | #include <asm/sibyte/sb1250.h> | ||
32 | #include <asm/sibyte/sb1250_regs.h> | ||
33 | #include <asm/sibyte/sb1250_dma.h> | ||
34 | |||
35 | #ifdef CONFIG_SB1_PASS_1_WORKAROUNDS | ||
36 | #define SB1_PREF_LOAD_STREAMED_HINT "0" | ||
37 | #define SB1_PREF_STORE_STREAMED_HINT "1" | ||
38 | #else | ||
39 | #define SB1_PREF_LOAD_STREAMED_HINT "4" | ||
40 | #define SB1_PREF_STORE_STREAMED_HINT "5" | ||
41 | #endif | ||
42 | |||
43 | static inline void clear_page_cpu(void *page) | ||
44 | { | ||
45 | unsigned char *addr = (unsigned char *) page; | ||
46 | unsigned char *end = addr + PAGE_SIZE; | ||
47 | |||
48 | /* | ||
49 | * JDCXXX - This should be bottlenecked by the write buffer, but these | ||
50 | * things tend to be mildly unpredictable...should check this on the | ||
51 | * performance model | ||
52 | * | ||
53 | * We prefetch 4 lines ahead. We're also "cheating" slightly here... | ||
54 | * since we know we're on an SB1, we force the assembler to take | ||
55 | * 64-bit operands to speed things up | ||
56 | */ | ||
57 | __asm__ __volatile__( | ||
58 | " .set push \n" | ||
59 | " .set mips4 \n" | ||
60 | " .set noreorder \n" | ||
61 | #ifdef CONFIG_CPU_HAS_PREFETCH | ||
62 | " daddiu %0, %0, 128 \n" | ||
63 | " pref " SB1_PREF_STORE_STREAMED_HINT ", -128(%0) \n" /* Prefetch the first 4 lines */ | ||
64 | " pref " SB1_PREF_STORE_STREAMED_HINT ", -96(%0) \n" | ||
65 | " pref " SB1_PREF_STORE_STREAMED_HINT ", -64(%0) \n" | ||
66 | " pref " SB1_PREF_STORE_STREAMED_HINT ", -32(%0) \n" | ||
67 | "1: sd $0, -128(%0) \n" /* Throw out a cacheline of 0's */ | ||
68 | " sd $0, -120(%0) \n" | ||
69 | " sd $0, -112(%0) \n" | ||
70 | " sd $0, -104(%0) \n" | ||
71 | " daddiu %0, %0, 32 \n" | ||
72 | " bnel %0, %1, 1b \n" | ||
73 | " pref " SB1_PREF_STORE_STREAMED_HINT ", -32(%0) \n" | ||
74 | " daddiu %0, %0, -128 \n" | ||
75 | #endif | ||
76 | " sd $0, 0(%0) \n" /* Throw out a cacheline of 0's */ | ||
77 | "1: sd $0, 8(%0) \n" | ||
78 | " sd $0, 16(%0) \n" | ||
79 | " sd $0, 24(%0) \n" | ||
80 | " daddiu %0, %0, 32 \n" | ||
81 | " bnel %0, %1, 1b \n" | ||
82 | " sd $0, 0(%0) \n" | ||
83 | " .set pop \n" | ||
84 | : "+r" (addr) | ||
85 | : "r" (end) | ||
86 | : "memory"); | ||
87 | } | ||
88 | |||
89 | static inline void copy_page_cpu(void *to, void *from) | ||
90 | { | ||
91 | unsigned char *src = (unsigned char *)from; | ||
92 | unsigned char *dst = (unsigned char *)to; | ||
93 | unsigned char *end = src + PAGE_SIZE; | ||
94 | |||
95 | /* | ||
96 | * The pref's used here are using "streaming" hints, which cause the | ||
97 | * copied data to be kicked out of the cache sooner. A page copy often | ||
98 | * ends up copying a lot more data than is commonly used, so this seems | ||
99 | * to make sense in terms of reducing cache pollution, but I've no real | ||
100 | * performance data to back this up | ||
101 | */ | ||
102 | __asm__ __volatile__( | ||
103 | " .set push \n" | ||
104 | " .set mips4 \n" | ||
105 | " .set noreorder \n" | ||
106 | #ifdef CONFIG_CPU_HAS_PREFETCH | ||
107 | " daddiu %0, %0, 128 \n" | ||
108 | " daddiu %1, %1, 128 \n" | ||
109 | " pref " SB1_PREF_LOAD_STREAMED_HINT ", -128(%0)\n" /* Prefetch the first 4 lines */ | ||
110 | " pref " SB1_PREF_STORE_STREAMED_HINT ", -128(%1)\n" | ||
111 | " pref " SB1_PREF_LOAD_STREAMED_HINT ", -96(%0)\n" | ||
112 | " pref " SB1_PREF_STORE_STREAMED_HINT ", -96(%1)\n" | ||
113 | " pref " SB1_PREF_LOAD_STREAMED_HINT ", -64(%0)\n" | ||
114 | " pref " SB1_PREF_STORE_STREAMED_HINT ", -64(%1)\n" | ||
115 | " pref " SB1_PREF_LOAD_STREAMED_HINT ", -32(%0)\n" | ||
116 | "1: pref " SB1_PREF_STORE_STREAMED_HINT ", -32(%1)\n" | ||
117 | # ifdef CONFIG_MIPS64 | ||
118 | " ld $8, -128(%0) \n" /* Block copy a cacheline */ | ||
119 | " ld $9, -120(%0) \n" | ||
120 | " ld $10, -112(%0) \n" | ||
121 | " ld $11, -104(%0) \n" | ||
122 | " sd $8, -128(%1) \n" | ||
123 | " sd $9, -120(%1) \n" | ||
124 | " sd $10, -112(%1) \n" | ||
125 | " sd $11, -104(%1) \n" | ||
126 | # else | ||
127 | " lw $2, -128(%0) \n" /* Block copy a cacheline */ | ||
128 | " lw $3, -124(%0) \n" | ||
129 | " lw $6, -120(%0) \n" | ||
130 | " lw $7, -116(%0) \n" | ||
131 | " lw $8, -112(%0) \n" | ||
132 | " lw $9, -108(%0) \n" | ||
133 | " lw $10, -104(%0) \n" | ||
134 | " lw $11, -100(%0) \n" | ||
135 | " sw $2, -128(%1) \n" | ||
136 | " sw $3, -124(%1) \n" | ||
137 | " sw $6, -120(%1) \n" | ||
138 | " sw $7, -116(%1) \n" | ||
139 | " sw $8, -112(%1) \n" | ||
140 | " sw $9, -108(%1) \n" | ||
141 | " sw $10, -104(%1) \n" | ||
142 | " sw $11, -100(%1) \n" | ||
143 | # endif | ||
144 | " daddiu %0, %0, 32 \n" | ||
145 | " daddiu %1, %1, 32 \n" | ||
146 | " bnel %0, %2, 1b \n" | ||
147 | " pref " SB1_PREF_LOAD_STREAMED_HINT ", -32(%0)\n" | ||
148 | " daddiu %0, %0, -128 \n" | ||
149 | " daddiu %1, %1, -128 \n" | ||
150 | #endif | ||
151 | #ifdef CONFIG_MIPS64 | ||
152 | " ld $8, 0(%0) \n" /* Block copy a cacheline */ | ||
153 | "1: ld $9, 8(%0) \n" | ||
154 | " ld $10, 16(%0) \n" | ||
155 | " ld $11, 24(%0) \n" | ||
156 | " sd $8, 0(%1) \n" | ||
157 | " sd $9, 8(%1) \n" | ||
158 | " sd $10, 16(%1) \n" | ||
159 | " sd $11, 24(%1) \n" | ||
160 | #else | ||
161 | " lw $2, 0(%0) \n" /* Block copy a cacheline */ | ||
162 | "1: lw $3, 4(%0) \n" | ||
163 | " lw $6, 8(%0) \n" | ||
164 | " lw $7, 12(%0) \n" | ||
165 | " lw $8, 16(%0) \n" | ||
166 | " lw $9, 20(%0) \n" | ||
167 | " lw $10, 24(%0) \n" | ||
168 | " lw $11, 28(%0) \n" | ||
169 | " sw $2, 0(%1) \n" | ||
170 | " sw $3, 4(%1) \n" | ||
171 | " sw $6, 8(%1) \n" | ||
172 | " sw $7, 12(%1) \n" | ||
173 | " sw $8, 16(%1) \n" | ||
174 | " sw $9, 20(%1) \n" | ||
175 | " sw $10, 24(%1) \n" | ||
176 | " sw $11, 28(%1) \n" | ||
177 | #endif | ||
178 | " daddiu %0, %0, 32 \n" | ||
179 | " daddiu %1, %1, 32 \n" | ||
180 | " bnel %0, %2, 1b \n" | ||
181 | #ifdef CONFIG_MIPS64 | ||
182 | " ld $8, 0(%0) \n" | ||
183 | #else | ||
184 | " lw $2, 0(%0) \n" | ||
185 | #endif | ||
186 | " .set pop \n" | ||
187 | : "+r" (src), "+r" (dst) | ||
188 | : "r" (end) | ||
189 | #ifdef CONFIG_MIPS64 | ||
190 | : "$8","$9","$10","$11","memory"); | ||
191 | #else | ||
192 | : "$2","$3","$6","$7","$8","$9","$10","$11","memory"); | ||
193 | #endif | ||
194 | } | ||
195 | |||
196 | |||
197 | #ifdef CONFIG_SIBYTE_DMA_PAGEOPS | ||
198 | |||
199 | /* | ||
200 | * Pad descriptors to cacheline, since each is exclusively owned by a | ||
201 | * particular CPU. | ||
202 | */ | ||
203 | typedef struct dmadscr_s { | ||
204 | u64 dscr_a; | ||
205 | u64 dscr_b; | ||
206 | u64 pad_a; | ||
207 | u64 pad_b; | ||
208 | } dmadscr_t; | ||
209 | |||
210 | static dmadscr_t page_descr[NR_CPUS] __attribute__((aligned(SMP_CACHE_BYTES))); | ||
211 | |||
212 | void sb1_dma_init(void) | ||
213 | { | ||
214 | int cpu = smp_processor_id(); | ||
215 | u64 base_val = CPHYSADDR(&page_descr[cpu]) | V_DM_DSCR_BASE_RINGSZ(1); | ||
216 | |||
217 | bus_writeq(base_val, | ||
218 | (void *)IOADDR(A_DM_REGISTER(cpu, R_DM_DSCR_BASE))); | ||
219 | bus_writeq(base_val | M_DM_DSCR_BASE_RESET, | ||
220 | (void *)IOADDR(A_DM_REGISTER(cpu, R_DM_DSCR_BASE))); | ||
221 | bus_writeq(base_val | M_DM_DSCR_BASE_ENABL, | ||
222 | (void *)IOADDR(A_DM_REGISTER(cpu, R_DM_DSCR_BASE))); | ||
223 | } | ||
224 | |||
225 | void clear_page(void *page) | ||
226 | { | ||
227 | int cpu = smp_processor_id(); | ||
228 | |||
229 | /* if the page is above Kseg0, use old way */ | ||
230 | if ((long)KSEGX(page) != (long)CKSEG0) | ||
231 | return clear_page_cpu(page); | ||
232 | |||
233 | page_descr[cpu].dscr_a = CPHYSADDR(page) | M_DM_DSCRA_ZERO_MEM | M_DM_DSCRA_L2C_DEST | M_DM_DSCRA_INTERRUPT; | ||
234 | page_descr[cpu].dscr_b = V_DM_DSCRB_SRC_LENGTH(PAGE_SIZE); | ||
235 | bus_writeq(1, (void *)IOADDR(A_DM_REGISTER(cpu, R_DM_DSCR_COUNT))); | ||
236 | |||
237 | /* | ||
238 | * Don't really want to do it this way, but there's no | ||
239 | * reliable way to delay completion detection. | ||
240 | */ | ||
241 | while (!(bus_readq((void *)(IOADDR(A_DM_REGISTER(cpu, R_DM_DSCR_BASE_DEBUG)) & | ||
242 | M_DM_DSCR_BASE_INTERRUPT)))) | ||
243 | ; | ||
244 | bus_readq((void *)IOADDR(A_DM_REGISTER(cpu, R_DM_DSCR_BASE))); | ||
245 | } | ||
246 | |||
247 | void copy_page(void *to, void *from) | ||
248 | { | ||
249 | unsigned long from_phys = CPHYSADDR(from); | ||
250 | unsigned long to_phys = CPHYSADDR(to); | ||
251 | int cpu = smp_processor_id(); | ||
252 | |||
253 | /* if either page is above Kseg0, use old way */ | ||
254 | if ((long)KSEGX(to) != (long)CKSEG0 | ||
255 | || (long)KSEGX(from) != (long)CKSEG0) | ||
256 | return copy_page_cpu(to, from); | ||
257 | |||
258 | page_descr[cpu].dscr_a = CPHYSADDR(to_phys) | M_DM_DSCRA_L2C_DEST | M_DM_DSCRA_INTERRUPT; | ||
259 | page_descr[cpu].dscr_b = CPHYSADDR(from_phys) | V_DM_DSCRB_SRC_LENGTH(PAGE_SIZE); | ||
260 | bus_writeq(1, (void *)IOADDR(A_DM_REGISTER(cpu, R_DM_DSCR_COUNT))); | ||
261 | |||
262 | /* | ||
263 | * Don't really want to do it this way, but there's no | ||
264 | * reliable way to delay completion detection. | ||
265 | */ | ||
266 | while (!(bus_readq((void *)(IOADDR(A_DM_REGISTER(cpu, R_DM_DSCR_BASE_DEBUG)) & | ||
267 | M_DM_DSCR_BASE_INTERRUPT)))) | ||
268 | ; | ||
269 | bus_readq((void *)IOADDR(A_DM_REGISTER(cpu, R_DM_DSCR_BASE))); | ||
270 | } | ||
271 | |||
272 | #else /* !CONFIG_SIBYTE_DMA_PAGEOPS */ | ||
273 | |||
274 | void clear_page(void *page) | ||
275 | { | ||
276 | return clear_page_cpu(page); | ||
277 | } | ||
278 | |||
279 | void copy_page(void *to, void *from) | ||
280 | { | ||
281 | return copy_page_cpu(to, from); | ||
282 | } | ||
283 | |||
284 | #endif /* !CONFIG_SIBYTE_DMA_PAGEOPS */ | ||
285 | |||
286 | EXPORT_SYMBOL(clear_page); | ||
287 | EXPORT_SYMBOL(copy_page); | ||