blob: 162bb155e695bb5fea4e52df86d5b48ac9b9255e [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
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
43static 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
89static 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"
Ralf Baechle875d43e2005-09-03 15:56:16 -0700117# ifdef CONFIG_64BIT
Linus Torvalds1da177e2005-04-16 15:20:36 -0700118 " 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
Ralf Baechle875d43e2005-09-03 15:56:16 -0700151#ifdef CONFIG_64BIT
Linus Torvalds1da177e2005-04-16 15:20:36 -0700152 " 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"
Ralf Baechle875d43e2005-09-03 15:56:16 -0700181#ifdef CONFIG_64BIT
Linus Torvalds1da177e2005-04-16 15:20:36 -0700182 " 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)
Ralf Baechle875d43e2005-09-03 15:56:16 -0700189#ifdef CONFIG_64BIT
Linus Torvalds1da177e2005-04-16 15:20:36 -0700190 : "$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
Ralf Baechle42a3b4f2005-09-03 15:56:17 -0700201 * particular CPU.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700202 */
203typedef struct dmadscr_s {
204 u64 dscr_a;
205 u64 dscr_b;
206 u64 pad_a;
207 u64 pad_b;
208} dmadscr_t;
209
Thiemo Seufer685f7792005-02-25 13:11:18 +0000210static dmadscr_t page_descr[DM_NUM_CHANNELS] __attribute__((aligned(SMP_CACHE_BYTES)));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700211
212void sb1_dma_init(void)
213{
Thiemo Seufer685f7792005-02-25 13:11:18 +0000214 int i;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700215
Thiemo Seufer685f7792005-02-25 13:11:18 +0000216 for (i = 0; i < DM_NUM_CHANNELS; i++) {
217 u64 base_val = (u64)CPHYSADDR(&page_descr[i]) | V_DM_DSCR_BASE_RINGSZ(1);
218 void *base_reg = (void *)IOADDR(A_DM_REGISTER(i, R_DM_DSCR_BASE));
219
220 __raw_writeq(base_val, base_reg);
221 __raw_writeq(base_val | M_DM_DSCR_BASE_RESET, base_reg);
222 __raw_writeq(base_val | M_DM_DSCR_BASE_ENABL, base_reg);
223 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700224}
225
226void clear_page(void *page)
227{
Thiemo Seufer685f7792005-02-25 13:11:18 +0000228 u64 to_phys = (u64)CPHYSADDR(page);
229 unsigned int cpu = smp_processor_id();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700230
Thiemo Seufer685f7792005-02-25 13:11:18 +0000231 /* if the page is not in KSEG0, use old way */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700232 if ((long)KSEGX(page) != (long)CKSEG0)
233 return clear_page_cpu(page);
234
Thiemo Seufer685f7792005-02-25 13:11:18 +0000235 page_descr[cpu].dscr_a = to_phys | M_DM_DSCRA_ZERO_MEM | M_DM_DSCRA_L2C_DEST | M_DM_DSCRA_INTERRUPT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700236 page_descr[cpu].dscr_b = V_DM_DSCRB_SRC_LENGTH(PAGE_SIZE);
Maciej W. Rozycki65bda1a2005-02-22 21:51:30 +0000237 __raw_writeq(1, IOADDR(A_DM_REGISTER(cpu, R_DM_DSCR_COUNT)));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700238
239 /*
240 * Don't really want to do it this way, but there's no
241 * reliable way to delay completion detection.
242 */
Maciej W. Rozycki65bda1a2005-02-22 21:51:30 +0000243 while (!(__raw_readq(IOADDR(A_DM_REGISTER(cpu, R_DM_DSCR_BASE_DEBUG)))
Thiemo Seufer685f7792005-02-25 13:11:18 +0000244 & M_DM_DSCR_BASE_INTERRUPT))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700245 ;
Maciej W. Rozycki65bda1a2005-02-22 21:51:30 +0000246 __raw_readq(IOADDR(A_DM_REGISTER(cpu, R_DM_DSCR_BASE)));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700247}
248
249void copy_page(void *to, void *from)
250{
Thiemo Seufer685f7792005-02-25 13:11:18 +0000251 u64 from_phys = (u64)CPHYSADDR(from);
252 u64 to_phys = (u64)CPHYSADDR(to);
253 unsigned int cpu = smp_processor_id();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700254
Thiemo Seufer685f7792005-02-25 13:11:18 +0000255 /* if any page is not in KSEG0, use old way */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700256 if ((long)KSEGX(to) != (long)CKSEG0
257 || (long)KSEGX(from) != (long)CKSEG0)
258 return copy_page_cpu(to, from);
259
Thiemo Seufer685f7792005-02-25 13:11:18 +0000260 page_descr[cpu].dscr_a = to_phys | M_DM_DSCRA_L2C_DEST | M_DM_DSCRA_INTERRUPT;
261 page_descr[cpu].dscr_b = from_phys | V_DM_DSCRB_SRC_LENGTH(PAGE_SIZE);
262 __raw_writeq(1, (void *)IOADDR(A_DM_REGISTER(cpu, R_DM_DSCR_COUNT)));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700263
264 /*
265 * Don't really want to do it this way, but there's no
266 * reliable way to delay completion detection.
267 */
Thiemo Seufer685f7792005-02-25 13:11:18 +0000268 while (!(__raw_readq((void *)IOADDR(A_DM_REGISTER(cpu, R_DM_DSCR_BASE_DEBUG)))
269 & M_DM_DSCR_BASE_INTERRUPT))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700270 ;
Maciej W. Rozycki65bda1a2005-02-22 21:51:30 +0000271 __raw_readq(IOADDR(A_DM_REGISTER(cpu, R_DM_DSCR_BASE)));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700272}
273
274#else /* !CONFIG_SIBYTE_DMA_PAGEOPS */
275
276void clear_page(void *page)
277{
278 return clear_page_cpu(page);
279}
280
281void copy_page(void *to, void *from)
282{
283 return copy_page_cpu(to, from);
284}
285
286#endif /* !CONFIG_SIBYTE_DMA_PAGEOPS */
287
288EXPORT_SYMBOL(clear_page);
289EXPORT_SYMBOL(copy_page);