diff options
author | Thiemo Seufer <ths@networkno.de> | 2008-02-18 14:32:49 -0500 |
---|---|---|
committer | Ralf Baechle <ralf@linux-mips.org> | 2008-04-28 12:14:23 -0400 |
commit | fb2a27e743cd565c25cd896911e494482a8b7251 (patch) | |
tree | 791190ead6211e829443185fb1d4b4109fbd9a34 | |
parent | 064922a805ec7aadfafdd27aa6b4908d737c3c1d (diff) |
[MIPS] Reimplement clear_page/copy_page
Fold the SB-1 specific implementation of clear_page/copy_page in the
generic version, and rewrite that one in tlbex style. The immediate
benefits:
- It converts the compile-time workaround for SB-1 pass 1 prefetches
to a more efficient run-time check.
- It allows adjustment of loop unfolling, which helps to reduce the
number of redundant cdex cache ops.
- It fixes some esoteric cornercases (the cache line length calculations
can go wrong, and support for 64k pages without prefetch instructions
will overflow the addiu immediate).
- Somewhat better guesses of "good" prefetch values.
Signed-off-by: Thiemo Seufer <ths@networkno.de>
Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
-rw-r--r-- | arch/mips/mm/Makefile | 37 | ||||
-rw-r--r-- | arch/mips/mm/page.c | 684 | ||||
-rw-r--r-- | arch/mips/mm/pg-r4k.c | 534 | ||||
-rw-r--r-- | arch/mips/mm/pg-sb1.c | 302 | ||||
-rw-r--r-- | arch/mips/mm/uasm.c | 26 | ||||
-rw-r--r-- | arch/mips/mm/uasm.h | 4 |
6 files changed, 725 insertions, 862 deletions
diff --git a/arch/mips/mm/Makefile b/arch/mips/mm/Makefile index c6f832e0f41c..48731020ca0e 100644 --- a/arch/mips/mm/Makefile +++ b/arch/mips/mm/Makefile | |||
@@ -4,30 +4,29 @@ | |||
4 | 4 | ||
5 | obj-y += cache.o dma-default.o extable.o fault.o \ | 5 | obj-y += cache.o dma-default.o extable.o fault.o \ |
6 | init.o pgtable.o tlbex.o tlbex-fault.o \ | 6 | init.o pgtable.o tlbex.o tlbex-fault.o \ |
7 | uasm.o | 7 | uasm.o page.o |
8 | 8 | ||
9 | obj-$(CONFIG_32BIT) += ioremap.o pgtable-32.o | 9 | obj-$(CONFIG_32BIT) += ioremap.o pgtable-32.o |
10 | obj-$(CONFIG_64BIT) += pgtable-64.o | 10 | obj-$(CONFIG_64BIT) += pgtable-64.o |
11 | obj-$(CONFIG_HIGHMEM) += highmem.o | 11 | obj-$(CONFIG_HIGHMEM) += highmem.o |
12 | 12 | ||
13 | obj-$(CONFIG_CPU_LOONGSON2) += c-r4k.o cex-gen.o pg-r4k.o tlb-r4k.o | 13 | obj-$(CONFIG_CPU_LOONGSON2) += c-r4k.o cex-gen.o tlb-r4k.o |
14 | obj-$(CONFIG_CPU_MIPS32) += c-r4k.o cex-gen.o pg-r4k.o tlb-r4k.o | 14 | obj-$(CONFIG_CPU_MIPS32) += c-r4k.o cex-gen.o tlb-r4k.o |
15 | obj-$(CONFIG_CPU_MIPS64) += c-r4k.o cex-gen.o pg-r4k.o tlb-r4k.o | 15 | obj-$(CONFIG_CPU_MIPS64) += c-r4k.o cex-gen.o tlb-r4k.o |
16 | obj-$(CONFIG_CPU_NEVADA) += c-r4k.o cex-gen.o pg-r4k.o tlb-r4k.o | 16 | obj-$(CONFIG_CPU_NEVADA) += c-r4k.o cex-gen.o tlb-r4k.o |
17 | obj-$(CONFIG_CPU_R10000) += c-r4k.o cex-gen.o pg-r4k.o tlb-r4k.o | 17 | obj-$(CONFIG_CPU_R10000) += c-r4k.o cex-gen.o tlb-r4k.o |
18 | obj-$(CONFIG_CPU_R3000) += c-r3k.o tlb-r3k.o pg-r4k.o | 18 | obj-$(CONFIG_CPU_R3000) += c-r3k.o tlb-r3k.o |
19 | obj-$(CONFIG_CPU_R4300) += c-r4k.o cex-gen.o pg-r4k.o tlb-r4k.o | 19 | obj-$(CONFIG_CPU_R4300) += c-r4k.o cex-gen.o tlb-r4k.o |
20 | obj-$(CONFIG_CPU_R4X00) += c-r4k.o cex-gen.o pg-r4k.o tlb-r4k.o | 20 | obj-$(CONFIG_CPU_R4X00) += c-r4k.o cex-gen.o tlb-r4k.o |
21 | obj-$(CONFIG_CPU_R5000) += c-r4k.o cex-gen.o pg-r4k.o tlb-r4k.o | 21 | obj-$(CONFIG_CPU_R5000) += c-r4k.o cex-gen.o tlb-r4k.o |
22 | obj-$(CONFIG_CPU_R5432) += c-r4k.o cex-gen.o pg-r4k.o tlb-r4k.o | 22 | obj-$(CONFIG_CPU_R5432) += c-r4k.o cex-gen.o tlb-r4k.o |
23 | obj-$(CONFIG_CPU_R8000) += c-r4k.o cex-gen.o pg-r4k.o tlb-r8k.o | 23 | obj-$(CONFIG_CPU_R8000) += c-r4k.o cex-gen.o tlb-r8k.o |
24 | obj-$(CONFIG_CPU_RM7000) += c-r4k.o cex-gen.o pg-r4k.o tlb-r4k.o | 24 | obj-$(CONFIG_CPU_RM7000) += c-r4k.o cex-gen.o tlb-r4k.o |
25 | obj-$(CONFIG_CPU_RM9000) += c-r4k.o cex-gen.o pg-r4k.o tlb-r4k.o | 25 | obj-$(CONFIG_CPU_RM9000) += c-r4k.o cex-gen.o tlb-r4k.o |
26 | obj-$(CONFIG_CPU_SB1) += c-r4k.o cerr-sb1.o cex-sb1.o pg-sb1.o \ | 26 | obj-$(CONFIG_CPU_SB1) += c-r4k.o cerr-sb1.o cex-sb1.o tlb-r4k.o |
27 | tlb-r4k.o | 27 | obj-$(CONFIG_CPU_TX39XX) += c-tx39.o tlb-r3k.o |
28 | obj-$(CONFIG_CPU_TX39XX) += c-tx39.o pg-r4k.o tlb-r3k.o | 28 | obj-$(CONFIG_CPU_TX49XX) += c-r4k.o cex-gen.o tlb-r4k.o |
29 | obj-$(CONFIG_CPU_TX49XX) += c-r4k.o cex-gen.o pg-r4k.o tlb-r4k.o | 29 | obj-$(CONFIG_CPU_VR41XX) += c-r4k.o cex-gen.o tlb-r4k.o |
30 | obj-$(CONFIG_CPU_VR41XX) += c-r4k.o cex-gen.o pg-r4k.o tlb-r4k.o | ||
31 | 30 | ||
32 | obj-$(CONFIG_IP22_CPU_SCACHE) += sc-ip22.o | 31 | obj-$(CONFIG_IP22_CPU_SCACHE) += sc-ip22.o |
33 | obj-$(CONFIG_R5000_CPU_SCACHE) += sc-r5k.o | 32 | obj-$(CONFIG_R5000_CPU_SCACHE) += sc-r5k.o |
diff --git a/arch/mips/mm/page.c b/arch/mips/mm/page.c new file mode 100644 index 000000000000..d827d6144369 --- /dev/null +++ b/arch/mips/mm/page.c | |||
@@ -0,0 +1,684 @@ | |||
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 | * Copyright (C) 2003, 04, 05 Ralf Baechle (ralf@linux-mips.org) | ||
7 | * Copyright (C) 2007 Maciej W. Rozycki | ||
8 | * Copyright (C) 2008 Thiemo Seufer | ||
9 | */ | ||
10 | #include <linux/init.h> | ||
11 | #include <linux/kernel.h> | ||
12 | #include <linux/sched.h> | ||
13 | #include <linux/mm.h> | ||
14 | #include <linux/module.h> | ||
15 | #include <linux/proc_fs.h> | ||
16 | |||
17 | #include <asm/bugs.h> | ||
18 | #include <asm/cacheops.h> | ||
19 | #include <asm/inst.h> | ||
20 | #include <asm/io.h> | ||
21 | #include <asm/page.h> | ||
22 | #include <asm/pgtable.h> | ||
23 | #include <asm/prefetch.h> | ||
24 | #include <asm/system.h> | ||
25 | #include <asm/bootinfo.h> | ||
26 | #include <asm/mipsregs.h> | ||
27 | #include <asm/mmu_context.h> | ||
28 | #include <asm/cpu.h> | ||
29 | #include <asm/war.h> | ||
30 | |||
31 | #ifdef CONFIG_SIBYTE_DMA_PAGEOPS | ||
32 | #include <asm/sibyte/sb1250.h> | ||
33 | #include <asm/sibyte/sb1250_regs.h> | ||
34 | #include <asm/sibyte/sb1250_dma.h> | ||
35 | #endif | ||
36 | |||
37 | #include "uasm.h" | ||
38 | |||
39 | /* Registers used in the assembled routines. */ | ||
40 | #define ZERO 0 | ||
41 | #define AT 2 | ||
42 | #define A0 4 | ||
43 | #define A1 5 | ||
44 | #define A2 6 | ||
45 | #define T0 8 | ||
46 | #define T1 9 | ||
47 | #define T2 10 | ||
48 | #define T3 11 | ||
49 | #define T9 25 | ||
50 | #define RA 31 | ||
51 | |||
52 | /* Handle labels (which must be positive integers). */ | ||
53 | enum label_id { | ||
54 | label_clear_nopref = 1, | ||
55 | label_clear_pref, | ||
56 | label_copy_nopref, | ||
57 | label_copy_pref_both, | ||
58 | label_copy_pref_store, | ||
59 | }; | ||
60 | |||
61 | UASM_L_LA(_clear_nopref) | ||
62 | UASM_L_LA(_clear_pref) | ||
63 | UASM_L_LA(_copy_nopref) | ||
64 | UASM_L_LA(_copy_pref_both) | ||
65 | UASM_L_LA(_copy_pref_store) | ||
66 | |||
67 | /* We need one branch and therefore one relocation per target label. */ | ||
68 | static struct uasm_label __cpuinitdata labels[5]; | ||
69 | static struct uasm_reloc __cpuinitdata relocs[5]; | ||
70 | |||
71 | #define cpu_is_r4600_v1_x() ((read_c0_prid() & 0xfffffff0) == 0x00002010) | ||
72 | #define cpu_is_r4600_v2_x() ((read_c0_prid() & 0xfffffff0) == 0x00002020) | ||
73 | |||
74 | /* | ||
75 | * Maximum sizes: | ||
76 | * | ||
77 | * R4000 128 bytes S-cache: 0x058 bytes | ||
78 | * R4600 v1.7: 0x05c bytes | ||
79 | * R4600 v2.0: 0x060 bytes | ||
80 | * With prefetching, 16 word strides 0x120 bytes | ||
81 | */ | ||
82 | |||
83 | static u32 clear_page_array[0x120 / 4]; | ||
84 | |||
85 | #ifdef CONFIG_SIBYTE_DMA_PAGEOPS | ||
86 | void clear_page_cpu(void *page) __attribute__((alias("clear_page_array"))); | ||
87 | #else | ||
88 | void clear_page(void *page) __attribute__((alias("clear_page_array"))); | ||
89 | #endif | ||
90 | |||
91 | EXPORT_SYMBOL(clear_page); | ||
92 | |||
93 | /* | ||
94 | * Maximum sizes: | ||
95 | * | ||
96 | * R4000 128 bytes S-cache: 0x11c bytes | ||
97 | * R4600 v1.7: 0x080 bytes | ||
98 | * R4600 v2.0: 0x07c bytes | ||
99 | * With prefetching, 16 word strides 0x540 bytes | ||
100 | */ | ||
101 | static u32 copy_page_array[0x540 / 4]; | ||
102 | |||
103 | #ifdef CONFIG_SIBYTE_DMA_PAGEOPS | ||
104 | void | ||
105 | copy_page_cpu(void *to, void *from) __attribute__((alias("copy_page_array"))); | ||
106 | #else | ||
107 | void copy_page(void *to, void *from) __attribute__((alias("copy_page_array"))); | ||
108 | #endif | ||
109 | |||
110 | EXPORT_SYMBOL(copy_page); | ||
111 | |||
112 | |||
113 | static int pref_bias_clear_store __cpuinitdata; | ||
114 | static int pref_bias_copy_load __cpuinitdata; | ||
115 | static int pref_bias_copy_store __cpuinitdata; | ||
116 | |||
117 | static u32 pref_src_mode __cpuinitdata; | ||
118 | static u32 pref_dst_mode __cpuinitdata; | ||
119 | |||
120 | static int clear_word_size __cpuinitdata; | ||
121 | static int copy_word_size __cpuinitdata; | ||
122 | |||
123 | static int half_clear_loop_size __cpuinitdata; | ||
124 | static int half_copy_loop_size __cpuinitdata; | ||
125 | |||
126 | static int cache_line_size __cpuinitdata; | ||
127 | #define cache_line_mask() (cache_line_size - 1) | ||
128 | |||
129 | static inline void __cpuinit | ||
130 | pg_addiu(u32 **buf, unsigned int reg1, unsigned int reg2, unsigned int off) | ||
131 | { | ||
132 | if (cpu_has_64bit_gp_regs && DADDI_WAR && r4k_daddiu_bug()) { | ||
133 | if (off > 0x7fff) { | ||
134 | uasm_i_lui(buf, T9, uasm_rel_hi(off)); | ||
135 | uasm_i_addiu(buf, T9, T9, uasm_rel_lo(off)); | ||
136 | } else | ||
137 | uasm_i_addiu(buf, T9, ZERO, off); | ||
138 | uasm_i_daddu(buf, reg1, reg2, T9); | ||
139 | } else { | ||
140 | if (off > 0x7fff) { | ||
141 | uasm_i_lui(buf, T9, uasm_rel_hi(off)); | ||
142 | uasm_i_addiu(buf, T9, T9, uasm_rel_lo(off)); | ||
143 | UASM_i_ADDU(buf, reg1, reg2, T9); | ||
144 | } else | ||
145 | UASM_i_ADDIU(buf, reg1, reg2, off); | ||
146 | } | ||
147 | } | ||
148 | |||
149 | static void __cpuinit set_prefetch_parameters(void) | ||
150 | { | ||
151 | if (cpu_has_64bit_gp_regs || cpu_has_64bit_zero_reg) | ||
152 | clear_word_size = 8; | ||
153 | else | ||
154 | clear_word_size = 4; | ||
155 | |||
156 | if (cpu_has_64bit_gp_regs) | ||
157 | copy_word_size = 8; | ||
158 | else | ||
159 | copy_word_size = 4; | ||
160 | |||
161 | /* | ||
162 | * The pref's used here are using "streaming" hints, which cause the | ||
163 | * copied data to be kicked out of the cache sooner. A page copy often | ||
164 | * ends up copying a lot more data than is commonly used, so this seems | ||
165 | * to make sense in terms of reducing cache pollution, but I've no real | ||
166 | * performance data to back this up. | ||
167 | */ | ||
168 | if (cpu_has_prefetch) { | ||
169 | /* | ||
170 | * XXX: Most prefetch bias values in here are based on | ||
171 | * guesswork. | ||
172 | */ | ||
173 | cache_line_size = cpu_dcache_line_size(); | ||
174 | switch (current_cpu_type()) { | ||
175 | case CPU_TX49XX: | ||
176 | /* TX49 supports only Pref_Load */ | ||
177 | pref_bias_copy_load = 256; | ||
178 | break; | ||
179 | |||
180 | case CPU_RM9000: | ||
181 | /* | ||
182 | * As a workaround for erratum G105 which make the | ||
183 | * PrepareForStore hint unusable we fall back to | ||
184 | * StoreRetained on the RM9000. Once it is known which | ||
185 | * versions of the RM9000 we'll be able to condition- | ||
186 | * alize this. | ||
187 | */ | ||
188 | |||
189 | case CPU_R10000: | ||
190 | case CPU_R12000: | ||
191 | case CPU_R14000: | ||
192 | /* | ||
193 | * Those values have been experimentally tuned for an | ||
194 | * Origin 200. | ||
195 | */ | ||
196 | pref_bias_clear_store = 512; | ||
197 | pref_bias_copy_load = 256; | ||
198 | pref_bias_copy_store = 256; | ||
199 | pref_src_mode = Pref_LoadStreamed; | ||
200 | pref_dst_mode = Pref_StoreStreamed; | ||
201 | break; | ||
202 | |||
203 | case CPU_SB1: | ||
204 | case CPU_SB1A: | ||
205 | pref_bias_clear_store = 128; | ||
206 | pref_bias_copy_load = 128; | ||
207 | pref_bias_copy_store = 128; | ||
208 | /* | ||
209 | * SB1 pass1 Pref_LoadStreamed/Pref_StoreStreamed | ||
210 | * hints are broken. | ||
211 | */ | ||
212 | if (current_cpu_type() == CPU_SB1 && | ||
213 | (current_cpu_data.processor_id & 0xff) < 0x02) { | ||
214 | pref_src_mode = Pref_Load; | ||
215 | pref_dst_mode = Pref_Store; | ||
216 | } else { | ||
217 | pref_src_mode = Pref_LoadStreamed; | ||
218 | pref_dst_mode = Pref_StoreStreamed; | ||
219 | } | ||
220 | break; | ||
221 | |||
222 | default: | ||
223 | pref_bias_clear_store = 128; | ||
224 | pref_bias_copy_load = 256; | ||
225 | pref_bias_copy_store = 128; | ||
226 | pref_src_mode = Pref_LoadStreamed; | ||
227 | pref_dst_mode = Pref_PrepareForStore; | ||
228 | break; | ||
229 | } | ||
230 | } else { | ||
231 | if (cpu_has_cache_cdex_s) | ||
232 | cache_line_size = cpu_scache_line_size(); | ||
233 | else if (cpu_has_cache_cdex_p) | ||
234 | cache_line_size = cpu_dcache_line_size(); | ||
235 | } | ||
236 | /* | ||
237 | * Too much unrolling will overflow the available space in | ||
238 | * clear_space_array / copy_page_array. 8 words sounds generous, | ||
239 | * but a R4000 with 128 byte L2 line length can exceed even that. | ||
240 | */ | ||
241 | half_clear_loop_size = min(8 * clear_word_size, | ||
242 | max(cache_line_size >> 1, | ||
243 | 4 * clear_word_size)); | ||
244 | half_copy_loop_size = min(8 * copy_word_size, | ||
245 | max(cache_line_size >> 1, | ||
246 | 4 * copy_word_size)); | ||
247 | } | ||
248 | |||
249 | static void __cpuinit build_clear_store(u32 **buf, int off) | ||
250 | { | ||
251 | if (cpu_has_64bit_gp_regs || cpu_has_64bit_zero_reg) { | ||
252 | uasm_i_sd(buf, ZERO, off, A0); | ||
253 | } else { | ||
254 | uasm_i_sw(buf, ZERO, off, A0); | ||
255 | } | ||
256 | } | ||
257 | |||
258 | static inline void __cpuinit build_clear_pref(u32 **buf, int off) | ||
259 | { | ||
260 | if (off & cache_line_mask()) | ||
261 | return; | ||
262 | |||
263 | if (pref_bias_clear_store) { | ||
264 | uasm_i_pref(buf, pref_dst_mode, pref_bias_clear_store + off, | ||
265 | A0); | ||
266 | } else if (cpu_has_cache_cdex_s) { | ||
267 | uasm_i_cache(buf, Create_Dirty_Excl_SD, off, A0); | ||
268 | } else if (cpu_has_cache_cdex_p) { | ||
269 | if (R4600_V1_HIT_CACHEOP_WAR && cpu_is_r4600_v1_x()) { | ||
270 | uasm_i_nop(buf); | ||
271 | uasm_i_nop(buf); | ||
272 | uasm_i_nop(buf); | ||
273 | uasm_i_nop(buf); | ||
274 | } | ||
275 | |||
276 | if (R4600_V2_HIT_CACHEOP_WAR && cpu_is_r4600_v2_x()) | ||
277 | uasm_i_lw(buf, ZERO, ZERO, AT); | ||
278 | |||
279 | uasm_i_cache(buf, Create_Dirty_Excl_D, off, A0); | ||
280 | } | ||
281 | } | ||
282 | |||
283 | void __cpuinit build_clear_page(void) | ||
284 | { | ||
285 | int off; | ||
286 | u32 *buf = (u32 *)&clear_page_array; | ||
287 | struct uasm_label *l = labels; | ||
288 | struct uasm_reloc *r = relocs; | ||
289 | int i; | ||
290 | |||
291 | memset(labels, 0, sizeof(labels)); | ||
292 | memset(relocs, 0, sizeof(relocs)); | ||
293 | |||
294 | set_prefetch_parameters(); | ||
295 | |||
296 | /* | ||
297 | * This algorithm makes the following assumptions: | ||
298 | * - The prefetch bias is a multiple of 2 words. | ||
299 | * - The prefetch bias is less than one page. | ||
300 | */ | ||
301 | BUG_ON(pref_bias_clear_store % (2 * clear_word_size)); | ||
302 | BUG_ON(PAGE_SIZE < pref_bias_clear_store); | ||
303 | |||
304 | off = PAGE_SIZE - pref_bias_clear_store; | ||
305 | if (off > 0xffff || !pref_bias_clear_store) | ||
306 | pg_addiu(&buf, A2, A0, off); | ||
307 | else | ||
308 | uasm_i_ori(&buf, A2, A0, off); | ||
309 | |||
310 | if (R4600_V2_HIT_CACHEOP_WAR && cpu_is_r4600_v2_x()) | ||
311 | uasm_i_lui(&buf, AT, 0xa000); | ||
312 | |||
313 | off = min(8, pref_bias_clear_store / cache_line_size) * | ||
314 | cache_line_size; | ||
315 | while (off) { | ||
316 | build_clear_pref(&buf, -off); | ||
317 | off -= cache_line_size; | ||
318 | } | ||
319 | uasm_l_clear_pref(&l, buf); | ||
320 | do { | ||
321 | build_clear_pref(&buf, off); | ||
322 | build_clear_store(&buf, off); | ||
323 | off += clear_word_size; | ||
324 | } while (off < half_clear_loop_size); | ||
325 | pg_addiu(&buf, A0, A0, 2 * off); | ||
326 | off = -off; | ||
327 | do { | ||
328 | build_clear_pref(&buf, off); | ||
329 | if (off == -clear_word_size) | ||
330 | uasm_il_bne(&buf, &r, A0, A2, label_clear_pref); | ||
331 | build_clear_store(&buf, off); | ||
332 | off += clear_word_size; | ||
333 | } while (off < 0); | ||
334 | |||
335 | if (pref_bias_clear_store) { | ||
336 | pg_addiu(&buf, A2, A0, pref_bias_clear_store); | ||
337 | uasm_l_clear_nopref(&l, buf); | ||
338 | off = 0; | ||
339 | do { | ||
340 | build_clear_store(&buf, off); | ||
341 | off += clear_word_size; | ||
342 | } while (off < half_clear_loop_size); | ||
343 | pg_addiu(&buf, A0, A0, 2 * off); | ||
344 | off = -off; | ||
345 | do { | ||
346 | if (off == -clear_word_size) | ||
347 | uasm_il_bne(&buf, &r, A0, A2, | ||
348 | label_clear_nopref); | ||
349 | build_clear_store(&buf, off); | ||
350 | off += clear_word_size; | ||
351 | } while (off < 0); | ||
352 | } | ||
353 | |||
354 | uasm_i_jr(&buf, RA); | ||
355 | uasm_i_nop(&buf); | ||
356 | |||
357 | BUG_ON(buf > clear_page_array + ARRAY_SIZE(clear_page_array)); | ||
358 | |||
359 | uasm_resolve_relocs(relocs, labels); | ||
360 | |||
361 | pr_debug("Synthesized clear page handler (%u instructions).\n", | ||
362 | (u32)(buf - clear_page_array)); | ||
363 | |||
364 | pr_debug("\t.set push\n"); | ||
365 | pr_debug("\t.set noreorder\n"); | ||
366 | for (i = 0; i < (buf - clear_page_array); i++) | ||
367 | pr_debug("\t.word 0x%08x\n", clear_page_array[i]); | ||
368 | pr_debug("\t.set pop\n"); | ||
369 | } | ||
370 | |||
371 | static void __cpuinit build_copy_load(u32 **buf, int reg, int off) | ||
372 | { | ||
373 | if (cpu_has_64bit_gp_regs) { | ||
374 | uasm_i_ld(buf, reg, off, A1); | ||
375 | } else { | ||
376 | uasm_i_lw(buf, reg, off, A1); | ||
377 | } | ||
378 | } | ||
379 | |||
380 | static void __cpuinit build_copy_store(u32 **buf, int reg, int off) | ||
381 | { | ||
382 | if (cpu_has_64bit_gp_regs) { | ||
383 | uasm_i_sd(buf, reg, off, A0); | ||
384 | } else { | ||
385 | uasm_i_sw(buf, reg, off, A0); | ||
386 | } | ||
387 | } | ||
388 | |||
389 | static inline void build_copy_load_pref(u32 **buf, int off) | ||
390 | { | ||
391 | if (off & cache_line_mask()) | ||
392 | return; | ||
393 | |||
394 | if (pref_bias_copy_load) | ||
395 | uasm_i_pref(buf, pref_src_mode, pref_bias_copy_load + off, A1); | ||
396 | } | ||
397 | |||
398 | static inline void build_copy_store_pref(u32 **buf, int off) | ||
399 | { | ||
400 | if (off & cache_line_mask()) | ||
401 | return; | ||
402 | |||
403 | if (pref_bias_copy_store) { | ||
404 | uasm_i_pref(buf, pref_dst_mode, pref_bias_copy_store + off, | ||
405 | A0); | ||
406 | } else if (cpu_has_cache_cdex_s) { | ||
407 | uasm_i_cache(buf, Create_Dirty_Excl_SD, off, A0); | ||
408 | } else if (cpu_has_cache_cdex_p) { | ||
409 | if (R4600_V1_HIT_CACHEOP_WAR && cpu_is_r4600_v1_x()) { | ||
410 | uasm_i_nop(buf); | ||
411 | uasm_i_nop(buf); | ||
412 | uasm_i_nop(buf); | ||
413 | uasm_i_nop(buf); | ||
414 | } | ||
415 | |||
416 | if (R4600_V2_HIT_CACHEOP_WAR && cpu_is_r4600_v2_x()) | ||
417 | uasm_i_lw(buf, ZERO, ZERO, AT); | ||
418 | |||
419 | uasm_i_cache(buf, Create_Dirty_Excl_D, off, A0); | ||
420 | } | ||
421 | } | ||
422 | |||
423 | void __cpuinit build_copy_page(void) | ||
424 | { | ||
425 | int off; | ||
426 | u32 *buf = (u32 *)©_page_array; | ||
427 | struct uasm_label *l = labels; | ||
428 | struct uasm_reloc *r = relocs; | ||
429 | int i; | ||
430 | |||
431 | memset(labels, 0, sizeof(labels)); | ||
432 | memset(relocs, 0, sizeof(relocs)); | ||
433 | |||
434 | set_prefetch_parameters(); | ||
435 | |||
436 | /* | ||
437 | * This algorithm makes the following assumptions: | ||
438 | * - All prefetch biases are multiples of 8 words. | ||
439 | * - The prefetch biases are less than one page. | ||
440 | * - The store prefetch bias isn't greater than the load | ||
441 | * prefetch bias. | ||
442 | */ | ||
443 | BUG_ON(pref_bias_copy_load % (8 * copy_word_size)); | ||
444 | BUG_ON(pref_bias_copy_store % (8 * copy_word_size)); | ||
445 | BUG_ON(PAGE_SIZE < pref_bias_copy_load); | ||
446 | BUG_ON(pref_bias_copy_store > pref_bias_copy_load); | ||
447 | |||
448 | off = PAGE_SIZE - pref_bias_copy_load; | ||
449 | if (off > 0xffff || !pref_bias_copy_load) | ||
450 | pg_addiu(&buf, A2, A0, off); | ||
451 | else | ||
452 | uasm_i_ori(&buf, A2, A0, off); | ||
453 | |||
454 | if (R4600_V2_HIT_CACHEOP_WAR && cpu_is_r4600_v2_x()) | ||
455 | uasm_i_lui(&buf, AT, 0xa000); | ||
456 | |||
457 | off = min(8, pref_bias_copy_load / cache_line_size) * cache_line_size; | ||
458 | while (off) { | ||
459 | build_copy_load_pref(&buf, -off); | ||
460 | off -= cache_line_size; | ||
461 | } | ||
462 | off = min(8, pref_bias_copy_store / cache_line_size) * cache_line_size; | ||
463 | while (off) { | ||
464 | build_copy_store_pref(&buf, -off); | ||
465 | off -= cache_line_size; | ||
466 | } | ||
467 | uasm_l_copy_pref_both(&l, buf); | ||
468 | do { | ||
469 | build_copy_load_pref(&buf, off); | ||
470 | build_copy_load(&buf, T0, off); | ||
471 | build_copy_load_pref(&buf, off + copy_word_size); | ||
472 | build_copy_load(&buf, T1, off + copy_word_size); | ||
473 | build_copy_load_pref(&buf, off + 2 * copy_word_size); | ||
474 | build_copy_load(&buf, T2, off + 2 * copy_word_size); | ||
475 | build_copy_load_pref(&buf, off + 3 * copy_word_size); | ||
476 | build_copy_load(&buf, T3, off + 3 * copy_word_size); | ||
477 | build_copy_store_pref(&buf, off); | ||
478 | build_copy_store(&buf, T0, off); | ||
479 | build_copy_store_pref(&buf, off + copy_word_size); | ||
480 | build_copy_store(&buf, T1, off + copy_word_size); | ||
481 | build_copy_store_pref(&buf, off + 2 * copy_word_size); | ||
482 | build_copy_store(&buf, T2, off + 2 * copy_word_size); | ||
483 | build_copy_store_pref(&buf, off + 3 * copy_word_size); | ||
484 | build_copy_store(&buf, T3, off + 3 * copy_word_size); | ||
485 | off += 4 * copy_word_size; | ||
486 | } while (off < half_copy_loop_size); | ||
487 | pg_addiu(&buf, A1, A1, 2 * off); | ||
488 | pg_addiu(&buf, A0, A0, 2 * off); | ||
489 | off = -off; | ||
490 | do { | ||
491 | build_copy_load_pref(&buf, off); | ||
492 | build_copy_load(&buf, T0, off); | ||
493 | build_copy_load_pref(&buf, off + copy_word_size); | ||
494 | build_copy_load(&buf, T1, off + copy_word_size); | ||
495 | build_copy_load_pref(&buf, off + 2 * copy_word_size); | ||
496 | build_copy_load(&buf, T2, off + 2 * copy_word_size); | ||
497 | build_copy_load_pref(&buf, off + 3 * copy_word_size); | ||
498 | build_copy_load(&buf, T3, off + 3 * copy_word_size); | ||
499 | build_copy_store_pref(&buf, off); | ||
500 | build_copy_store(&buf, T0, off); | ||
501 | build_copy_store_pref(&buf, off + copy_word_size); | ||
502 | build_copy_store(&buf, T1, off + copy_word_size); | ||
503 | build_copy_store_pref(&buf, off + 2 * copy_word_size); | ||
504 | build_copy_store(&buf, T2, off + 2 * copy_word_size); | ||
505 | build_copy_store_pref(&buf, off + 3 * copy_word_size); | ||
506 | if (off == -(4 * copy_word_size)) | ||
507 | uasm_il_bne(&buf, &r, A2, A0, label_copy_pref_both); | ||
508 | build_copy_store(&buf, T3, off + 3 * copy_word_size); | ||
509 | off += 4 * copy_word_size; | ||
510 | } while (off < 0); | ||
511 | |||
512 | if (pref_bias_copy_load - pref_bias_copy_store) { | ||
513 | pg_addiu(&buf, A2, A0, | ||
514 | pref_bias_copy_load - pref_bias_copy_store); | ||
515 | uasm_l_copy_pref_store(&l, buf); | ||
516 | off = 0; | ||
517 | do { | ||
518 | build_copy_load(&buf, T0, off); | ||
519 | build_copy_load(&buf, T1, off + copy_word_size); | ||
520 | build_copy_load(&buf, T2, off + 2 * copy_word_size); | ||
521 | build_copy_load(&buf, T3, off + 3 * copy_word_size); | ||
522 | build_copy_store_pref(&buf, off); | ||
523 | build_copy_store(&buf, T0, off); | ||
524 | build_copy_store_pref(&buf, off + copy_word_size); | ||
525 | build_copy_store(&buf, T1, off + copy_word_size); | ||
526 | build_copy_store_pref(&buf, off + 2 * copy_word_size); | ||
527 | build_copy_store(&buf, T2, off + 2 * copy_word_size); | ||
528 | build_copy_store_pref(&buf, off + 3 * copy_word_size); | ||
529 | build_copy_store(&buf, T3, off + 3 * copy_word_size); | ||
530 | off += 4 * copy_word_size; | ||
531 | } while (off < half_copy_loop_size); | ||
532 | pg_addiu(&buf, A1, A1, 2 * off); | ||
533 | pg_addiu(&buf, A0, A0, 2 * off); | ||
534 | off = -off; | ||
535 | do { | ||
536 | build_copy_load(&buf, T0, off); | ||
537 | build_copy_load(&buf, T1, off + copy_word_size); | ||
538 | build_copy_load(&buf, T2, off + 2 * copy_word_size); | ||
539 | build_copy_load(&buf, T3, off + 3 * copy_word_size); | ||
540 | build_copy_store_pref(&buf, off); | ||
541 | build_copy_store(&buf, T0, off); | ||
542 | build_copy_store_pref(&buf, off + copy_word_size); | ||
543 | build_copy_store(&buf, T1, off + copy_word_size); | ||
544 | build_copy_store_pref(&buf, off + 2 * copy_word_size); | ||
545 | build_copy_store(&buf, T2, off + 2 * copy_word_size); | ||
546 | build_copy_store_pref(&buf, off + 3 * copy_word_size); | ||
547 | if (off == -(4 * copy_word_size)) | ||
548 | uasm_il_bne(&buf, &r, A2, A0, | ||
549 | label_copy_pref_store); | ||
550 | build_copy_store(&buf, T3, off + 3 * copy_word_size); | ||
551 | off += 4 * copy_word_size; | ||
552 | } while (off < 0); | ||
553 | } | ||
554 | |||
555 | if (pref_bias_copy_store) { | ||
556 | pg_addiu(&buf, A2, A0, pref_bias_copy_store); | ||
557 | uasm_l_copy_nopref(&l, buf); | ||
558 | off = 0; | ||
559 | do { | ||
560 | build_copy_load(&buf, T0, off); | ||
561 | build_copy_load(&buf, T1, off + copy_word_size); | ||
562 | build_copy_load(&buf, T2, off + 2 * copy_word_size); | ||
563 | build_copy_load(&buf, T3, off + 3 * copy_word_size); | ||
564 | build_copy_store(&buf, T0, off); | ||
565 | build_copy_store(&buf, T1, off + copy_word_size); | ||
566 | build_copy_store(&buf, T2, off + 2 * copy_word_size); | ||
567 | build_copy_store(&buf, T3, off + 3 * copy_word_size); | ||
568 | off += 4 * copy_word_size; | ||
569 | } while (off < half_copy_loop_size); | ||
570 | pg_addiu(&buf, A1, A1, 2 * off); | ||
571 | pg_addiu(&buf, A0, A0, 2 * off); | ||
572 | off = -off; | ||
573 | do { | ||
574 | build_copy_load(&buf, T0, off); | ||
575 | build_copy_load(&buf, T1, off + copy_word_size); | ||
576 | build_copy_load(&buf, T2, off + 2 * copy_word_size); | ||
577 | build_copy_load(&buf, T3, off + 3 * copy_word_size); | ||
578 | build_copy_store(&buf, T0, off); | ||
579 | build_copy_store(&buf, T1, off + copy_word_size); | ||
580 | build_copy_store(&buf, T2, off + 2 * copy_word_size); | ||
581 | if (off == -(4 * copy_word_size)) | ||
582 | uasm_il_bne(&buf, &r, A2, A0, | ||
583 | label_copy_nopref); | ||
584 | build_copy_store(&buf, T3, off + 3 * copy_word_size); | ||
585 | off += 4 * copy_word_size; | ||
586 | } while (off < 0); | ||
587 | } | ||
588 | |||
589 | uasm_i_jr(&buf, RA); | ||
590 | uasm_i_nop(&buf); | ||
591 | |||
592 | BUG_ON(buf > copy_page_array + ARRAY_SIZE(copy_page_array)); | ||
593 | |||
594 | uasm_resolve_relocs(relocs, labels); | ||
595 | |||
596 | pr_debug("Synthesized copy page handler (%u instructions).\n", | ||
597 | (u32)(buf - copy_page_array)); | ||
598 | |||
599 | pr_debug("\t.set push\n"); | ||
600 | pr_debug("\t.set noreorder\n"); | ||
601 | for (i = 0; i < (buf - copy_page_array); i++) | ||
602 | pr_debug("\t.word 0x%08x\n", copy_page_array[i]); | ||
603 | pr_debug("\t.set pop\n"); | ||
604 | } | ||
605 | |||
606 | #ifdef CONFIG_SIBYTE_DMA_PAGEOPS | ||
607 | |||
608 | /* | ||
609 | * Pad descriptors to cacheline, since each is exclusively owned by a | ||
610 | * particular CPU. | ||
611 | */ | ||
612 | struct dmadscr { | ||
613 | u64 dscr_a; | ||
614 | u64 dscr_b; | ||
615 | u64 pad_a; | ||
616 | u64 pad_b; | ||
617 | } ____cacheline_aligned_in_smp page_descr[DM_NUM_CHANNELS]; | ||
618 | |||
619 | void sb1_dma_init(void) | ||
620 | { | ||
621 | int i; | ||
622 | |||
623 | for (i = 0; i < DM_NUM_CHANNELS; i++) { | ||
624 | const u64 base_val = CPHYSADDR((unsigned long)&page_descr[i]) | | ||
625 | V_DM_DSCR_BASE_RINGSZ(1); | ||
626 | void *base_reg = IOADDR(A_DM_REGISTER(i, R_DM_DSCR_BASE)); | ||
627 | |||
628 | __raw_writeq(base_val, base_reg); | ||
629 | __raw_writeq(base_val | M_DM_DSCR_BASE_RESET, base_reg); | ||
630 | __raw_writeq(base_val | M_DM_DSCR_BASE_ENABL, base_reg); | ||
631 | } | ||
632 | } | ||
633 | |||
634 | void clear_page(void *page) | ||
635 | { | ||
636 | u64 to_phys = CPHYSADDR((unsigned long)page); | ||
637 | unsigned int cpu = smp_processor_id(); | ||
638 | |||
639 | /* if the page is not in KSEG0, use old way */ | ||
640 | if ((long)KSEGX((unsigned long)page) != (long)CKSEG0) | ||
641 | return clear_page_cpu(page); | ||
642 | |||
643 | page_descr[cpu].dscr_a = to_phys | M_DM_DSCRA_ZERO_MEM | | ||
644 | M_DM_DSCRA_L2C_DEST | M_DM_DSCRA_INTERRUPT; | ||
645 | page_descr[cpu].dscr_b = V_DM_DSCRB_SRC_LENGTH(PAGE_SIZE); | ||
646 | __raw_writeq(1, IOADDR(A_DM_REGISTER(cpu, R_DM_DSCR_COUNT))); | ||
647 | |||
648 | /* | ||
649 | * Don't really want to do it this way, but there's no | ||
650 | * reliable way to delay completion detection. | ||
651 | */ | ||
652 | while (!(__raw_readq(IOADDR(A_DM_REGISTER(cpu, R_DM_DSCR_BASE_DEBUG))) | ||
653 | & M_DM_DSCR_BASE_INTERRUPT)) | ||
654 | ; | ||
655 | __raw_readq(IOADDR(A_DM_REGISTER(cpu, R_DM_DSCR_BASE))); | ||
656 | } | ||
657 | |||
658 | void copy_page(void *to, void *from) | ||
659 | { | ||
660 | u64 from_phys = CPHYSADDR((unsigned long)from); | ||
661 | u64 to_phys = CPHYSADDR((unsigned long)to); | ||
662 | unsigned int cpu = smp_processor_id(); | ||
663 | |||
664 | /* if any page is not in KSEG0, use old way */ | ||
665 | if ((long)KSEGX((unsigned long)to) != (long)CKSEG0 | ||
666 | || (long)KSEGX((unsigned long)from) != (long)CKSEG0) | ||
667 | return copy_page_cpu(to, from); | ||
668 | |||
669 | page_descr[cpu].dscr_a = to_phys | M_DM_DSCRA_L2C_DEST | | ||
670 | M_DM_DSCRA_INTERRUPT; | ||
671 | page_descr[cpu].dscr_b = from_phys | V_DM_DSCRB_SRC_LENGTH(PAGE_SIZE); | ||
672 | __raw_writeq(1, IOADDR(A_DM_REGISTER(cpu, R_DM_DSCR_COUNT))); | ||
673 | |||
674 | /* | ||
675 | * Don't really want to do it this way, but there's no | ||
676 | * reliable way to delay completion detection. | ||
677 | */ | ||
678 | while (!(__raw_readq(IOADDR(A_DM_REGISTER(cpu, R_DM_DSCR_BASE_DEBUG))) | ||
679 | & M_DM_DSCR_BASE_INTERRUPT)) | ||
680 | ; | ||
681 | __raw_readq(IOADDR(A_DM_REGISTER(cpu, R_DM_DSCR_BASE))); | ||
682 | } | ||
683 | |||
684 | #endif /* CONFIG_SIBYTE_DMA_PAGEOPS */ | ||
diff --git a/arch/mips/mm/pg-r4k.c b/arch/mips/mm/pg-r4k.c deleted file mode 100644 index 455dedb5b39e..000000000000 --- a/arch/mips/mm/pg-r4k.c +++ /dev/null | |||
@@ -1,534 +0,0 @@ | |||
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 | * Copyright (C) 2003, 04, 05 Ralf Baechle (ralf@linux-mips.org) | ||
7 | * Copyright (C) 2007 Maciej W. Rozycki | ||
8 | */ | ||
9 | #include <linux/init.h> | ||
10 | #include <linux/kernel.h> | ||
11 | #include <linux/sched.h> | ||
12 | #include <linux/mm.h> | ||
13 | #include <linux/module.h> | ||
14 | #include <linux/proc_fs.h> | ||
15 | |||
16 | #include <asm/bugs.h> | ||
17 | #include <asm/cacheops.h> | ||
18 | #include <asm/inst.h> | ||
19 | #include <asm/io.h> | ||
20 | #include <asm/page.h> | ||
21 | #include <asm/pgtable.h> | ||
22 | #include <asm/prefetch.h> | ||
23 | #include <asm/system.h> | ||
24 | #include <asm/bootinfo.h> | ||
25 | #include <asm/mipsregs.h> | ||
26 | #include <asm/mmu_context.h> | ||
27 | #include <asm/cpu.h> | ||
28 | #include <asm/war.h> | ||
29 | |||
30 | #define half_scache_line_size() (cpu_scache_line_size() >> 1) | ||
31 | #define cpu_is_r4600_v1_x() ((read_c0_prid() & 0xfffffff0) == 0x00002010) | ||
32 | #define cpu_is_r4600_v2_x() ((read_c0_prid() & 0xfffffff0) == 0x00002020) | ||
33 | |||
34 | |||
35 | /* | ||
36 | * Maximum sizes: | ||
37 | * | ||
38 | * R4000 128 bytes S-cache: 0x58 bytes | ||
39 | * R4600 v1.7: 0x5c bytes | ||
40 | * R4600 v2.0: 0x60 bytes | ||
41 | * With prefetching, 16 byte strides 0xa0 bytes | ||
42 | */ | ||
43 | |||
44 | static unsigned int clear_page_array[0x130 / 4]; | ||
45 | |||
46 | void clear_page(void * page) __attribute__((alias("clear_page_array"))); | ||
47 | |||
48 | EXPORT_SYMBOL(clear_page); | ||
49 | |||
50 | /* | ||
51 | * Maximum sizes: | ||
52 | * | ||
53 | * R4000 128 bytes S-cache: 0x11c bytes | ||
54 | * R4600 v1.7: 0x080 bytes | ||
55 | * R4600 v2.0: 0x07c bytes | ||
56 | * With prefetching, 16 byte strides 0x0b8 bytes | ||
57 | */ | ||
58 | static unsigned int copy_page_array[0x148 / 4]; | ||
59 | |||
60 | void copy_page(void *to, void *from) __attribute__((alias("copy_page_array"))); | ||
61 | |||
62 | EXPORT_SYMBOL(copy_page); | ||
63 | |||
64 | /* | ||
65 | * This is suboptimal for 32-bit kernels; we assume that R10000 is only used | ||
66 | * with 64-bit kernels. The prefetch offsets have been experimentally tuned | ||
67 | * an Origin 200. | ||
68 | */ | ||
69 | static int pref_offset_clear __cpuinitdata = 512; | ||
70 | static int pref_offset_copy __cpuinitdata = 256; | ||
71 | |||
72 | static unsigned int pref_src_mode __cpuinitdata; | ||
73 | static unsigned int pref_dst_mode __cpuinitdata; | ||
74 | |||
75 | static int load_offset __cpuinitdata; | ||
76 | static int store_offset __cpuinitdata; | ||
77 | |||
78 | static unsigned int __cpuinitdata *dest, *epc; | ||
79 | |||
80 | static unsigned int instruction_pending; | ||
81 | static union mips_instruction delayed_mi; | ||
82 | |||
83 | static void __cpuinit emit_instruction(union mips_instruction mi) | ||
84 | { | ||
85 | if (instruction_pending) | ||
86 | *epc++ = delayed_mi.word; | ||
87 | |||
88 | instruction_pending = 1; | ||
89 | delayed_mi = mi; | ||
90 | } | ||
91 | |||
92 | static inline void flush_delay_slot_or_nop(void) | ||
93 | { | ||
94 | if (instruction_pending) { | ||
95 | *epc++ = delayed_mi.word; | ||
96 | instruction_pending = 0; | ||
97 | return; | ||
98 | } | ||
99 | |||
100 | *epc++ = 0; | ||
101 | } | ||
102 | |||
103 | static inline unsigned int *label(void) | ||
104 | { | ||
105 | if (instruction_pending) { | ||
106 | *epc++ = delayed_mi.word; | ||
107 | instruction_pending = 0; | ||
108 | } | ||
109 | |||
110 | return epc; | ||
111 | } | ||
112 | |||
113 | static inline void build_insn_word(unsigned int word) | ||
114 | { | ||
115 | union mips_instruction mi; | ||
116 | |||
117 | mi.word = word; | ||
118 | |||
119 | emit_instruction(mi); | ||
120 | } | ||
121 | |||
122 | static inline void build_nop(void) | ||
123 | { | ||
124 | build_insn_word(0); /* nop */ | ||
125 | } | ||
126 | |||
127 | static inline void build_src_pref(int advance) | ||
128 | { | ||
129 | if (!(load_offset & (cpu_dcache_line_size() - 1)) && advance) { | ||
130 | union mips_instruction mi; | ||
131 | |||
132 | mi.i_format.opcode = pref_op; | ||
133 | mi.i_format.rs = 5; /* $a1 */ | ||
134 | mi.i_format.rt = pref_src_mode; | ||
135 | mi.i_format.simmediate = load_offset + advance; | ||
136 | |||
137 | emit_instruction(mi); | ||
138 | } | ||
139 | } | ||
140 | |||
141 | static inline void __build_load_reg(int reg) | ||
142 | { | ||
143 | union mips_instruction mi; | ||
144 | unsigned int width; | ||
145 | |||
146 | if (cpu_has_64bit_gp_regs) { | ||
147 | mi.i_format.opcode = ld_op; | ||
148 | width = 8; | ||
149 | } else { | ||
150 | mi.i_format.opcode = lw_op; | ||
151 | width = 4; | ||
152 | } | ||
153 | mi.i_format.rs = 5; /* $a1 */ | ||
154 | mi.i_format.rt = reg; /* $reg */ | ||
155 | mi.i_format.simmediate = load_offset; | ||
156 | |||
157 | load_offset += width; | ||
158 | emit_instruction(mi); | ||
159 | } | ||
160 | |||
161 | static inline void build_load_reg(int reg) | ||
162 | { | ||
163 | if (cpu_has_prefetch) | ||
164 | build_src_pref(pref_offset_copy); | ||
165 | |||
166 | __build_load_reg(reg); | ||
167 | } | ||
168 | |||
169 | static inline void build_dst_pref(int advance) | ||
170 | { | ||
171 | if (!(store_offset & (cpu_dcache_line_size() - 1)) && advance) { | ||
172 | union mips_instruction mi; | ||
173 | |||
174 | mi.i_format.opcode = pref_op; | ||
175 | mi.i_format.rs = 4; /* $a0 */ | ||
176 | mi.i_format.rt = pref_dst_mode; | ||
177 | mi.i_format.simmediate = store_offset + advance; | ||
178 | |||
179 | emit_instruction(mi); | ||
180 | } | ||
181 | } | ||
182 | |||
183 | static inline void build_cdex_s(void) | ||
184 | { | ||
185 | union mips_instruction mi; | ||
186 | |||
187 | if ((store_offset & (cpu_scache_line_size() - 1))) | ||
188 | return; | ||
189 | |||
190 | mi.c_format.opcode = cache_op; | ||
191 | mi.c_format.rs = 4; /* $a0 */ | ||
192 | mi.c_format.c_op = 3; /* Create Dirty Exclusive */ | ||
193 | mi.c_format.cache = 3; /* Secondary Data Cache */ | ||
194 | mi.c_format.simmediate = store_offset; | ||
195 | |||
196 | emit_instruction(mi); | ||
197 | } | ||
198 | |||
199 | static inline void build_cdex_p(void) | ||
200 | { | ||
201 | union mips_instruction mi; | ||
202 | |||
203 | if (store_offset & (cpu_dcache_line_size() - 1)) | ||
204 | return; | ||
205 | |||
206 | if (R4600_V1_HIT_CACHEOP_WAR && cpu_is_r4600_v1_x()) { | ||
207 | build_nop(); | ||
208 | build_nop(); | ||
209 | build_nop(); | ||
210 | build_nop(); | ||
211 | } | ||
212 | |||
213 | if (R4600_V2_HIT_CACHEOP_WAR && cpu_is_r4600_v2_x()) | ||
214 | build_insn_word(0x8c200000); /* lw $zero, ($at) */ | ||
215 | |||
216 | mi.c_format.opcode = cache_op; | ||
217 | mi.c_format.rs = 4; /* $a0 */ | ||
218 | mi.c_format.c_op = 3; /* Create Dirty Exclusive */ | ||
219 | mi.c_format.cache = 1; /* Data Cache */ | ||
220 | mi.c_format.simmediate = store_offset; | ||
221 | |||
222 | emit_instruction(mi); | ||
223 | } | ||
224 | |||
225 | static void __cpuinit __build_store_reg(int reg) | ||
226 | { | ||
227 | union mips_instruction mi; | ||
228 | unsigned int width; | ||
229 | |||
230 | if (cpu_has_64bit_gp_regs || | ||
231 | (cpu_has_64bit_zero_reg && reg == 0)) { | ||
232 | mi.i_format.opcode = sd_op; | ||
233 | width = 8; | ||
234 | } else { | ||
235 | mi.i_format.opcode = sw_op; | ||
236 | width = 4; | ||
237 | } | ||
238 | mi.i_format.rs = 4; /* $a0 */ | ||
239 | mi.i_format.rt = reg; /* $reg */ | ||
240 | mi.i_format.simmediate = store_offset; | ||
241 | |||
242 | store_offset += width; | ||
243 | emit_instruction(mi); | ||
244 | } | ||
245 | |||
246 | static inline void build_store_reg(int reg) | ||
247 | { | ||
248 | int pref_off = cpu_has_prefetch ? | ||
249 | (reg ? pref_offset_copy : pref_offset_clear) : 0; | ||
250 | if (pref_off) | ||
251 | build_dst_pref(pref_off); | ||
252 | else if (cpu_has_cache_cdex_s) | ||
253 | build_cdex_s(); | ||
254 | else if (cpu_has_cache_cdex_p) | ||
255 | build_cdex_p(); | ||
256 | |||
257 | __build_store_reg(reg); | ||
258 | } | ||
259 | |||
260 | static inline void build_addiu_rt_rs(unsigned int rt, unsigned int rs, | ||
261 | unsigned long offset) | ||
262 | { | ||
263 | union mips_instruction mi; | ||
264 | |||
265 | BUG_ON(offset > 0x7fff); | ||
266 | |||
267 | if (cpu_has_64bit_gp_regs && DADDI_WAR && r4k_daddiu_bug()) { | ||
268 | mi.i_format.opcode = addiu_op; | ||
269 | mi.i_format.rs = 0; /* $zero */ | ||
270 | mi.i_format.rt = 25; /* $t9 */ | ||
271 | mi.i_format.simmediate = offset; | ||
272 | emit_instruction(mi); | ||
273 | |||
274 | mi.r_format.opcode = spec_op; | ||
275 | mi.r_format.rs = rs; | ||
276 | mi.r_format.rt = 25; /* $t9 */ | ||
277 | mi.r_format.rd = rt; | ||
278 | mi.r_format.re = 0; | ||
279 | mi.r_format.func = daddu_op; | ||
280 | } else { | ||
281 | mi.i_format.opcode = cpu_has_64bit_gp_regs ? | ||
282 | daddiu_op : addiu_op; | ||
283 | mi.i_format.rs = rs; | ||
284 | mi.i_format.rt = rt; | ||
285 | mi.i_format.simmediate = offset; | ||
286 | } | ||
287 | emit_instruction(mi); | ||
288 | } | ||
289 | |||
290 | static inline void build_addiu_a2_a0(unsigned long offset) | ||
291 | { | ||
292 | build_addiu_rt_rs(6, 4, offset); /* $a2, $a0, offset */ | ||
293 | } | ||
294 | |||
295 | static inline void build_addiu_a2(unsigned long offset) | ||
296 | { | ||
297 | build_addiu_rt_rs(6, 6, offset); /* $a2, $a2, offset */ | ||
298 | } | ||
299 | |||
300 | static inline void build_addiu_a1(unsigned long offset) | ||
301 | { | ||
302 | build_addiu_rt_rs(5, 5, offset); /* $a1, $a1, offset */ | ||
303 | |||
304 | load_offset -= offset; | ||
305 | } | ||
306 | |||
307 | static inline void build_addiu_a0(unsigned long offset) | ||
308 | { | ||
309 | build_addiu_rt_rs(4, 4, offset); /* $a0, $a0, offset */ | ||
310 | |||
311 | store_offset -= offset; | ||
312 | } | ||
313 | |||
314 | static inline void build_bne(unsigned int *dest) | ||
315 | { | ||
316 | union mips_instruction mi; | ||
317 | |||
318 | mi.i_format.opcode = bne_op; | ||
319 | mi.i_format.rs = 6; /* $a2 */ | ||
320 | mi.i_format.rt = 4; /* $a0 */ | ||
321 | mi.i_format.simmediate = dest - epc - 1; | ||
322 | |||
323 | *epc++ = mi.word; | ||
324 | flush_delay_slot_or_nop(); | ||
325 | } | ||
326 | |||
327 | static inline void build_jr_ra(void) | ||
328 | { | ||
329 | union mips_instruction mi; | ||
330 | |||
331 | mi.r_format.opcode = spec_op; | ||
332 | mi.r_format.rs = 31; | ||
333 | mi.r_format.rt = 0; | ||
334 | mi.r_format.rd = 0; | ||
335 | mi.r_format.re = 0; | ||
336 | mi.r_format.func = jr_op; | ||
337 | |||
338 | *epc++ = mi.word; | ||
339 | flush_delay_slot_or_nop(); | ||
340 | } | ||
341 | |||
342 | void __cpuinit build_clear_page(void) | ||
343 | { | ||
344 | unsigned int loop_start; | ||
345 | unsigned long off; | ||
346 | int i; | ||
347 | |||
348 | epc = (unsigned int *) &clear_page_array; | ||
349 | instruction_pending = 0; | ||
350 | store_offset = 0; | ||
351 | |||
352 | if (cpu_has_prefetch) { | ||
353 | switch (current_cpu_type()) { | ||
354 | case CPU_TX49XX: | ||
355 | /* TX49 supports only Pref_Load */ | ||
356 | pref_offset_clear = 0; | ||
357 | pref_offset_copy = 0; | ||
358 | break; | ||
359 | |||
360 | case CPU_RM9000: | ||
361 | /* | ||
362 | * As a workaround for erratum G105 which make the | ||
363 | * PrepareForStore hint unusable we fall back to | ||
364 | * StoreRetained on the RM9000. Once it is known which | ||
365 | * versions of the RM9000 we'll be able to condition- | ||
366 | * alize this. | ||
367 | */ | ||
368 | |||
369 | case CPU_R10000: | ||
370 | case CPU_R12000: | ||
371 | case CPU_R14000: | ||
372 | pref_src_mode = Pref_LoadStreamed; | ||
373 | pref_dst_mode = Pref_StoreStreamed; | ||
374 | break; | ||
375 | |||
376 | default: | ||
377 | pref_src_mode = Pref_LoadStreamed; | ||
378 | pref_dst_mode = Pref_PrepareForStore; | ||
379 | break; | ||
380 | } | ||
381 | } | ||
382 | |||
383 | off = PAGE_SIZE - (cpu_has_prefetch ? pref_offset_clear : 0); | ||
384 | if (off > 0x7fff) { | ||
385 | build_addiu_a2_a0(off >> 1); | ||
386 | build_addiu_a2(off >> 1); | ||
387 | } else | ||
388 | build_addiu_a2_a0(off); | ||
389 | |||
390 | if (R4600_V2_HIT_CACHEOP_WAR && cpu_is_r4600_v2_x()) | ||
391 | build_insn_word(0x3c01a000); /* lui $at, 0xa000 */ | ||
392 | |||
393 | dest = label(); | ||
394 | do { | ||
395 | build_store_reg(0); | ||
396 | build_store_reg(0); | ||
397 | build_store_reg(0); | ||
398 | build_store_reg(0); | ||
399 | } while (store_offset < half_scache_line_size()); | ||
400 | build_addiu_a0(2 * store_offset); | ||
401 | loop_start = store_offset; | ||
402 | do { | ||
403 | build_store_reg(0); | ||
404 | build_store_reg(0); | ||
405 | build_store_reg(0); | ||
406 | build_store_reg(0); | ||
407 | } while ((store_offset - loop_start) < half_scache_line_size()); | ||
408 | build_bne(dest); | ||
409 | |||
410 | if (cpu_has_prefetch && pref_offset_clear) { | ||
411 | build_addiu_a2_a0(pref_offset_clear); | ||
412 | dest = label(); | ||
413 | loop_start = store_offset; | ||
414 | do { | ||
415 | __build_store_reg(0); | ||
416 | __build_store_reg(0); | ||
417 | __build_store_reg(0); | ||
418 | __build_store_reg(0); | ||
419 | } while ((store_offset - loop_start) < half_scache_line_size()); | ||
420 | build_addiu_a0(2 * store_offset); | ||
421 | loop_start = store_offset; | ||
422 | do { | ||
423 | __build_store_reg(0); | ||
424 | __build_store_reg(0); | ||
425 | __build_store_reg(0); | ||
426 | __build_store_reg(0); | ||
427 | } while ((store_offset - loop_start) < half_scache_line_size()); | ||
428 | build_bne(dest); | ||
429 | } | ||
430 | |||
431 | build_jr_ra(); | ||
432 | |||
433 | BUG_ON(epc > clear_page_array + ARRAY_SIZE(clear_page_array)); | ||
434 | |||
435 | pr_info("Synthesized clear page handler (%u instructions).\n", | ||
436 | (unsigned int)(epc - clear_page_array)); | ||
437 | |||
438 | pr_debug("\t.set push\n"); | ||
439 | pr_debug("\t.set noreorder\n"); | ||
440 | for (i = 0; i < (epc - clear_page_array); i++) | ||
441 | pr_debug("\t.word 0x%08x\n", clear_page_array[i]); | ||
442 | pr_debug("\t.set pop\n"); | ||
443 | } | ||
444 | |||
445 | void __cpuinit build_copy_page(void) | ||
446 | { | ||
447 | unsigned int loop_start; | ||
448 | unsigned long off; | ||
449 | int i; | ||
450 | |||
451 | epc = (unsigned int *) ©_page_array; | ||
452 | store_offset = load_offset = 0; | ||
453 | instruction_pending = 0; | ||
454 | |||
455 | off = PAGE_SIZE - (cpu_has_prefetch ? pref_offset_copy : 0); | ||
456 | if (off > 0x7fff) { | ||
457 | build_addiu_a2_a0(off >> 1); | ||
458 | build_addiu_a2(off >> 1); | ||
459 | } else | ||
460 | build_addiu_a2_a0(off); | ||
461 | |||
462 | if (R4600_V2_HIT_CACHEOP_WAR && cpu_is_r4600_v2_x()) | ||
463 | build_insn_word(0x3c01a000); /* lui $at, 0xa000 */ | ||
464 | |||
465 | dest = label(); | ||
466 | loop_start = store_offset; | ||
467 | do { | ||
468 | build_load_reg( 8); | ||
469 | build_load_reg( 9); | ||
470 | build_load_reg(10); | ||
471 | build_load_reg(11); | ||
472 | build_store_reg( 8); | ||
473 | build_store_reg( 9); | ||
474 | build_store_reg(10); | ||
475 | build_store_reg(11); | ||
476 | } while ((store_offset - loop_start) < half_scache_line_size()); | ||
477 | build_addiu_a0(2 * store_offset); | ||
478 | build_addiu_a1(2 * load_offset); | ||
479 | loop_start = store_offset; | ||
480 | do { | ||
481 | build_load_reg( 8); | ||
482 | build_load_reg( 9); | ||
483 | build_load_reg(10); | ||
484 | build_load_reg(11); | ||
485 | build_store_reg( 8); | ||
486 | build_store_reg( 9); | ||
487 | build_store_reg(10); | ||
488 | build_store_reg(11); | ||
489 | } while ((store_offset - loop_start) < half_scache_line_size()); | ||
490 | build_bne(dest); | ||
491 | |||
492 | if (cpu_has_prefetch && pref_offset_copy) { | ||
493 | build_addiu_a2_a0(pref_offset_copy); | ||
494 | dest = label(); | ||
495 | loop_start = store_offset; | ||
496 | do { | ||
497 | __build_load_reg( 8); | ||
498 | __build_load_reg( 9); | ||
499 | __build_load_reg(10); | ||
500 | __build_load_reg(11); | ||
501 | __build_store_reg( 8); | ||
502 | __build_store_reg( 9); | ||
503 | __build_store_reg(10); | ||
504 | __build_store_reg(11); | ||
505 | } while ((store_offset - loop_start) < half_scache_line_size()); | ||
506 | build_addiu_a0(2 * store_offset); | ||
507 | build_addiu_a1(2 * load_offset); | ||
508 | loop_start = store_offset; | ||
509 | do { | ||
510 | __build_load_reg( 8); | ||
511 | __build_load_reg( 9); | ||
512 | __build_load_reg(10); | ||
513 | __build_load_reg(11); | ||
514 | __build_store_reg( 8); | ||
515 | __build_store_reg( 9); | ||
516 | __build_store_reg(10); | ||
517 | __build_store_reg(11); | ||
518 | } while ((store_offset - loop_start) < half_scache_line_size()); | ||
519 | build_bne(dest); | ||
520 | } | ||
521 | |||
522 | build_jr_ra(); | ||
523 | |||
524 | BUG_ON(epc > copy_page_array + ARRAY_SIZE(copy_page_array)); | ||
525 | |||
526 | pr_info("Synthesized copy page handler (%u instructions).\n", | ||
527 | (unsigned int)(epc - copy_page_array)); | ||
528 | |||
529 | pr_debug("\t.set push\n"); | ||
530 | pr_debug("\t.set noreorder\n"); | ||
531 | for (i = 0; i < (epc - copy_page_array); i++) | ||
532 | pr_debug("\t.word 0x%08x\n", copy_page_array[i]); | ||
533 | pr_debug("\t.set pop\n"); | ||
534 | } | ||
diff --git a/arch/mips/mm/pg-sb1.c b/arch/mips/mm/pg-sb1.c deleted file mode 100644 index 49e289d05414..000000000000 --- a/arch/mips/mm/pg-sb1.c +++ /dev/null | |||
@@ -1,302 +0,0 @@ | |||
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/module.h> | ||
26 | #include <linux/sched.h> | ||
27 | #include <linux/smp.h> | ||
28 | |||
29 | #include <asm/io.h> | ||
30 | #include <asm/sibyte/sb1250.h> | ||
31 | #include <asm/sibyte/sb1250_regs.h> | ||
32 | #include <asm/sibyte/sb1250_dma.h> | ||
33 | |||
34 | #ifdef CONFIG_SB1_PASS_1_WORKAROUNDS | ||
35 | #define SB1_PREF_LOAD_STREAMED_HINT "0" | ||
36 | #define SB1_PREF_STORE_STREAMED_HINT "1" | ||
37 | #else | ||
38 | #define SB1_PREF_LOAD_STREAMED_HINT "4" | ||
39 | #define SB1_PREF_STORE_STREAMED_HINT "5" | ||
40 | #endif | ||
41 | |||
42 | static inline void clear_page_cpu(void *page) | ||
43 | { | ||
44 | unsigned char *addr = (unsigned char *) page; | ||
45 | unsigned char *end = addr + PAGE_SIZE; | ||
46 | |||
47 | /* | ||
48 | * JDCXXX - This should be bottlenecked by the write buffer, but these | ||
49 | * things tend to be mildly unpredictable...should check this on the | ||
50 | * performance model | ||
51 | * | ||
52 | * We prefetch 4 lines ahead. We're also "cheating" slightly here... | ||
53 | * since we know we're on an SB1, we force the assembler to take | ||
54 | * 64-bit operands to speed things up | ||
55 | */ | ||
56 | __asm__ __volatile__( | ||
57 | " .set push \n" | ||
58 | " .set mips4 \n" | ||
59 | " .set noreorder \n" | ||
60 | #ifdef CONFIG_CPU_HAS_PREFETCH | ||
61 | " daddiu %0, %0, 128 \n" | ||
62 | " pref " SB1_PREF_STORE_STREAMED_HINT ", -128(%0) \n" | ||
63 | /* 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" | ||
110 | /* Prefetch the first 4 lines */ | ||
111 | " pref " SB1_PREF_STORE_STREAMED_HINT ", -128(%1)\n" | ||
112 | " pref " SB1_PREF_LOAD_STREAMED_HINT ", -96(%0)\n" | ||
113 | " pref " SB1_PREF_STORE_STREAMED_HINT ", -96(%1)\n" | ||
114 | " pref " SB1_PREF_LOAD_STREAMED_HINT ", -64(%0)\n" | ||
115 | " pref " SB1_PREF_STORE_STREAMED_HINT ", -64(%1)\n" | ||
116 | " pref " SB1_PREF_LOAD_STREAMED_HINT ", -32(%0)\n" | ||
117 | "1: pref " SB1_PREF_STORE_STREAMED_HINT ", -32(%1)\n" | ||
118 | # ifdef CONFIG_64BIT | ||
119 | " ld $8, -128(%0) \n" /* Block copy a cacheline */ | ||
120 | " ld $9, -120(%0) \n" | ||
121 | " ld $10, -112(%0) \n" | ||
122 | " ld $11, -104(%0) \n" | ||
123 | " sd $8, -128(%1) \n" | ||
124 | " sd $9, -120(%1) \n" | ||
125 | " sd $10, -112(%1) \n" | ||
126 | " sd $11, -104(%1) \n" | ||
127 | # else | ||
128 | " lw $2, -128(%0) \n" /* Block copy a cacheline */ | ||
129 | " lw $3, -124(%0) \n" | ||
130 | " lw $6, -120(%0) \n" | ||
131 | " lw $7, -116(%0) \n" | ||
132 | " lw $8, -112(%0) \n" | ||
133 | " lw $9, -108(%0) \n" | ||
134 | " lw $10, -104(%0) \n" | ||
135 | " lw $11, -100(%0) \n" | ||
136 | " sw $2, -128(%1) \n" | ||
137 | " sw $3, -124(%1) \n" | ||
138 | " sw $6, -120(%1) \n" | ||
139 | " sw $7, -116(%1) \n" | ||
140 | " sw $8, -112(%1) \n" | ||
141 | " sw $9, -108(%1) \n" | ||
142 | " sw $10, -104(%1) \n" | ||
143 | " sw $11, -100(%1) \n" | ||
144 | # endif | ||
145 | " daddiu %0, %0, 32 \n" | ||
146 | " daddiu %1, %1, 32 \n" | ||
147 | " bnel %0, %2, 1b \n" | ||
148 | " pref " SB1_PREF_LOAD_STREAMED_HINT ", -32(%0)\n" | ||
149 | " daddiu %0, %0, -128 \n" | ||
150 | " daddiu %1, %1, -128 \n" | ||
151 | #endif | ||
152 | #ifdef CONFIG_64BIT | ||
153 | " ld $8, 0(%0) \n" /* Block copy a cacheline */ | ||
154 | "1: ld $9, 8(%0) \n" | ||
155 | " ld $10, 16(%0) \n" | ||
156 | " ld $11, 24(%0) \n" | ||
157 | " sd $8, 0(%1) \n" | ||
158 | " sd $9, 8(%1) \n" | ||
159 | " sd $10, 16(%1) \n" | ||
160 | " sd $11, 24(%1) \n" | ||
161 | #else | ||
162 | " lw $2, 0(%0) \n" /* Block copy a cacheline */ | ||
163 | "1: lw $3, 4(%0) \n" | ||
164 | " lw $6, 8(%0) \n" | ||
165 | " lw $7, 12(%0) \n" | ||
166 | " lw $8, 16(%0) \n" | ||
167 | " lw $9, 20(%0) \n" | ||
168 | " lw $10, 24(%0) \n" | ||
169 | " lw $11, 28(%0) \n" | ||
170 | " sw $2, 0(%1) \n" | ||
171 | " sw $3, 4(%1) \n" | ||
172 | " sw $6, 8(%1) \n" | ||
173 | " sw $7, 12(%1) \n" | ||
174 | " sw $8, 16(%1) \n" | ||
175 | " sw $9, 20(%1) \n" | ||
176 | " sw $10, 24(%1) \n" | ||
177 | " sw $11, 28(%1) \n" | ||
178 | #endif | ||
179 | " daddiu %0, %0, 32 \n" | ||
180 | " daddiu %1, %1, 32 \n" | ||
181 | " bnel %0, %2, 1b \n" | ||
182 | #ifdef CONFIG_64BIT | ||
183 | " ld $8, 0(%0) \n" | ||
184 | #else | ||
185 | " lw $2, 0(%0) \n" | ||
186 | #endif | ||
187 | " .set pop \n" | ||
188 | : "+r" (src), "+r" (dst) | ||
189 | : "r" (end) | ||
190 | #ifdef CONFIG_64BIT | ||
191 | : "$8", "$9", "$10", "$11", "memory"); | ||
192 | #else | ||
193 | : "$2", "$3", "$6", "$7", "$8", "$9", "$10", "$11", "memory"); | ||
194 | #endif | ||
195 | } | ||
196 | |||
197 | |||
198 | #ifdef CONFIG_SIBYTE_DMA_PAGEOPS | ||
199 | |||
200 | /* | ||
201 | * Pad descriptors to cacheline, since each is exclusively owned by a | ||
202 | * particular CPU. | ||
203 | */ | ||
204 | typedef struct dmadscr_s { | ||
205 | u64 dscr_a; | ||
206 | u64 dscr_b; | ||
207 | u64 pad_a; | ||
208 | u64 pad_b; | ||
209 | } dmadscr_t; | ||
210 | |||
211 | static dmadscr_t page_descr[DM_NUM_CHANNELS] | ||
212 | __attribute__((aligned(SMP_CACHE_BYTES))); | ||
213 | |||
214 | void sb1_dma_init(void) | ||
215 | { | ||
216 | int i; | ||
217 | |||
218 | for (i = 0; i < DM_NUM_CHANNELS; i++) { | ||
219 | const u64 base_val = CPHYSADDR((unsigned long)&page_descr[i]) | | ||
220 | V_DM_DSCR_BASE_RINGSZ(1); | ||
221 | void *base_reg = IOADDR(A_DM_REGISTER(i, R_DM_DSCR_BASE)); | ||
222 | |||
223 | __raw_writeq(base_val, base_reg); | ||
224 | __raw_writeq(base_val | M_DM_DSCR_BASE_RESET, base_reg); | ||
225 | __raw_writeq(base_val | M_DM_DSCR_BASE_ENABL, base_reg); | ||
226 | } | ||
227 | } | ||
228 | |||
229 | void clear_page(void *page) | ||
230 | { | ||
231 | u64 to_phys = CPHYSADDR((unsigned long)page); | ||
232 | unsigned int cpu = smp_processor_id(); | ||
233 | |||
234 | /* if the page is not in KSEG0, use old way */ | ||
235 | if ((long)KSEGX((unsigned long)page) != (long)CKSEG0) | ||
236 | return clear_page_cpu(page); | ||
237 | |||
238 | page_descr[cpu].dscr_a = to_phys | M_DM_DSCRA_ZERO_MEM | | ||
239 | M_DM_DSCRA_L2C_DEST | M_DM_DSCRA_INTERRUPT; | ||
240 | page_descr[cpu].dscr_b = V_DM_DSCRB_SRC_LENGTH(PAGE_SIZE); | ||
241 | __raw_writeq(1, IOADDR(A_DM_REGISTER(cpu, R_DM_DSCR_COUNT))); | ||
242 | |||
243 | /* | ||
244 | * Don't really want to do it this way, but there's no | ||
245 | * reliable way to delay completion detection. | ||
246 | */ | ||
247 | while (!(__raw_readq(IOADDR(A_DM_REGISTER(cpu, R_DM_DSCR_BASE_DEBUG))) | ||
248 | & M_DM_DSCR_BASE_INTERRUPT)) | ||
249 | ; | ||
250 | __raw_readq(IOADDR(A_DM_REGISTER(cpu, R_DM_DSCR_BASE))); | ||
251 | } | ||
252 | |||
253 | void copy_page(void *to, void *from) | ||
254 | { | ||
255 | u64 from_phys = CPHYSADDR((unsigned long)from); | ||
256 | u64 to_phys = CPHYSADDR((unsigned long)to); | ||
257 | unsigned int cpu = smp_processor_id(); | ||
258 | |||
259 | /* if any page is not in KSEG0, use old way */ | ||
260 | if ((long)KSEGX((unsigned long)to) != (long)CKSEG0 | ||
261 | || (long)KSEGX((unsigned long)from) != (long)CKSEG0) | ||
262 | return copy_page_cpu(to, from); | ||
263 | |||
264 | page_descr[cpu].dscr_a = to_phys | M_DM_DSCRA_L2C_DEST | | ||
265 | M_DM_DSCRA_INTERRUPT; | ||
266 | page_descr[cpu].dscr_b = from_phys | V_DM_DSCRB_SRC_LENGTH(PAGE_SIZE); | ||
267 | __raw_writeq(1, IOADDR(A_DM_REGISTER(cpu, R_DM_DSCR_COUNT))); | ||
268 | |||
269 | /* | ||
270 | * Don't really want to do it this way, but there's no | ||
271 | * reliable way to delay completion detection. | ||
272 | */ | ||
273 | while (!(__raw_readq(IOADDR(A_DM_REGISTER(cpu, R_DM_DSCR_BASE_DEBUG))) | ||
274 | & M_DM_DSCR_BASE_INTERRUPT)) | ||
275 | ; | ||
276 | __raw_readq(IOADDR(A_DM_REGISTER(cpu, R_DM_DSCR_BASE))); | ||
277 | } | ||
278 | |||
279 | #else /* !CONFIG_SIBYTE_DMA_PAGEOPS */ | ||
280 | |||
281 | void clear_page(void *page) | ||
282 | { | ||
283 | return clear_page_cpu(page); | ||
284 | } | ||
285 | |||
286 | void copy_page(void *to, void *from) | ||
287 | { | ||
288 | return copy_page_cpu(to, from); | ||
289 | } | ||
290 | |||
291 | #endif /* !CONFIG_SIBYTE_DMA_PAGEOPS */ | ||
292 | |||
293 | EXPORT_SYMBOL(clear_page); | ||
294 | EXPORT_SYMBOL(copy_page); | ||
295 | |||
296 | void __cpuinit build_clear_page(void) | ||
297 | { | ||
298 | } | ||
299 | |||
300 | void __cpuinit build_copy_page(void) | ||
301 | { | ||
302 | } | ||
diff --git a/arch/mips/mm/uasm.c b/arch/mips/mm/uasm.c index 1a6f7704cc89..1655aa69e133 100644 --- a/arch/mips/mm/uasm.c +++ b/arch/mips/mm/uasm.c | |||
@@ -58,13 +58,13 @@ enum opcode { | |||
58 | insn_invalid, | 58 | insn_invalid, |
59 | insn_addu, insn_addiu, insn_and, insn_andi, insn_beq, | 59 | insn_addu, insn_addiu, insn_and, insn_andi, insn_beq, |
60 | insn_beql, insn_bgez, insn_bgezl, insn_bltz, insn_bltzl, | 60 | insn_beql, insn_bgez, insn_bgezl, insn_bltz, insn_bltzl, |
61 | insn_bne, insn_daddu, insn_daddiu, insn_dmfc0, insn_dmtc0, | 61 | insn_bne, insn_cache, insn_daddu, insn_daddiu, insn_dmfc0, |
62 | insn_dsll, insn_dsll32, insn_dsra, insn_dsrl, insn_dsrl32, | 62 | insn_dmtc0, insn_dsll, insn_dsll32, insn_dsra, insn_dsrl, |
63 | insn_dsubu, insn_eret, insn_j, insn_jal, insn_jr, insn_ld, | 63 | insn_dsrl32, insn_dsubu, insn_eret, insn_j, insn_jal, insn_jr, |
64 | insn_ll, insn_lld, insn_lui, insn_lw, insn_mfc0, insn_mtc0, | 64 | insn_ld, insn_ll, insn_lld, insn_lui, insn_lw, insn_mfc0, |
65 | insn_ori, insn_rfe, insn_sc, insn_scd, insn_sd, insn_sll, | 65 | insn_mtc0, insn_ori, insn_pref, insn_rfe, insn_sc, insn_scd, |
66 | insn_sra, insn_srl, insn_subu, insn_sw, insn_tlbp, insn_tlbwi, | 66 | insn_sd, insn_sll, insn_sra, insn_srl, insn_subu, insn_sw, |
67 | insn_tlbwr, insn_xor, insn_xori | 67 | insn_tlbp, insn_tlbwi, insn_tlbwr, insn_xor, insn_xori |
68 | }; | 68 | }; |
69 | 69 | ||
70 | struct insn { | 70 | struct insn { |
@@ -94,6 +94,7 @@ static struct insn insn_table[] __cpuinitdata = { | |||
94 | { insn_bltz, M(bcond_op, 0, bltz_op, 0, 0, 0), RS | BIMM }, | 94 | { insn_bltz, M(bcond_op, 0, bltz_op, 0, 0, 0), RS | BIMM }, |
95 | { insn_bltzl, M(bcond_op, 0, bltzl_op, 0, 0, 0), RS | BIMM }, | 95 | { insn_bltzl, M(bcond_op, 0, bltzl_op, 0, 0, 0), RS | BIMM }, |
96 | { insn_bne, M(bne_op, 0, 0, 0, 0, 0), RS | RT | BIMM }, | 96 | { insn_bne, M(bne_op, 0, 0, 0, 0, 0), RS | RT | BIMM }, |
97 | { insn_cache, M(cache_op, 0, 0, 0, 0, 0), RS | RT | SIMM }, | ||
97 | { insn_daddiu, M(daddiu_op, 0, 0, 0, 0, 0), RS | RT | SIMM }, | 98 | { insn_daddiu, M(daddiu_op, 0, 0, 0, 0, 0), RS | RT | SIMM }, |
98 | { insn_daddu, M(spec_op, 0, 0, 0, 0, daddu_op), RS | RT | RD }, | 99 | { insn_daddu, M(spec_op, 0, 0, 0, 0, daddu_op), RS | RT | RD }, |
99 | { insn_dmfc0, M(cop0_op, dmfc_op, 0, 0, 0, 0), RT | RD | SET}, | 100 | { insn_dmfc0, M(cop0_op, dmfc_op, 0, 0, 0, 0), RT | RD | SET}, |
@@ -116,6 +117,7 @@ static struct insn insn_table[] __cpuinitdata = { | |||
116 | { insn_mfc0, M(cop0_op, mfc_op, 0, 0, 0, 0), RT | RD | SET}, | 117 | { insn_mfc0, M(cop0_op, mfc_op, 0, 0, 0, 0), RT | RD | SET}, |
117 | { insn_mtc0, M(cop0_op, mtc_op, 0, 0, 0, 0), RT | RD | SET}, | 118 | { insn_mtc0, M(cop0_op, mtc_op, 0, 0, 0, 0), RT | RD | SET}, |
118 | { insn_ori, M(ori_op, 0, 0, 0, 0, 0), RS | RT | UIMM }, | 119 | { insn_ori, M(ori_op, 0, 0, 0, 0, 0), RS | RT | UIMM }, |
120 | { insn_pref, M(pref_op, 0, 0, 0, 0, 0), RS | RT | SIMM }, | ||
119 | { insn_rfe, M(cop0_op, cop_op, 0, 0, 0, rfe_op), 0 }, | 121 | { insn_rfe, M(cop0_op, cop_op, 0, 0, 0, rfe_op), 0 }, |
120 | { insn_sc, M(sc_op, 0, 0, 0, 0, 0), RS | RT | SIMM }, | 122 | { insn_sc, M(sc_op, 0, 0, 0, 0, 0), RS | RT | SIMM }, |
121 | { insn_scd, M(scd_op, 0, 0, 0, 0, 0), RS | RT | SIMM }, | 123 | { insn_scd, M(scd_op, 0, 0, 0, 0, 0), RS | RT | SIMM }, |
@@ -337,6 +339,7 @@ I_u1s2(_bgezl) | |||
337 | I_u1s2(_bltz) | 339 | I_u1s2(_bltz) |
338 | I_u1s2(_bltzl) | 340 | I_u1s2(_bltzl) |
339 | I_u1u2s3(_bne) | 341 | I_u1u2s3(_bne) |
342 | I_u2s3u1(_cache) | ||
340 | I_u1u2u3(_dmfc0) | 343 | I_u1u2u3(_dmfc0) |
341 | I_u1u2u3(_dmtc0) | 344 | I_u1u2u3(_dmtc0) |
342 | I_u2u1s3(_daddiu) | 345 | I_u2u1s3(_daddiu) |
@@ -359,6 +362,7 @@ I_u2s3u1(_lw) | |||
359 | I_u1u2u3(_mfc0) | 362 | I_u1u2u3(_mfc0) |
360 | I_u1u2u3(_mtc0) | 363 | I_u1u2u3(_mtc0) |
361 | I_u2u1u3(_ori) | 364 | I_u2u1u3(_ori) |
365 | I_u2s3u1(_pref) | ||
362 | I_0(_rfe) | 366 | I_0(_rfe) |
363 | I_u2s3u1(_sc) | 367 | I_u2s3u1(_sc) |
364 | I_u2s3u1(_scd) | 368 | I_u2s3u1(_scd) |
@@ -555,6 +559,14 @@ uasm_il_beqzl(u32 **p, struct uasm_reloc **r, unsigned int reg, int lid) | |||
555 | } | 559 | } |
556 | 560 | ||
557 | void __cpuinit | 561 | void __cpuinit |
562 | uasm_il_bne(u32 **p, struct uasm_reloc **r, unsigned int reg1, | ||
563 | unsigned int reg2, int lid) | ||
564 | { | ||
565 | uasm_r_mips_pc16(r, *p, lid); | ||
566 | uasm_i_bne(p, reg1, reg2, 0); | ||
567 | } | ||
568 | |||
569 | void __cpuinit | ||
558 | uasm_il_bnez(u32 **p, struct uasm_reloc **r, unsigned int reg, int lid) | 570 | uasm_il_bnez(u32 **p, struct uasm_reloc **r, unsigned int reg, int lid) |
559 | { | 571 | { |
560 | uasm_r_mips_pc16(r, *p, lid); | 572 | uasm_r_mips_pc16(r, *p, lid); |
diff --git a/arch/mips/mm/uasm.h b/arch/mips/mm/uasm.h index fe0574f6e77d..0d6a66f32030 100644 --- a/arch/mips/mm/uasm.h +++ b/arch/mips/mm/uasm.h | |||
@@ -55,6 +55,7 @@ Ip_u1s2(_bgezl); | |||
55 | Ip_u1s2(_bltz); | 55 | Ip_u1s2(_bltz); |
56 | Ip_u1s2(_bltzl); | 56 | Ip_u1s2(_bltzl); |
57 | Ip_u1u2s3(_bne); | 57 | Ip_u1u2s3(_bne); |
58 | Ip_u2s3u1(_cache); | ||
58 | Ip_u1u2u3(_dmfc0); | 59 | Ip_u1u2u3(_dmfc0); |
59 | Ip_u1u2u3(_dmtc0); | 60 | Ip_u1u2u3(_dmtc0); |
60 | Ip_u2u1s3(_daddiu); | 61 | Ip_u2u1s3(_daddiu); |
@@ -77,6 +78,7 @@ Ip_u2s3u1(_lw); | |||
77 | Ip_u1u2u3(_mfc0); | 78 | Ip_u1u2u3(_mfc0); |
78 | Ip_u1u2u3(_mtc0); | 79 | Ip_u1u2u3(_mtc0); |
79 | Ip_u2u1u3(_ori); | 80 | Ip_u2u1u3(_ori); |
81 | Ip_u2s3u1(_pref); | ||
80 | Ip_0(_rfe); | 82 | Ip_0(_rfe); |
81 | Ip_u2s3u1(_sc); | 83 | Ip_u2s3u1(_sc); |
82 | Ip_u2s3u1(_scd); | 84 | Ip_u2s3u1(_scd); |
@@ -177,6 +179,8 @@ void uasm_il_bltz(u32 **p, struct uasm_reloc **r, unsigned int reg, int lid); | |||
177 | void uasm_il_b(u32 **p, struct uasm_reloc **r, int lid); | 179 | void uasm_il_b(u32 **p, struct uasm_reloc **r, int lid); |
178 | void uasm_il_beqz(u32 **p, struct uasm_reloc **r, unsigned int reg, int lid); | 180 | void uasm_il_beqz(u32 **p, struct uasm_reloc **r, unsigned int reg, int lid); |
179 | void uasm_il_beqzl(u32 **p, struct uasm_reloc **r, unsigned int reg, int lid); | 181 | void uasm_il_beqzl(u32 **p, struct uasm_reloc **r, unsigned int reg, int lid); |
182 | void uasm_il_bne(u32 **p, struct uasm_reloc **r, unsigned int reg1, | ||
183 | unsigned int reg2, int lid); | ||
180 | void uasm_il_bnez(u32 **p, struct uasm_reloc **r, unsigned int reg, int lid); | 184 | void uasm_il_bnez(u32 **p, struct uasm_reloc **r, unsigned int reg, int lid); |
181 | void uasm_il_bgezl(u32 **p, struct uasm_reloc **r, unsigned int reg, int lid); | 185 | void uasm_il_bgezl(u32 **p, struct uasm_reloc **r, unsigned int reg, int lid); |
182 | void uasm_il_bgez(u32 **p, struct uasm_reloc **r, unsigned int reg, int lid); | 186 | void uasm_il_bgez(u32 **p, struct uasm_reloc **r, unsigned int reg, int lid); |