Commit | Line | Data |
---|---|---|
fb2a27e7 TS |
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 | |
c0226306 | 9 | * Copyright (C) 2012 MIPS Technologies, Inc. |
fb2a27e7 TS |
10 | */ |
11 | #include <linux/init.h> | |
12 | #include <linux/kernel.h> | |
13 | #include <linux/sched.h> | |
631330f5 | 14 | #include <linux/smp.h> |
fb2a27e7 TS |
15 | #include <linux/mm.h> |
16 | #include <linux/module.h> | |
17 | #include <linux/proc_fs.h> | |
18 | ||
19 | #include <asm/bugs.h> | |
20 | #include <asm/cacheops.h> | |
69f24d17 | 21 | #include <asm/cpu-type.h> |
fb2a27e7 TS |
22 | #include <asm/inst.h> |
23 | #include <asm/io.h> | |
24 | #include <asm/page.h> | |
25 | #include <asm/pgtable.h> | |
26 | #include <asm/prefetch.h> | |
fb2a27e7 TS |
27 | #include <asm/bootinfo.h> |
28 | #include <asm/mipsregs.h> | |
29 | #include <asm/mmu_context.h> | |
30 | #include <asm/cpu.h> | |
31 | #include <asm/war.h> | |
32 | ||
33 | #ifdef CONFIG_SIBYTE_DMA_PAGEOPS | |
34 | #include <asm/sibyte/sb1250.h> | |
35 | #include <asm/sibyte/sb1250_regs.h> | |
36 | #include <asm/sibyte/sb1250_dma.h> | |
37 | #endif | |
38 | ||
3482d713 | 39 | #include <asm/uasm.h> |
fb2a27e7 TS |
40 | |
41 | /* Registers used in the assembled routines. */ | |
42 | #define ZERO 0 | |
43 | #define AT 2 | |
44 | #define A0 4 | |
45 | #define A1 5 | |
46 | #define A2 6 | |
47 | #define T0 8 | |
48 | #define T1 9 | |
49 | #define T2 10 | |
50 | #define T3 11 | |
51 | #define T9 25 | |
52 | #define RA 31 | |
53 | ||
54 | /* Handle labels (which must be positive integers). */ | |
55 | enum label_id { | |
56 | label_clear_nopref = 1, | |
57 | label_clear_pref, | |
58 | label_copy_nopref, | |
59 | label_copy_pref_both, | |
60 | label_copy_pref_store, | |
61 | }; | |
62 | ||
63 | UASM_L_LA(_clear_nopref) | |
64 | UASM_L_LA(_clear_pref) | |
65 | UASM_L_LA(_copy_nopref) | |
66 | UASM_L_LA(_copy_pref_both) | |
67 | UASM_L_LA(_copy_pref_store) | |
68 | ||
69 | /* We need one branch and therefore one relocation per target label. */ | |
078a55fc PG |
70 | static struct uasm_label labels[5]; |
71 | static struct uasm_reloc relocs[5]; | |
fb2a27e7 TS |
72 | |
73 | #define cpu_is_r4600_v1_x() ((read_c0_prid() & 0xfffffff0) == 0x00002010) | |
74 | #define cpu_is_r4600_v2_x() ((read_c0_prid() & 0xfffffff0) == 0x00002020) | |
75 | ||
078a55fc PG |
76 | static int pref_bias_clear_store; |
77 | static int pref_bias_copy_load; | |
78 | static int pref_bias_copy_store; | |
fb2a27e7 | 79 | |
078a55fc PG |
80 | static u32 pref_src_mode; |
81 | static u32 pref_dst_mode; | |
fb2a27e7 | 82 | |
078a55fc PG |
83 | static int clear_word_size; |
84 | static int copy_word_size; | |
fb2a27e7 | 85 | |
078a55fc PG |
86 | static int half_clear_loop_size; |
87 | static int half_copy_loop_size; | |
fb2a27e7 | 88 | |
078a55fc | 89 | static int cache_line_size; |
fb2a27e7 TS |
90 | #define cache_line_mask() (cache_line_size - 1) |
91 | ||
078a55fc | 92 | static inline void |
fb2a27e7 TS |
93 | pg_addiu(u32 **buf, unsigned int reg1, unsigned int reg2, unsigned int off) |
94 | { | |
95 | if (cpu_has_64bit_gp_regs && DADDI_WAR && r4k_daddiu_bug()) { | |
96 | if (off > 0x7fff) { | |
97 | uasm_i_lui(buf, T9, uasm_rel_hi(off)); | |
98 | uasm_i_addiu(buf, T9, T9, uasm_rel_lo(off)); | |
99 | } else | |
100 | uasm_i_addiu(buf, T9, ZERO, off); | |
101 | uasm_i_daddu(buf, reg1, reg2, T9); | |
102 | } else { | |
103 | if (off > 0x7fff) { | |
104 | uasm_i_lui(buf, T9, uasm_rel_hi(off)); | |
105 | uasm_i_addiu(buf, T9, T9, uasm_rel_lo(off)); | |
106 | UASM_i_ADDU(buf, reg1, reg2, T9); | |
107 | } else | |
108 | UASM_i_ADDIU(buf, reg1, reg2, off); | |
109 | } | |
110 | } | |
111 | ||
078a55fc | 112 | static void set_prefetch_parameters(void) |
fb2a27e7 TS |
113 | { |
114 | if (cpu_has_64bit_gp_regs || cpu_has_64bit_zero_reg) | |
115 | clear_word_size = 8; | |
116 | else | |
117 | clear_word_size = 4; | |
118 | ||
119 | if (cpu_has_64bit_gp_regs) | |
120 | copy_word_size = 8; | |
121 | else | |
122 | copy_word_size = 4; | |
123 | ||
124 | /* | |
125 | * The pref's used here are using "streaming" hints, which cause the | |
126 | * copied data to be kicked out of the cache sooner. A page copy often | |
127 | * ends up copying a lot more data than is commonly used, so this seems | |
128 | * to make sense in terms of reducing cache pollution, but I've no real | |
129 | * performance data to back this up. | |
130 | */ | |
131 | if (cpu_has_prefetch) { | |
132 | /* | |
133 | * XXX: Most prefetch bias values in here are based on | |
134 | * guesswork. | |
135 | */ | |
136 | cache_line_size = cpu_dcache_line_size(); | |
137 | switch (current_cpu_type()) { | |
a644b277 | 138 | case CPU_R5500: |
fb2a27e7 | 139 | case CPU_TX49XX: |
a644b277 | 140 | /* These processors only support the Pref_Load. */ |
fb2a27e7 TS |
141 | pref_bias_copy_load = 256; |
142 | break; | |
143 | ||
fb2a27e7 TS |
144 | case CPU_R10000: |
145 | case CPU_R12000: | |
146 | case CPU_R14000: | |
147 | /* | |
148 | * Those values have been experimentally tuned for an | |
149 | * Origin 200. | |
150 | */ | |
151 | pref_bias_clear_store = 512; | |
152 | pref_bias_copy_load = 256; | |
153 | pref_bias_copy_store = 256; | |
154 | pref_src_mode = Pref_LoadStreamed; | |
155 | pref_dst_mode = Pref_StoreStreamed; | |
156 | break; | |
157 | ||
158 | case CPU_SB1: | |
159 | case CPU_SB1A: | |
160 | pref_bias_clear_store = 128; | |
161 | pref_bias_copy_load = 128; | |
162 | pref_bias_copy_store = 128; | |
163 | /* | |
164 | * SB1 pass1 Pref_LoadStreamed/Pref_StoreStreamed | |
165 | * hints are broken. | |
166 | */ | |
167 | if (current_cpu_type() == CPU_SB1 && | |
168 | (current_cpu_data.processor_id & 0xff) < 0x02) { | |
169 | pref_src_mode = Pref_Load; | |
170 | pref_dst_mode = Pref_Store; | |
171 | } else { | |
172 | pref_src_mode = Pref_LoadStreamed; | |
173 | pref_dst_mode = Pref_StoreStreamed; | |
174 | } | |
175 | break; | |
176 | ||
177 | default: | |
178 | pref_bias_clear_store = 128; | |
179 | pref_bias_copy_load = 256; | |
180 | pref_bias_copy_store = 128; | |
181 | pref_src_mode = Pref_LoadStreamed; | |
182 | pref_dst_mode = Pref_PrepareForStore; | |
183 | break; | |
184 | } | |
185 | } else { | |
186 | if (cpu_has_cache_cdex_s) | |
187 | cache_line_size = cpu_scache_line_size(); | |
188 | else if (cpu_has_cache_cdex_p) | |
189 | cache_line_size = cpu_dcache_line_size(); | |
190 | } | |
191 | /* | |
192 | * Too much unrolling will overflow the available space in | |
14defd90 | 193 | * clear_space_array / copy_page_array. |
fb2a27e7 | 194 | */ |
14defd90 | 195 | half_clear_loop_size = min(16 * clear_word_size, |
fb2a27e7 TS |
196 | max(cache_line_size >> 1, |
197 | 4 * clear_word_size)); | |
14defd90 | 198 | half_copy_loop_size = min(16 * copy_word_size, |
fb2a27e7 TS |
199 | max(cache_line_size >> 1, |
200 | 4 * copy_word_size)); | |
201 | } | |
202 | ||
078a55fc | 203 | static void build_clear_store(u32 **buf, int off) |
fb2a27e7 TS |
204 | { |
205 | if (cpu_has_64bit_gp_regs || cpu_has_64bit_zero_reg) { | |
206 | uasm_i_sd(buf, ZERO, off, A0); | |
207 | } else { | |
208 | uasm_i_sw(buf, ZERO, off, A0); | |
209 | } | |
210 | } | |
211 | ||
078a55fc | 212 | static inline void build_clear_pref(u32 **buf, int off) |
fb2a27e7 TS |
213 | { |
214 | if (off & cache_line_mask()) | |
215 | return; | |
216 | ||
217 | if (pref_bias_clear_store) { | |
218 | uasm_i_pref(buf, pref_dst_mode, pref_bias_clear_store + off, | |
219 | A0); | |
14defd90 TB |
220 | } else if (cache_line_size == (half_clear_loop_size << 1)) { |
221 | if (cpu_has_cache_cdex_s) { | |
222 | uasm_i_cache(buf, Create_Dirty_Excl_SD, off, A0); | |
223 | } else if (cpu_has_cache_cdex_p) { | |
224 | if (R4600_V1_HIT_CACHEOP_WAR && cpu_is_r4600_v1_x()) { | |
225 | uasm_i_nop(buf); | |
226 | uasm_i_nop(buf); | |
227 | uasm_i_nop(buf); | |
228 | uasm_i_nop(buf); | |
229 | } | |
fb2a27e7 | 230 | |
14defd90 TB |
231 | if (R4600_V2_HIT_CACHEOP_WAR && cpu_is_r4600_v2_x()) |
232 | uasm_i_lw(buf, ZERO, ZERO, AT); | |
fb2a27e7 | 233 | |
14defd90 TB |
234 | uasm_i_cache(buf, Create_Dirty_Excl_D, off, A0); |
235 | } | |
fc192e50 | 236 | } |
fb2a27e7 TS |
237 | } |
238 | ||
c0226306 SH |
239 | extern u32 __clear_page_start; |
240 | extern u32 __clear_page_end; | |
241 | extern u32 __copy_page_start; | |
242 | extern u32 __copy_page_end; | |
243 | ||
078a55fc | 244 | void build_clear_page(void) |
fb2a27e7 TS |
245 | { |
246 | int off; | |
c0226306 | 247 | u32 *buf = &__clear_page_start; |
fb2a27e7 TS |
248 | struct uasm_label *l = labels; |
249 | struct uasm_reloc *r = relocs; | |
250 | int i; | |
8759934e HC |
251 | static atomic_t run_once = ATOMIC_INIT(0); |
252 | ||
253 | if (atomic_xchg(&run_once, 1)) { | |
254 | return; | |
255 | } | |
fb2a27e7 TS |
256 | |
257 | memset(labels, 0, sizeof(labels)); | |
258 | memset(relocs, 0, sizeof(relocs)); | |
259 | ||
260 | set_prefetch_parameters(); | |
261 | ||
262 | /* | |
263 | * This algorithm makes the following assumptions: | |
264 | * - The prefetch bias is a multiple of 2 words. | |
265 | * - The prefetch bias is less than one page. | |
266 | */ | |
267 | BUG_ON(pref_bias_clear_store % (2 * clear_word_size)); | |
268 | BUG_ON(PAGE_SIZE < pref_bias_clear_store); | |
269 | ||
270 | off = PAGE_SIZE - pref_bias_clear_store; | |
271 | if (off > 0xffff || !pref_bias_clear_store) | |
272 | pg_addiu(&buf, A2, A0, off); | |
273 | else | |
274 | uasm_i_ori(&buf, A2, A0, off); | |
275 | ||
276 | if (R4600_V2_HIT_CACHEOP_WAR && cpu_is_r4600_v2_x()) | |
277 | uasm_i_lui(&buf, AT, 0xa000); | |
278 | ||
cd9da13d | 279 | off = cache_line_size ? min(8, pref_bias_clear_store / cache_line_size) |
70342287 | 280 | * cache_line_size : 0; |
fb2a27e7 TS |
281 | while (off) { |
282 | build_clear_pref(&buf, -off); | |
283 | off -= cache_line_size; | |
284 | } | |
285 | uasm_l_clear_pref(&l, buf); | |
286 | do { | |
287 | build_clear_pref(&buf, off); | |
288 | build_clear_store(&buf, off); | |
289 | off += clear_word_size; | |
290 | } while (off < half_clear_loop_size); | |
291 | pg_addiu(&buf, A0, A0, 2 * off); | |
292 | off = -off; | |
293 | do { | |
294 | build_clear_pref(&buf, off); | |
295 | if (off == -clear_word_size) | |
296 | uasm_il_bne(&buf, &r, A0, A2, label_clear_pref); | |
297 | build_clear_store(&buf, off); | |
298 | off += clear_word_size; | |
299 | } while (off < 0); | |
300 | ||
301 | if (pref_bias_clear_store) { | |
302 | pg_addiu(&buf, A2, A0, pref_bias_clear_store); | |
303 | uasm_l_clear_nopref(&l, buf); | |
304 | off = 0; | |
305 | do { | |
306 | build_clear_store(&buf, off); | |
307 | off += clear_word_size; | |
308 | } while (off < half_clear_loop_size); | |
309 | pg_addiu(&buf, A0, A0, 2 * off); | |
310 | off = -off; | |
311 | do { | |
312 | if (off == -clear_word_size) | |
313 | uasm_il_bne(&buf, &r, A0, A2, | |
314 | label_clear_nopref); | |
315 | build_clear_store(&buf, off); | |
316 | off += clear_word_size; | |
317 | } while (off < 0); | |
318 | } | |
319 | ||
320 | uasm_i_jr(&buf, RA); | |
321 | uasm_i_nop(&buf); | |
322 | ||
c0226306 | 323 | BUG_ON(buf > &__clear_page_end); |
fb2a27e7 TS |
324 | |
325 | uasm_resolve_relocs(relocs, labels); | |
326 | ||
327 | pr_debug("Synthesized clear page handler (%u instructions).\n", | |
c0226306 | 328 | (u32)(buf - &__clear_page_start)); |
fb2a27e7 TS |
329 | |
330 | pr_debug("\t.set push\n"); | |
331 | pr_debug("\t.set noreorder\n"); | |
c0226306 SH |
332 | for (i = 0; i < (buf - &__clear_page_start); i++) |
333 | pr_debug("\t.word 0x%08x\n", (&__clear_page_start)[i]); | |
fb2a27e7 TS |
334 | pr_debug("\t.set pop\n"); |
335 | } | |
336 | ||
078a55fc | 337 | static void build_copy_load(u32 **buf, int reg, int off) |
fb2a27e7 TS |
338 | { |
339 | if (cpu_has_64bit_gp_regs) { | |
340 | uasm_i_ld(buf, reg, off, A1); | |
341 | } else { | |
342 | uasm_i_lw(buf, reg, off, A1); | |
343 | } | |
344 | } | |
345 | ||
078a55fc | 346 | static void build_copy_store(u32 **buf, int reg, int off) |
fb2a27e7 TS |
347 | { |
348 | if (cpu_has_64bit_gp_regs) { | |
349 | uasm_i_sd(buf, reg, off, A0); | |
350 | } else { | |
351 | uasm_i_sw(buf, reg, off, A0); | |
352 | } | |
353 | } | |
354 | ||
355 | static inline void build_copy_load_pref(u32 **buf, int off) | |
356 | { | |
357 | if (off & cache_line_mask()) | |
358 | return; | |
359 | ||
360 | if (pref_bias_copy_load) | |
361 | uasm_i_pref(buf, pref_src_mode, pref_bias_copy_load + off, A1); | |
362 | } | |
363 | ||
364 | static inline void build_copy_store_pref(u32 **buf, int off) | |
365 | { | |
366 | if (off & cache_line_mask()) | |
367 | return; | |
368 | ||
369 | if (pref_bias_copy_store) { | |
370 | uasm_i_pref(buf, pref_dst_mode, pref_bias_copy_store + off, | |
371 | A0); | |
14defd90 TB |
372 | } else if (cache_line_size == (half_copy_loop_size << 1)) { |
373 | if (cpu_has_cache_cdex_s) { | |
374 | uasm_i_cache(buf, Create_Dirty_Excl_SD, off, A0); | |
375 | } else if (cpu_has_cache_cdex_p) { | |
376 | if (R4600_V1_HIT_CACHEOP_WAR && cpu_is_r4600_v1_x()) { | |
377 | uasm_i_nop(buf); | |
378 | uasm_i_nop(buf); | |
379 | uasm_i_nop(buf); | |
380 | uasm_i_nop(buf); | |
381 | } | |
fb2a27e7 | 382 | |
14defd90 TB |
383 | if (R4600_V2_HIT_CACHEOP_WAR && cpu_is_r4600_v2_x()) |
384 | uasm_i_lw(buf, ZERO, ZERO, AT); | |
fb2a27e7 | 385 | |
14defd90 TB |
386 | uasm_i_cache(buf, Create_Dirty_Excl_D, off, A0); |
387 | } | |
fb2a27e7 TS |
388 | } |
389 | } | |
390 | ||
078a55fc | 391 | void build_copy_page(void) |
fb2a27e7 TS |
392 | { |
393 | int off; | |
c0226306 | 394 | u32 *buf = &__copy_page_start; |
fb2a27e7 TS |
395 | struct uasm_label *l = labels; |
396 | struct uasm_reloc *r = relocs; | |
397 | int i; | |
8759934e HC |
398 | static atomic_t run_once = ATOMIC_INIT(0); |
399 | ||
400 | if (atomic_xchg(&run_once, 1)) { | |
401 | return; | |
402 | } | |
fb2a27e7 TS |
403 | |
404 | memset(labels, 0, sizeof(labels)); | |
405 | memset(relocs, 0, sizeof(relocs)); | |
406 | ||
407 | set_prefetch_parameters(); | |
408 | ||
409 | /* | |
410 | * This algorithm makes the following assumptions: | |
411 | * - All prefetch biases are multiples of 8 words. | |
412 | * - The prefetch biases are less than one page. | |
413 | * - The store prefetch bias isn't greater than the load | |
414 | * prefetch bias. | |
415 | */ | |
416 | BUG_ON(pref_bias_copy_load % (8 * copy_word_size)); | |
417 | BUG_ON(pref_bias_copy_store % (8 * copy_word_size)); | |
418 | BUG_ON(PAGE_SIZE < pref_bias_copy_load); | |
419 | BUG_ON(pref_bias_copy_store > pref_bias_copy_load); | |
420 | ||
421 | off = PAGE_SIZE - pref_bias_copy_load; | |
422 | if (off > 0xffff || !pref_bias_copy_load) | |
423 | pg_addiu(&buf, A2, A0, off); | |
424 | else | |
425 | uasm_i_ori(&buf, A2, A0, off); | |
426 | ||
427 | if (R4600_V2_HIT_CACHEOP_WAR && cpu_is_r4600_v2_x()) | |
428 | uasm_i_lui(&buf, AT, 0xa000); | |
429 | ||
cd9da13d | 430 | off = cache_line_size ? min(8, pref_bias_copy_load / cache_line_size) * |
70342287 | 431 | cache_line_size : 0; |
fb2a27e7 TS |
432 | while (off) { |
433 | build_copy_load_pref(&buf, -off); | |
434 | off -= cache_line_size; | |
435 | } | |
7bd0fea2 | 436 | off = cache_line_size ? min(8, pref_bias_copy_store / cache_line_size) * |
70342287 | 437 | cache_line_size : 0; |
fb2a27e7 TS |
438 | while (off) { |
439 | build_copy_store_pref(&buf, -off); | |
440 | off -= cache_line_size; | |
441 | } | |
442 | uasm_l_copy_pref_both(&l, buf); | |
443 | do { | |
444 | build_copy_load_pref(&buf, off); | |
445 | build_copy_load(&buf, T0, off); | |
446 | build_copy_load_pref(&buf, off + copy_word_size); | |
447 | build_copy_load(&buf, T1, off + copy_word_size); | |
448 | build_copy_load_pref(&buf, off + 2 * copy_word_size); | |
449 | build_copy_load(&buf, T2, off + 2 * copy_word_size); | |
450 | build_copy_load_pref(&buf, off + 3 * copy_word_size); | |
451 | build_copy_load(&buf, T3, off + 3 * copy_word_size); | |
452 | build_copy_store_pref(&buf, off); | |
453 | build_copy_store(&buf, T0, off); | |
454 | build_copy_store_pref(&buf, off + copy_word_size); | |
455 | build_copy_store(&buf, T1, off + copy_word_size); | |
456 | build_copy_store_pref(&buf, off + 2 * copy_word_size); | |
457 | build_copy_store(&buf, T2, off + 2 * copy_word_size); | |
458 | build_copy_store_pref(&buf, off + 3 * copy_word_size); | |
459 | build_copy_store(&buf, T3, off + 3 * copy_word_size); | |
460 | off += 4 * copy_word_size; | |
461 | } while (off < half_copy_loop_size); | |
462 | pg_addiu(&buf, A1, A1, 2 * off); | |
463 | pg_addiu(&buf, A0, A0, 2 * off); | |
464 | off = -off; | |
465 | do { | |
466 | build_copy_load_pref(&buf, off); | |
467 | build_copy_load(&buf, T0, off); | |
468 | build_copy_load_pref(&buf, off + copy_word_size); | |
469 | build_copy_load(&buf, T1, off + copy_word_size); | |
470 | build_copy_load_pref(&buf, off + 2 * copy_word_size); | |
471 | build_copy_load(&buf, T2, off + 2 * copy_word_size); | |
472 | build_copy_load_pref(&buf, off + 3 * copy_word_size); | |
473 | build_copy_load(&buf, T3, off + 3 * copy_word_size); | |
474 | build_copy_store_pref(&buf, off); | |
475 | build_copy_store(&buf, T0, off); | |
476 | build_copy_store_pref(&buf, off + copy_word_size); | |
477 | build_copy_store(&buf, T1, off + copy_word_size); | |
478 | build_copy_store_pref(&buf, off + 2 * copy_word_size); | |
479 | build_copy_store(&buf, T2, off + 2 * copy_word_size); | |
480 | build_copy_store_pref(&buf, off + 3 * copy_word_size); | |
481 | if (off == -(4 * copy_word_size)) | |
482 | uasm_il_bne(&buf, &r, A2, A0, label_copy_pref_both); | |
483 | build_copy_store(&buf, T3, off + 3 * copy_word_size); | |
484 | off += 4 * copy_word_size; | |
485 | } while (off < 0); | |
486 | ||
487 | if (pref_bias_copy_load - pref_bias_copy_store) { | |
488 | pg_addiu(&buf, A2, A0, | |
489 | pref_bias_copy_load - pref_bias_copy_store); | |
490 | uasm_l_copy_pref_store(&l, buf); | |
491 | off = 0; | |
492 | do { | |
493 | build_copy_load(&buf, T0, off); | |
494 | build_copy_load(&buf, T1, off + copy_word_size); | |
495 | build_copy_load(&buf, T2, off + 2 * copy_word_size); | |
496 | build_copy_load(&buf, T3, off + 3 * copy_word_size); | |
497 | build_copy_store_pref(&buf, off); | |
498 | build_copy_store(&buf, T0, off); | |
499 | build_copy_store_pref(&buf, off + copy_word_size); | |
500 | build_copy_store(&buf, T1, off + copy_word_size); | |
501 | build_copy_store_pref(&buf, off + 2 * copy_word_size); | |
502 | build_copy_store(&buf, T2, off + 2 * copy_word_size); | |
503 | build_copy_store_pref(&buf, off + 3 * copy_word_size); | |
504 | build_copy_store(&buf, T3, off + 3 * copy_word_size); | |
505 | off += 4 * copy_word_size; | |
506 | } while (off < half_copy_loop_size); | |
507 | pg_addiu(&buf, A1, A1, 2 * off); | |
508 | pg_addiu(&buf, A0, A0, 2 * off); | |
509 | off = -off; | |
510 | do { | |
511 | build_copy_load(&buf, T0, off); | |
512 | build_copy_load(&buf, T1, off + copy_word_size); | |
513 | build_copy_load(&buf, T2, off + 2 * copy_word_size); | |
514 | build_copy_load(&buf, T3, off + 3 * copy_word_size); | |
515 | build_copy_store_pref(&buf, off); | |
516 | build_copy_store(&buf, T0, off); | |
517 | build_copy_store_pref(&buf, off + copy_word_size); | |
518 | build_copy_store(&buf, T1, off + copy_word_size); | |
519 | build_copy_store_pref(&buf, off + 2 * copy_word_size); | |
520 | build_copy_store(&buf, T2, off + 2 * copy_word_size); | |
521 | build_copy_store_pref(&buf, off + 3 * copy_word_size); | |
522 | if (off == -(4 * copy_word_size)) | |
523 | uasm_il_bne(&buf, &r, A2, A0, | |
524 | label_copy_pref_store); | |
525 | build_copy_store(&buf, T3, off + 3 * copy_word_size); | |
526 | off += 4 * copy_word_size; | |
527 | } while (off < 0); | |
528 | } | |
529 | ||
530 | if (pref_bias_copy_store) { | |
531 | pg_addiu(&buf, A2, A0, pref_bias_copy_store); | |
532 | uasm_l_copy_nopref(&l, buf); | |
533 | off = 0; | |
534 | do { | |
535 | build_copy_load(&buf, T0, off); | |
536 | build_copy_load(&buf, T1, off + copy_word_size); | |
537 | build_copy_load(&buf, T2, off + 2 * copy_word_size); | |
538 | build_copy_load(&buf, T3, off + 3 * copy_word_size); | |
539 | build_copy_store(&buf, T0, off); | |
540 | build_copy_store(&buf, T1, off + copy_word_size); | |
541 | build_copy_store(&buf, T2, off + 2 * copy_word_size); | |
542 | build_copy_store(&buf, T3, off + 3 * copy_word_size); | |
543 | off += 4 * copy_word_size; | |
544 | } while (off < half_copy_loop_size); | |
545 | pg_addiu(&buf, A1, A1, 2 * off); | |
546 | pg_addiu(&buf, A0, A0, 2 * off); | |
547 | off = -off; | |
548 | do { | |
549 | build_copy_load(&buf, T0, off); | |
550 | build_copy_load(&buf, T1, off + copy_word_size); | |
551 | build_copy_load(&buf, T2, off + 2 * copy_word_size); | |
552 | build_copy_load(&buf, T3, off + 3 * copy_word_size); | |
553 | build_copy_store(&buf, T0, off); | |
554 | build_copy_store(&buf, T1, off + copy_word_size); | |
555 | build_copy_store(&buf, T2, off + 2 * copy_word_size); | |
556 | if (off == -(4 * copy_word_size)) | |
557 | uasm_il_bne(&buf, &r, A2, A0, | |
558 | label_copy_nopref); | |
559 | build_copy_store(&buf, T3, off + 3 * copy_word_size); | |
560 | off += 4 * copy_word_size; | |
561 | } while (off < 0); | |
562 | } | |
563 | ||
564 | uasm_i_jr(&buf, RA); | |
565 | uasm_i_nop(&buf); | |
566 | ||
c0226306 | 567 | BUG_ON(buf > &__copy_page_end); |
fb2a27e7 TS |
568 | |
569 | uasm_resolve_relocs(relocs, labels); | |
570 | ||
571 | pr_debug("Synthesized copy page handler (%u instructions).\n", | |
c0226306 | 572 | (u32)(buf - &__copy_page_start)); |
fb2a27e7 TS |
573 | |
574 | pr_debug("\t.set push\n"); | |
575 | pr_debug("\t.set noreorder\n"); | |
c0226306 SH |
576 | for (i = 0; i < (buf - &__copy_page_start); i++) |
577 | pr_debug("\t.word 0x%08x\n", (&__copy_page_start)[i]); | |
fb2a27e7 TS |
578 | pr_debug("\t.set pop\n"); |
579 | } | |
580 | ||
581 | #ifdef CONFIG_SIBYTE_DMA_PAGEOPS | |
c0226306 SH |
582 | extern void clear_page_cpu(void *page); |
583 | extern void copy_page_cpu(void *to, void *from); | |
fb2a27e7 TS |
584 | |
585 | /* | |
586 | * Pad descriptors to cacheline, since each is exclusively owned by a | |
587 | * particular CPU. | |
588 | */ | |
589 | struct dmadscr { | |
590 | u64 dscr_a; | |
591 | u64 dscr_b; | |
592 | u64 pad_a; | |
593 | u64 pad_b; | |
594 | } ____cacheline_aligned_in_smp page_descr[DM_NUM_CHANNELS]; | |
595 | ||
596 | void sb1_dma_init(void) | |
597 | { | |
598 | int i; | |
599 | ||
600 | for (i = 0; i < DM_NUM_CHANNELS; i++) { | |
601 | const u64 base_val = CPHYSADDR((unsigned long)&page_descr[i]) | | |
602 | V_DM_DSCR_BASE_RINGSZ(1); | |
603 | void *base_reg = IOADDR(A_DM_REGISTER(i, R_DM_DSCR_BASE)); | |
604 | ||
605 | __raw_writeq(base_val, base_reg); | |
606 | __raw_writeq(base_val | M_DM_DSCR_BASE_RESET, base_reg); | |
607 | __raw_writeq(base_val | M_DM_DSCR_BASE_ENABL, base_reg); | |
608 | } | |
609 | } | |
610 | ||
611 | void clear_page(void *page) | |
612 | { | |
613 | u64 to_phys = CPHYSADDR((unsigned long)page); | |
614 | unsigned int cpu = smp_processor_id(); | |
615 | ||
616 | /* if the page is not in KSEG0, use old way */ | |
617 | if ((long)KSEGX((unsigned long)page) != (long)CKSEG0) | |
618 | return clear_page_cpu(page); | |
619 | ||
620 | page_descr[cpu].dscr_a = to_phys | M_DM_DSCRA_ZERO_MEM | | |
621 | M_DM_DSCRA_L2C_DEST | M_DM_DSCRA_INTERRUPT; | |
622 | page_descr[cpu].dscr_b = V_DM_DSCRB_SRC_LENGTH(PAGE_SIZE); | |
623 | __raw_writeq(1, IOADDR(A_DM_REGISTER(cpu, R_DM_DSCR_COUNT))); | |
624 | ||
625 | /* | |
626 | * Don't really want to do it this way, but there's no | |
627 | * reliable way to delay completion detection. | |
628 | */ | |
629 | while (!(__raw_readq(IOADDR(A_DM_REGISTER(cpu, R_DM_DSCR_BASE_DEBUG))) | |
630 | & M_DM_DSCR_BASE_INTERRUPT)) | |
631 | ; | |
632 | __raw_readq(IOADDR(A_DM_REGISTER(cpu, R_DM_DSCR_BASE))); | |
633 | } | |
634 | ||
635 | void copy_page(void *to, void *from) | |
636 | { | |
637 | u64 from_phys = CPHYSADDR((unsigned long)from); | |
638 | u64 to_phys = CPHYSADDR((unsigned long)to); | |
639 | unsigned int cpu = smp_processor_id(); | |
640 | ||
641 | /* if any page is not in KSEG0, use old way */ | |
642 | if ((long)KSEGX((unsigned long)to) != (long)CKSEG0 | |
643 | || (long)KSEGX((unsigned long)from) != (long)CKSEG0) | |
644 | return copy_page_cpu(to, from); | |
645 | ||
646 | page_descr[cpu].dscr_a = to_phys | M_DM_DSCRA_L2C_DEST | | |
647 | M_DM_DSCRA_INTERRUPT; | |
648 | page_descr[cpu].dscr_b = from_phys | V_DM_DSCRB_SRC_LENGTH(PAGE_SIZE); | |
649 | __raw_writeq(1, IOADDR(A_DM_REGISTER(cpu, R_DM_DSCR_COUNT))); | |
650 | ||
651 | /* | |
652 | * Don't really want to do it this way, but there's no | |
653 | * reliable way to delay completion detection. | |
654 | */ | |
655 | while (!(__raw_readq(IOADDR(A_DM_REGISTER(cpu, R_DM_DSCR_BASE_DEBUG))) | |
656 | & M_DM_DSCR_BASE_INTERRUPT)) | |
657 | ; | |
658 | __raw_readq(IOADDR(A_DM_REGISTER(cpu, R_DM_DSCR_BASE))); | |
659 | } | |
660 | ||
661 | #endif /* CONFIG_SIBYTE_DMA_PAGEOPS */ |