Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 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 | */ |
| 8 | #include <linux/init.h> |
| 9 | #include <linux/kernel.h> |
| 10 | #include <linux/sched.h> |
| 11 | #include <linux/mm.h> |
| 12 | #include <linux/module.h> |
| 13 | #include <linux/proc_fs.h> |
| 14 | |
| 15 | #include <asm/cacheops.h> |
| 16 | #include <asm/inst.h> |
| 17 | #include <asm/io.h> |
| 18 | #include <asm/page.h> |
| 19 | #include <asm/pgtable.h> |
| 20 | #include <asm/prefetch.h> |
| 21 | #include <asm/system.h> |
| 22 | #include <asm/bootinfo.h> |
| 23 | #include <asm/mipsregs.h> |
| 24 | #include <asm/mmu_context.h> |
| 25 | #include <asm/cpu.h> |
| 26 | #include <asm/war.h> |
| 27 | |
Thiemo Seufer | 330cfe0 | 2005-09-01 18:33:58 +0000 | [diff] [blame] | 28 | #define half_scache_line_size() (cpu_scache_line_size() >> 1) |
| 29 | #define cpu_is_r4600_v1_x() ((read_c0_prid() & 0xfffffff0) == 0x00002010) |
| 30 | #define cpu_is_r4600_v2_x() ((read_c0_prid() & 0xfffffff0) == 0x00002020) |
| 31 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 32 | |
| 33 | /* |
| 34 | * Maximum sizes: |
| 35 | * |
| 36 | * R4000 128 bytes S-cache: 0x58 bytes |
| 37 | * R4600 v1.7: 0x5c bytes |
| 38 | * R4600 v2.0: 0x60 bytes |
| 39 | * With prefetching, 16 byte strides 0xa0 bytes |
| 40 | */ |
| 41 | |
| 42 | static unsigned int clear_page_array[0x130 / 4]; |
| 43 | |
| 44 | void clear_page(void * page) __attribute__((alias("clear_page_array"))); |
| 45 | |
| 46 | EXPORT_SYMBOL(clear_page); |
| 47 | |
| 48 | /* |
| 49 | * Maximum sizes: |
| 50 | * |
| 51 | * R4000 128 bytes S-cache: 0x11c bytes |
| 52 | * R4600 v1.7: 0x080 bytes |
| 53 | * R4600 v2.0: 0x07c bytes |
| 54 | * With prefetching, 16 byte strides 0x0b8 bytes |
| 55 | */ |
| 56 | static unsigned int copy_page_array[0x148 / 4]; |
| 57 | |
| 58 | void copy_page(void *to, void *from) __attribute__((alias("copy_page_array"))); |
| 59 | |
| 60 | EXPORT_SYMBOL(copy_page); |
| 61 | |
| 62 | /* |
| 63 | * This is suboptimal for 32-bit kernels; we assume that R10000 is only used |
| 64 | * with 64-bit kernels. The prefetch offsets have been experimentally tuned |
| 65 | * an Origin 200. |
| 66 | */ |
| 67 | static int pref_offset_clear __initdata = 512; |
| 68 | static int pref_offset_copy __initdata = 256; |
| 69 | |
| 70 | static unsigned int pref_src_mode __initdata; |
| 71 | static unsigned int pref_dst_mode __initdata; |
| 72 | |
| 73 | static int load_offset __initdata; |
| 74 | static int store_offset __initdata; |
| 75 | |
| 76 | static unsigned int __initdata *dest, *epc; |
| 77 | |
| 78 | static unsigned int instruction_pending; |
| 79 | static union mips_instruction delayed_mi; |
| 80 | |
| 81 | static void __init emit_instruction(union mips_instruction mi) |
| 82 | { |
| 83 | if (instruction_pending) |
| 84 | *epc++ = delayed_mi.word; |
| 85 | |
| 86 | instruction_pending = 1; |
| 87 | delayed_mi = mi; |
| 88 | } |
| 89 | |
| 90 | static inline void flush_delay_slot_or_nop(void) |
| 91 | { |
| 92 | if (instruction_pending) { |
| 93 | *epc++ = delayed_mi.word; |
| 94 | instruction_pending = 0; |
| 95 | return; |
| 96 | } |
| 97 | |
| 98 | *epc++ = 0; |
| 99 | } |
| 100 | |
| 101 | static inline unsigned int *label(void) |
| 102 | { |
| 103 | if (instruction_pending) { |
| 104 | *epc++ = delayed_mi.word; |
| 105 | instruction_pending = 0; |
| 106 | } |
| 107 | |
| 108 | return epc; |
| 109 | } |
| 110 | |
| 111 | static inline void build_insn_word(unsigned int word) |
| 112 | { |
| 113 | union mips_instruction mi; |
| 114 | |
| 115 | mi.word = word; |
| 116 | |
| 117 | emit_instruction(mi); |
| 118 | } |
| 119 | |
| 120 | static inline void build_nop(void) |
| 121 | { |
| 122 | build_insn_word(0); /* nop */ |
| 123 | } |
| 124 | |
| 125 | static inline void build_src_pref(int advance) |
| 126 | { |
Atsushi Nemoto | de862b4 | 2006-03-17 12:59:22 +0900 | [diff] [blame] | 127 | if (!(load_offset & (cpu_dcache_line_size() - 1)) && advance) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 128 | union mips_instruction mi; |
| 129 | |
| 130 | mi.i_format.opcode = pref_op; |
| 131 | mi.i_format.rs = 5; /* $a1 */ |
| 132 | mi.i_format.rt = pref_src_mode; |
| 133 | mi.i_format.simmediate = load_offset + advance; |
| 134 | |
| 135 | emit_instruction(mi); |
| 136 | } |
| 137 | } |
| 138 | |
| 139 | static inline void __build_load_reg(int reg) |
| 140 | { |
| 141 | union mips_instruction mi; |
| 142 | unsigned int width; |
| 143 | |
| 144 | if (cpu_has_64bit_gp_regs) { |
| 145 | mi.i_format.opcode = ld_op; |
| 146 | width = 8; |
| 147 | } else { |
| 148 | mi.i_format.opcode = lw_op; |
| 149 | width = 4; |
| 150 | } |
| 151 | mi.i_format.rs = 5; /* $a1 */ |
| 152 | mi.i_format.rt = reg; /* $reg */ |
| 153 | mi.i_format.simmediate = load_offset; |
| 154 | |
| 155 | load_offset += width; |
| 156 | emit_instruction(mi); |
| 157 | } |
| 158 | |
| 159 | static inline void build_load_reg(int reg) |
| 160 | { |
| 161 | if (cpu_has_prefetch) |
| 162 | build_src_pref(pref_offset_copy); |
| 163 | |
| 164 | __build_load_reg(reg); |
| 165 | } |
| 166 | |
| 167 | static inline void build_dst_pref(int advance) |
| 168 | { |
Atsushi Nemoto | de862b4 | 2006-03-17 12:59:22 +0900 | [diff] [blame] | 169 | if (!(store_offset & (cpu_dcache_line_size() - 1)) && advance) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 170 | union mips_instruction mi; |
| 171 | |
| 172 | mi.i_format.opcode = pref_op; |
| 173 | mi.i_format.rs = 4; /* $a0 */ |
| 174 | mi.i_format.rt = pref_dst_mode; |
| 175 | mi.i_format.simmediate = store_offset + advance; |
| 176 | |
| 177 | emit_instruction(mi); |
| 178 | } |
| 179 | } |
| 180 | |
| 181 | static inline void build_cdex_s(void) |
| 182 | { |
| 183 | union mips_instruction mi; |
| 184 | |
| 185 | if ((store_offset & (cpu_scache_line_size() - 1))) |
| 186 | return; |
| 187 | |
| 188 | mi.c_format.opcode = cache_op; |
| 189 | mi.c_format.rs = 4; /* $a0 */ |
| 190 | mi.c_format.c_op = 3; /* Create Dirty Exclusive */ |
| 191 | mi.c_format.cache = 3; /* Secondary Data Cache */ |
| 192 | mi.c_format.simmediate = store_offset; |
| 193 | |
| 194 | emit_instruction(mi); |
| 195 | } |
| 196 | |
| 197 | static inline void build_cdex_p(void) |
| 198 | { |
| 199 | union mips_instruction mi; |
| 200 | |
| 201 | if (store_offset & (cpu_dcache_line_size() - 1)) |
| 202 | return; |
| 203 | |
Thiemo Seufer | 330cfe0 | 2005-09-01 18:33:58 +0000 | [diff] [blame] | 204 | if (R4600_V1_HIT_CACHEOP_WAR && cpu_is_r4600_v1_x()) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 205 | build_nop(); |
| 206 | build_nop(); |
| 207 | build_nop(); |
| 208 | build_nop(); |
| 209 | } |
| 210 | |
Thiemo Seufer | 330cfe0 | 2005-09-01 18:33:58 +0000 | [diff] [blame] | 211 | if (R4600_V2_HIT_CACHEOP_WAR && cpu_is_r4600_v2_x()) |
Thiemo Seufer | 10a3dab | 2005-09-09 20:26:54 +0000 | [diff] [blame] | 212 | build_insn_word(0x3c01a000); /* lui $at, 0xa000 */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 213 | |
| 214 | mi.c_format.opcode = cache_op; |
| 215 | mi.c_format.rs = 4; /* $a0 */ |
| 216 | mi.c_format.c_op = 3; /* Create Dirty Exclusive */ |
| 217 | mi.c_format.cache = 1; /* Data Cache */ |
| 218 | mi.c_format.simmediate = store_offset; |
| 219 | |
| 220 | emit_instruction(mi); |
| 221 | } |
| 222 | |
| 223 | static void __init __build_store_reg(int reg) |
| 224 | { |
| 225 | union mips_instruction mi; |
| 226 | unsigned int width; |
| 227 | |
| 228 | if (cpu_has_64bit_gp_regs || |
| 229 | (cpu_has_64bit_zero_reg && reg == 0)) { |
| 230 | mi.i_format.opcode = sd_op; |
| 231 | width = 8; |
| 232 | } else { |
| 233 | mi.i_format.opcode = sw_op; |
| 234 | width = 4; |
| 235 | } |
| 236 | mi.i_format.rs = 4; /* $a0 */ |
| 237 | mi.i_format.rt = reg; /* $reg */ |
| 238 | mi.i_format.simmediate = store_offset; |
| 239 | |
| 240 | store_offset += width; |
| 241 | emit_instruction(mi); |
| 242 | } |
| 243 | |
| 244 | static inline void build_store_reg(int reg) |
| 245 | { |
| 246 | if (cpu_has_prefetch) |
| 247 | if (reg) |
| 248 | build_dst_pref(pref_offset_copy); |
| 249 | else |
| 250 | build_dst_pref(pref_offset_clear); |
| 251 | else if (cpu_has_cache_cdex_s) |
| 252 | build_cdex_s(); |
| 253 | else if (cpu_has_cache_cdex_p) |
| 254 | build_cdex_p(); |
| 255 | |
| 256 | __build_store_reg(reg); |
| 257 | } |
| 258 | |
| 259 | static inline void build_addiu_a2_a0(unsigned long offset) |
| 260 | { |
| 261 | union mips_instruction mi; |
| 262 | |
| 263 | BUG_ON(offset > 0x7fff); |
| 264 | |
| 265 | mi.i_format.opcode = cpu_has_64bit_gp_regs ? daddiu_op : addiu_op; |
| 266 | mi.i_format.rs = 4; /* $a0 */ |
| 267 | mi.i_format.rt = 6; /* $a2 */ |
| 268 | mi.i_format.simmediate = offset; |
| 269 | |
| 270 | emit_instruction(mi); |
| 271 | } |
| 272 | |
| 273 | static inline void build_addiu_a1(unsigned long offset) |
| 274 | { |
| 275 | union mips_instruction mi; |
| 276 | |
| 277 | BUG_ON(offset > 0x7fff); |
| 278 | |
| 279 | mi.i_format.opcode = cpu_has_64bit_gp_regs ? daddiu_op : addiu_op; |
| 280 | mi.i_format.rs = 5; /* $a1 */ |
| 281 | mi.i_format.rt = 5; /* $a1 */ |
| 282 | mi.i_format.simmediate = offset; |
| 283 | |
| 284 | load_offset -= offset; |
| 285 | |
| 286 | emit_instruction(mi); |
| 287 | } |
| 288 | |
| 289 | static inline void build_addiu_a0(unsigned long offset) |
| 290 | { |
| 291 | union mips_instruction mi; |
| 292 | |
| 293 | BUG_ON(offset > 0x7fff); |
| 294 | |
| 295 | mi.i_format.opcode = cpu_has_64bit_gp_regs ? daddiu_op : addiu_op; |
| 296 | mi.i_format.rs = 4; /* $a0 */ |
| 297 | mi.i_format.rt = 4; /* $a0 */ |
| 298 | mi.i_format.simmediate = offset; |
| 299 | |
| 300 | store_offset -= offset; |
| 301 | |
| 302 | emit_instruction(mi); |
| 303 | } |
| 304 | |
| 305 | static inline void build_bne(unsigned int *dest) |
| 306 | { |
| 307 | union mips_instruction mi; |
| 308 | |
| 309 | mi.i_format.opcode = bne_op; |
| 310 | mi.i_format.rs = 6; /* $a2 */ |
| 311 | mi.i_format.rt = 4; /* $a0 */ |
| 312 | mi.i_format.simmediate = dest - epc - 1; |
| 313 | |
| 314 | *epc++ = mi.word; |
| 315 | flush_delay_slot_or_nop(); |
| 316 | } |
| 317 | |
| 318 | static inline void build_jr_ra(void) |
| 319 | { |
| 320 | union mips_instruction mi; |
| 321 | |
| 322 | mi.r_format.opcode = spec_op; |
| 323 | mi.r_format.rs = 31; |
| 324 | mi.r_format.rt = 0; |
| 325 | mi.r_format.rd = 0; |
| 326 | mi.r_format.re = 0; |
| 327 | mi.r_format.func = jr_op; |
| 328 | |
| 329 | *epc++ = mi.word; |
| 330 | flush_delay_slot_or_nop(); |
| 331 | } |
| 332 | |
| 333 | void __init build_clear_page(void) |
| 334 | { |
| 335 | unsigned int loop_start; |
| 336 | |
| 337 | epc = (unsigned int *) &clear_page_array; |
| 338 | instruction_pending = 0; |
| 339 | store_offset = 0; |
| 340 | |
| 341 | if (cpu_has_prefetch) { |
| 342 | switch (current_cpu_data.cputype) { |
Atsushi Nemoto | de862b4 | 2006-03-17 12:59:22 +0900 | [diff] [blame] | 343 | case CPU_TX49XX: |
| 344 | /* TX49 supports only Pref_Load */ |
| 345 | pref_offset_clear = 0; |
| 346 | pref_offset_copy = 0; |
| 347 | break; |
| 348 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 349 | case CPU_RM9000: |
| 350 | /* |
| 351 | * As a workaround for erratum G105 which make the |
| 352 | * PrepareForStore hint unusable we fall back to |
| 353 | * StoreRetained on the RM9000. Once it is known which |
| 354 | * versions of the RM9000 we'll be able to condition- |
| 355 | * alize this. |
| 356 | */ |
| 357 | |
| 358 | case CPU_R10000: |
| 359 | case CPU_R12000: |
| 360 | pref_src_mode = Pref_LoadStreamed; |
| 361 | pref_dst_mode = Pref_StoreStreamed; |
| 362 | break; |
| 363 | |
| 364 | default: |
| 365 | pref_src_mode = Pref_LoadStreamed; |
| 366 | pref_dst_mode = Pref_PrepareForStore; |
| 367 | break; |
| 368 | } |
| 369 | } |
| 370 | |
| 371 | build_addiu_a2_a0(PAGE_SIZE - (cpu_has_prefetch ? pref_offset_clear : 0)); |
| 372 | |
Thiemo Seufer | 330cfe0 | 2005-09-01 18:33:58 +0000 | [diff] [blame] | 373 | if (R4600_V2_HIT_CACHEOP_WAR && cpu_is_r4600_v2_x()) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 374 | build_insn_word(0x3c01a000); /* lui $at, 0xa000 */ |
| 375 | |
| 376 | dest = label(); |
| 377 | do { |
| 378 | build_store_reg(0); |
| 379 | build_store_reg(0); |
| 380 | build_store_reg(0); |
| 381 | build_store_reg(0); |
| 382 | } while (store_offset < half_scache_line_size()); |
| 383 | build_addiu_a0(2 * store_offset); |
| 384 | loop_start = store_offset; |
| 385 | do { |
| 386 | build_store_reg(0); |
| 387 | build_store_reg(0); |
| 388 | build_store_reg(0); |
| 389 | build_store_reg(0); |
| 390 | } while ((store_offset - loop_start) < half_scache_line_size()); |
| 391 | build_bne(dest); |
| 392 | |
| 393 | if (cpu_has_prefetch && pref_offset_clear) { |
| 394 | build_addiu_a2_a0(pref_offset_clear); |
| 395 | dest = label(); |
| 396 | loop_start = store_offset; |
| 397 | do { |
| 398 | __build_store_reg(0); |
| 399 | __build_store_reg(0); |
| 400 | __build_store_reg(0); |
| 401 | __build_store_reg(0); |
| 402 | } while ((store_offset - loop_start) < half_scache_line_size()); |
| 403 | build_addiu_a0(2 * store_offset); |
| 404 | loop_start = store_offset; |
| 405 | do { |
| 406 | __build_store_reg(0); |
| 407 | __build_store_reg(0); |
| 408 | __build_store_reg(0); |
| 409 | __build_store_reg(0); |
| 410 | } while ((store_offset - loop_start) < half_scache_line_size()); |
| 411 | build_bne(dest); |
| 412 | } |
| 413 | |
| 414 | build_jr_ra(); |
| 415 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 416 | BUG_ON(epc > clear_page_array + ARRAY_SIZE(clear_page_array)); |
| 417 | } |
| 418 | |
| 419 | void __init build_copy_page(void) |
| 420 | { |
| 421 | unsigned int loop_start; |
| 422 | |
| 423 | epc = (unsigned int *) ©_page_array; |
| 424 | store_offset = load_offset = 0; |
| 425 | instruction_pending = 0; |
| 426 | |
| 427 | build_addiu_a2_a0(PAGE_SIZE - (cpu_has_prefetch ? pref_offset_copy : 0)); |
| 428 | |
Thiemo Seufer | 330cfe0 | 2005-09-01 18:33:58 +0000 | [diff] [blame] | 429 | if (R4600_V2_HIT_CACHEOP_WAR && cpu_is_r4600_v2_x()) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 430 | build_insn_word(0x3c01a000); /* lui $at, 0xa000 */ |
| 431 | |
| 432 | dest = label(); |
| 433 | loop_start = store_offset; |
| 434 | do { |
| 435 | build_load_reg( 8); |
| 436 | build_load_reg( 9); |
| 437 | build_load_reg(10); |
| 438 | build_load_reg(11); |
| 439 | build_store_reg( 8); |
| 440 | build_store_reg( 9); |
| 441 | build_store_reg(10); |
| 442 | build_store_reg(11); |
| 443 | } while ((store_offset - loop_start) < half_scache_line_size()); |
| 444 | build_addiu_a0(2 * store_offset); |
| 445 | build_addiu_a1(2 * load_offset); |
| 446 | loop_start = store_offset; |
| 447 | do { |
| 448 | build_load_reg( 8); |
| 449 | build_load_reg( 9); |
| 450 | build_load_reg(10); |
| 451 | build_load_reg(11); |
| 452 | build_store_reg( 8); |
| 453 | build_store_reg( 9); |
| 454 | build_store_reg(10); |
| 455 | build_store_reg(11); |
| 456 | } while ((store_offset - loop_start) < half_scache_line_size()); |
| 457 | build_bne(dest); |
| 458 | |
| 459 | if (cpu_has_prefetch && pref_offset_copy) { |
| 460 | build_addiu_a2_a0(pref_offset_copy); |
| 461 | dest = label(); |
| 462 | loop_start = store_offset; |
| 463 | do { |
| 464 | __build_load_reg( 8); |
| 465 | __build_load_reg( 9); |
| 466 | __build_load_reg(10); |
| 467 | __build_load_reg(11); |
| 468 | __build_store_reg( 8); |
| 469 | __build_store_reg( 9); |
| 470 | __build_store_reg(10); |
| 471 | __build_store_reg(11); |
| 472 | } while ((store_offset - loop_start) < half_scache_line_size()); |
| 473 | build_addiu_a0(2 * store_offset); |
| 474 | build_addiu_a1(2 * load_offset); |
| 475 | loop_start = store_offset; |
| 476 | do { |
| 477 | __build_load_reg( 8); |
| 478 | __build_load_reg( 9); |
| 479 | __build_load_reg(10); |
| 480 | __build_load_reg(11); |
| 481 | __build_store_reg( 8); |
| 482 | __build_store_reg( 9); |
| 483 | __build_store_reg(10); |
| 484 | __build_store_reg(11); |
| 485 | } while ((store_offset - loop_start) < half_scache_line_size()); |
| 486 | build_bne(dest); |
| 487 | } |
| 488 | |
| 489 | build_jr_ra(); |
| 490 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 491 | BUG_ON(epc > copy_page_array + ARRAY_SIZE(copy_page_array)); |
| 492 | } |