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) 1994, 1995 Waldorf GmbH 7 * Copyright (C) 1994 - 2000, 06 Ralf Baechle 8 * Copyright (C) 1999, 2000 Silicon Graphics, Inc. 9 * Copyright (C) 2004, 2005 MIPS Technologies, Inc. All rights reserved. 10 * Author: Maciej W. Rozycki <macro@mips.com> 11 */ 12 #ifndef _ASM_IO_H 13 #define _ASM_IO_H 14 15 #include <linux/compiler.h> 16 #include <linux/kernel.h> 17 #include <linux/types.h> 18 19 #include <asm/addrspace.h> 20 #include <asm/byteorder.h> 21 #include <asm/cpu.h> 22 #include <asm/cpu-features.h> 23 #include <asm-generic/iomap.h> 24 #include <asm/page.h> 25 #include <asm/pgtable-bits.h> 26 #include <asm/processor.h> 27 #include <asm/string.h> 28 29 #include <ioremap.h> 30 #include <mangle-port.h> 31 32 /* 33 * Slowdown I/O port space accesses for antique hardware. 34 */ 35 #undef CONF_SLOWDOWN_IO 36 37 /* 38 * Raw operations are never swapped in software. OTOH values that raw 39 * operations are working on may or may not have been swapped by the bus 40 * hardware. An example use would be for flash memory that's used for 41 * execute in place. 42 */ 43 # define __raw_ioswabb(a, x) (x) 44 # define __raw_ioswabw(a, x) (x) 45 # define __raw_ioswabl(a, x) (x) 46 # define __raw_ioswabq(a, x) (x) 47 # define ____raw_ioswabq(a, x) (x) 48 49 /* ioswab[bwlq], __mem_ioswab[bwlq] are defined in mangle-port.h */ 50 51 #define IO_SPACE_LIMIT 0xffff 52 53 /* 54 * On MIPS I/O ports are memory mapped, so we access them using normal 55 * load/store instructions. mips_io_port_base is the virtual address to 56 * which all ports are being mapped. For sake of efficiency some code 57 * assumes that this is an address that can be loaded with a single lui 58 * instruction, so the lower 16 bits must be zero. Should be true on 59 * on any sane architecture; generic code does not use this assumption. 60 */ 61 extern const unsigned long mips_io_port_base; 62 63 /* 64 * Gcc will generate code to load the value of mips_io_port_base after each 65 * function call which may be fairly wasteful in some cases. So we don't 66 * play quite by the book. We tell gcc mips_io_port_base is a long variable 67 * which solves the code generation issue. Now we need to violate the 68 * aliasing rules a little to make initialization possible and finally we 69 * will need the barrier() to fight side effects of the aliasing chat. 70 * This trickery will eventually collapse under gcc's optimizer. Oh well. 71 */ 72 static inline void set_io_port_base(unsigned long base) 73 { 74 * (unsigned long *) &mips_io_port_base = base; 75 barrier(); 76 } 77 78 /* 79 * Thanks to James van Artsdalen for a better timing-fix than 80 * the two short jumps: using outb's to a nonexistent port seems 81 * to guarantee better timings even on fast machines. 82 * 83 * On the other hand, I'd like to be sure of a non-existent port: 84 * I feel a bit unsafe about using 0x80 (should be safe, though) 85 * 86 * Linus 87 * 88 */ 89 90 #define __SLOW_DOWN_IO \ 91 __asm__ __volatile__( \ 92 "sb\t$0,0x80(%0)" \ 93 : : "r" (mips_io_port_base)); 94 95 #ifdef CONF_SLOWDOWN_IO 96 #ifdef REALLY_SLOW_IO 97 #define SLOW_DOWN_IO { __SLOW_DOWN_IO; __SLOW_DOWN_IO; __SLOW_DOWN_IO; __SLOW_DOWN_IO; } 98 #else 99 #define SLOW_DOWN_IO __SLOW_DOWN_IO 100 #endif 101 #else 102 #define SLOW_DOWN_IO 103 #endif 104 105 /* 106 * virt_to_phys - map virtual addresses to physical 107 * @address: address to remap 108 * 109 * The returned physical address is the physical (CPU) mapping for 110 * the memory address given. It is only valid to use this function on 111 * addresses directly mapped or allocated via kmalloc. 112 * 113 * This function does not give bus mappings for DMA transfers. In 114 * almost all conceivable cases a device driver should not be using 115 * this function 116 */ 117 static inline unsigned long virt_to_phys(volatile const void *address) 118 { 119 return (unsigned long)address - PAGE_OFFSET + PHYS_OFFSET; 120 } 121 122 /* 123 * phys_to_virt - map physical address to virtual 124 * @address: address to remap 125 * 126 * The returned virtual address is a current CPU mapping for 127 * the memory address given. It is only valid to use this function on 128 * addresses that have a kernel mapping 129 * 130 * This function does not handle bus mappings for DMA transfers. In 131 * almost all conceivable cases a device driver should not be using 132 * this function 133 */ 134 static inline void * phys_to_virt(unsigned long address) 135 { 136 return (void *)(address + PAGE_OFFSET - PHYS_OFFSET); 137 } 138 139 /* 140 * ISA I/O bus memory addresses are 1:1 with the physical address. 141 */ 142 static inline unsigned long isa_virt_to_bus(volatile void * address) 143 { 144 return (unsigned long)address - PAGE_OFFSET; 145 } 146 147 static inline void * isa_bus_to_virt(unsigned long address) 148 { 149 return (void *)(address + PAGE_OFFSET); 150 } 151 152 #define isa_page_to_bus page_to_phys 153 154 /* 155 * However PCI ones are not necessarily 1:1 and therefore these interfaces 156 * are forbidden in portable PCI drivers. 157 * 158 * Allow them for x86 for legacy drivers, though. 159 */ 160 #define virt_to_bus virt_to_phys 161 #define bus_to_virt phys_to_virt 162 163 /* 164 * Change "struct page" to physical address. 165 */ 166 #define page_to_phys(page) ((dma_addr_t)page_to_pfn(page) << PAGE_SHIFT) 167 168 extern void __iomem * __ioremap(phys_t offset, phys_t size, unsigned long flags); 169 extern void __iounmap(const volatile void __iomem *addr); 170 171 static inline void __iomem * __ioremap_mode(phys_t offset, unsigned long size, 172 unsigned long flags) 173 { 174 void __iomem *addr = plat_ioremap(offset, size, flags); 175 176 if (addr) 177 return addr; 178 179 #define __IS_LOW512(addr) (!((phys_t)(addr) & (phys_t) ~0x1fffffffULL)) 180 181 if (cpu_has_64bit_addresses) { 182 u64 base = UNCAC_BASE; 183 184 /* 185 * R10000 supports a 2 bit uncached attribute therefore 186 * UNCAC_BASE may not equal IO_BASE. 187 */ 188 if (flags == _CACHE_UNCACHED) 189 base = (u64) IO_BASE; 190 return (void __iomem *) (unsigned long) (base + offset); 191 } else if (__builtin_constant_p(offset) && 192 __builtin_constant_p(size) && __builtin_constant_p(flags)) { 193 phys_t phys_addr, last_addr; 194 195 phys_addr = fixup_bigphys_addr(offset, size); 196 197 /* Don't allow wraparound or zero size. */ 198 last_addr = phys_addr + size - 1; 199 if (!size || last_addr < phys_addr) 200 return NULL; 201 202 /* 203 * Map uncached objects in the low 512MB of address 204 * space using KSEG1. 205 */ 206 if (__IS_LOW512(phys_addr) && __IS_LOW512(last_addr) && 207 flags == _CACHE_UNCACHED) 208 return (void __iomem *) 209 (unsigned long)CKSEG1ADDR(phys_addr); 210 } 211 212 return __ioremap(offset, size, flags); 213 214 #undef __IS_LOW512 215 } 216 217 /* 218 * ioremap - map bus memory into CPU space 219 * @offset: bus address of the memory 220 * @size: size of the resource to map 221 * 222 * ioremap performs a platform specific sequence of operations to 223 * make bus memory CPU accessible via the readb/readw/readl/writeb/ 224 * writew/writel functions and the other mmio helpers. The returned 225 * address is not guaranteed to be usable directly as a virtual 226 * address. 227 */ 228 #define ioremap(offset, size) \ 229 __ioremap_mode((offset), (size), _CACHE_UNCACHED) 230 231 /* 232 * ioremap_nocache - map bus memory into CPU space 233 * @offset: bus address of the memory 234 * @size: size of the resource to map 235 * 236 * ioremap_nocache performs a platform specific sequence of operations to 237 * make bus memory CPU accessible via the readb/readw/readl/writeb/ 238 * writew/writel functions and the other mmio helpers. The returned 239 * address is not guaranteed to be usable directly as a virtual 240 * address. 241 * 242 * This version of ioremap ensures that the memory is marked uncachable 243 * on the CPU as well as honouring existing caching rules from things like 244 * the PCI bus. Note that there are other caches and buffers on many 245 * busses. In paticular driver authors should read up on PCI writes 246 * 247 * It's useful if some control registers are in such an area and 248 * write combining or read caching is not desirable: 249 */ 250 #define ioremap_nocache(offset, size) \ 251 __ioremap_mode((offset), (size), _CACHE_UNCACHED) 252 253 /* 254 * ioremap_cachable - map bus memory into CPU space 255 * @offset: bus address of the memory 256 * @size: size of the resource to map 257 * 258 * ioremap_nocache performs a platform specific sequence of operations to 259 * make bus memory CPU accessible via the readb/readw/readl/writeb/ 260 * writew/writel functions and the other mmio helpers. The returned 261 * address is not guaranteed to be usable directly as a virtual 262 * address. 263 * 264 * This version of ioremap ensures that the memory is marked cachable by 265 * the CPU. Also enables full write-combining. Useful for some 266 * memory-like regions on I/O busses. 267 */ 268 #define ioremap_cachable(offset, size) \ 269 __ioremap_mode((offset), (size), _page_cachable_default) 270 271 /* 272 * These two are MIPS specific ioremap variant. ioremap_cacheable_cow 273 * requests a cachable mapping, ioremap_uncached_accelerated requests a 274 * mapping using the uncached accelerated mode which isn't supported on 275 * all processors. 276 */ 277 #define ioremap_cacheable_cow(offset, size) \ 278 __ioremap_mode((offset), (size), _CACHE_CACHABLE_COW) 279 #define ioremap_uncached_accelerated(offset, size) \ 280 __ioremap_mode((offset), (size), _CACHE_UNCACHED_ACCELERATED) 281 282 static inline void iounmap(const volatile void __iomem *addr) 283 { 284 if (plat_iounmap(addr)) 285 return; 286 287 #define __IS_KSEG1(addr) (((unsigned long)(addr) & ~0x1fffffffUL) == CKSEG1) 288 289 if (cpu_has_64bit_addresses || 290 (__builtin_constant_p(addr) && __IS_KSEG1(addr))) 291 return; 292 293 __iounmap(addr); 294 295 #undef __IS_KSEG1 296 } 297 298 #ifdef CONFIG_CPU_CAVIUM_OCTEON 299 #define war_octeon_io_reorder_wmb() wmb() 300 #else 301 #define war_octeon_io_reorder_wmb() do { } while (0) 302 #endif 303 304 #define __BUILD_MEMORY_SINGLE(pfx, bwlq, type, irq) \ 305 \ 306 static inline void pfx##write##bwlq(type val, \ 307 volatile void __iomem *mem) \ 308 { \ 309 volatile type *__mem; \ 310 type __val; \ 311 \ 312 war_octeon_io_reorder_wmb(); \ 313 \ 314 __mem = (void *)__swizzle_addr_##bwlq((unsigned long)(mem)); \ 315 \ 316 __val = pfx##ioswab##bwlq(__mem, val); \ 317 \ 318 if (sizeof(type) != sizeof(u64) || sizeof(u64) == sizeof(long)) \ 319 *__mem = __val; \ 320 else if (cpu_has_64bits) { \ 321 unsigned long __flags; \ 322 type __tmp; \ 323 \ 324 if (irq) \ 325 local_irq_save(__flags); \ 326 __asm__ __volatile__( \ 327 ".set mips3" "\t\t# __writeq""\n\t" \ 328 "dsll32 %L0, %L0, 0" "\n\t" \ 329 "dsrl32 %L0, %L0, 0" "\n\t" \ 330 "dsll32 %M0, %M0, 0" "\n\t" \ 331 "or %L0, %L0, %M0" "\n\t" \ 332 "sd %L0, %2" "\n\t" \ 333 ".set mips0" "\n" \ 334 : "=r" (__tmp) \ 335 : "0" (__val), "m" (*__mem)); \ 336 if (irq) \ 337 local_irq_restore(__flags); \ 338 } else \ 339 BUG(); \ 340 } \ 341 \ 342 static inline type pfx##read##bwlq(const volatile void __iomem *mem) \ 343 { \ 344 volatile type *__mem; \ 345 type __val; \ 346 \ 347 __mem = (void *)__swizzle_addr_##bwlq((unsigned long)(mem)); \ 348 \ 349 if (sizeof(type) != sizeof(u64) || sizeof(u64) == sizeof(long)) \ 350 __val = *__mem; \ 351 else if (cpu_has_64bits) { \ 352 unsigned long __flags; \ 353 \ 354 if (irq) \ 355 local_irq_save(__flags); \ 356 __asm__ __volatile__( \ 357 ".set mips3" "\t\t# __readq" "\n\t" \ 358 "ld %L0, %1" "\n\t" \ 359 "dsra32 %M0, %L0, 0" "\n\t" \ 360 "sll %L0, %L0, 0" "\n\t" \ 361 ".set mips0" "\n" \ 362 : "=r" (__val) \ 363 : "m" (*__mem)); \ 364 if (irq) \ 365 local_irq_restore(__flags); \ 366 } else { \ 367 __val = 0; \ 368 BUG(); \ 369 } \ 370 \ 371 return pfx##ioswab##bwlq(__mem, __val); \ 372 } 373 374 #define __BUILD_IOPORT_SINGLE(pfx, bwlq, type, p, slow) \ 375 \ 376 static inline void pfx##out##bwlq##p(type val, unsigned long port) \ 377 { \ 378 volatile type *__addr; \ 379 type __val; \ 380 \ 381 war_octeon_io_reorder_wmb(); \ 382 \ 383 __addr = (void *)__swizzle_addr_##bwlq(mips_io_port_base + port); \ 384 \ 385 __val = pfx##ioswab##bwlq(__addr, val); \ 386 \ 387 /* Really, we want this to be atomic */ \ 388 BUILD_BUG_ON(sizeof(type) > sizeof(unsigned long)); \ 389 \ 390 *__addr = __val; \ 391 slow; \ 392 } \ 393 \ 394 static inline type pfx##in##bwlq##p(unsigned long port) \ 395 { \ 396 volatile type *__addr; \ 397 type __val; \ 398 \ 399 __addr = (void *)__swizzle_addr_##bwlq(mips_io_port_base + port); \ 400 \ 401 BUILD_BUG_ON(sizeof(type) > sizeof(unsigned long)); \ 402 \ 403 __val = *__addr; \ 404 slow; \ 405 \ 406 return pfx##ioswab##bwlq(__addr, __val); \ 407 } 408 409 #define __BUILD_MEMORY_PFX(bus, bwlq, type) \ 410 \ 411 __BUILD_MEMORY_SINGLE(bus, bwlq, type, 1) 412 413 #define BUILDIO_MEM(bwlq, type) \ 414 \ 415 __BUILD_MEMORY_PFX(__raw_, bwlq, type) \ 416 __BUILD_MEMORY_PFX(, bwlq, type) \ 417 __BUILD_MEMORY_PFX(__mem_, bwlq, type) \ 418 419 BUILDIO_MEM(b, u8) 420 BUILDIO_MEM(w, u16) 421 BUILDIO_MEM(l, u32) 422 BUILDIO_MEM(q, u64) 423 424 #define __BUILD_IOPORT_PFX(bus, bwlq, type) \ 425 __BUILD_IOPORT_SINGLE(bus, bwlq, type, ,) \ 426 __BUILD_IOPORT_SINGLE(bus, bwlq, type, _p, SLOW_DOWN_IO) 427 428 #define BUILDIO_IOPORT(bwlq, type) \ 429 __BUILD_IOPORT_PFX(, bwlq, type) \ 430 __BUILD_IOPORT_PFX(__mem_, bwlq, type) 431 432 BUILDIO_IOPORT(b, u8) 433 BUILDIO_IOPORT(w, u16) 434 BUILDIO_IOPORT(l, u32) 435 #ifdef CONFIG_64BIT 436 BUILDIO_IOPORT(q, u64) 437 #endif 438 439 #define __BUILDIO(bwlq, type) \ 440 \ 441 __BUILD_MEMORY_SINGLE(____raw_, bwlq, type, 0) 442 443 __BUILDIO(q, u64) 444 445 #define readb_relaxed readb 446 #define readw_relaxed readw 447 #define readl_relaxed readl 448 #define readq_relaxed readq 449 450 /* 451 * Some code tests for these symbols 452 */ 453 #define readq readq 454 #define writeq writeq 455 456 #define __BUILD_MEMORY_STRING(bwlq, type) \ 457 \ 458 static inline void writes##bwlq(volatile void __iomem *mem, \ 459 const void *addr, unsigned int count) \ 460 { \ 461 const volatile type *__addr = addr; \ 462 \ 463 while (count--) { \ 464 __mem_write##bwlq(*__addr, mem); \ 465 __addr++; \ 466 } \ 467 } \ 468 \ 469 static inline void reads##bwlq(volatile void __iomem *mem, void *addr, \ 470 unsigned int count) \ 471 { \ 472 volatile type *__addr = addr; \ 473 \ 474 while (count--) { \ 475 *__addr = __mem_read##bwlq(mem); \ 476 __addr++; \ 477 } \ 478 } 479 480 #define __BUILD_IOPORT_STRING(bwlq, type) \ 481 \ 482 static inline void outs##bwlq(unsigned long port, const void *addr, \ 483 unsigned int count) \ 484 { \ 485 const volatile type *__addr = addr; \ 486 \ 487 while (count--) { \ 488 __mem_out##bwlq(*__addr, port); \ 489 __addr++; \ 490 } \ 491 } \ 492 \ 493 static inline void ins##bwlq(unsigned long port, void *addr, \ 494 unsigned int count) \ 495 { \ 496 volatile type *__addr = addr; \ 497 \ 498 while (count--) { \ 499 *__addr = __mem_in##bwlq(port); \ 500 __addr++; \ 501 } \ 502 } 503 504 #define BUILDSTRING(bwlq, type) \ 505 \ 506 __BUILD_MEMORY_STRING(bwlq, type) \ 507 __BUILD_IOPORT_STRING(bwlq, type) 508 509 BUILDSTRING(b, u8) 510 BUILDSTRING(w, u16) 511 BUILDSTRING(l, u32) 512 #ifdef CONFIG_64BIT 513 BUILDSTRING(q, u64) 514 #endif 515 516 517 #ifdef CONFIG_CPU_CAVIUM_OCTEON 518 #define mmiowb() wmb() 519 #else 520 /* Depends on MIPS II instruction set */ 521 #define mmiowb() asm volatile ("sync" ::: "memory") 522 #endif 523 524 static inline void memset_io(volatile void __iomem *addr, unsigned char val, int count) 525 { 526 memset((void __force *) addr, val, count); 527 } 528 static inline void memcpy_fromio(void *dst, const volatile void __iomem *src, int count) 529 { 530 memcpy(dst, (void __force *) src, count); 531 } 532 static inline void memcpy_toio(volatile void __iomem *dst, const void *src, int count) 533 { 534 memcpy((void __force *) dst, src, count); 535 } 536 537 /* 538 * The caches on some architectures aren't dma-coherent and have need to 539 * handle this in software. There are three types of operations that 540 * can be applied to dma buffers. 541 * 542 * - dma_cache_wback_inv(start, size) makes caches and coherent by 543 * writing the content of the caches back to memory, if necessary. 544 * The function also invalidates the affected part of the caches as 545 * necessary before DMA transfers from outside to memory. 546 * - dma_cache_wback(start, size) makes caches and coherent by 547 * writing the content of the caches back to memory, if necessary. 548 * The function also invalidates the affected part of the caches as 549 * necessary before DMA transfers from outside to memory. 550 * - dma_cache_inv(start, size) invalidates the affected parts of the 551 * caches. Dirty lines of the caches may be written back or simply 552 * be discarded. This operation is necessary before dma operations 553 * to the memory. 554 * 555 * This API used to be exported; it now is for arch code internal use only. 556 */ 557 #ifdef CONFIG_DMA_NONCOHERENT 558 559 extern void (*_dma_cache_wback_inv)(unsigned long start, unsigned long size); 560 extern void (*_dma_cache_wback)(unsigned long start, unsigned long size); 561 extern void (*_dma_cache_inv)(unsigned long start, unsigned long size); 562 563 #define dma_cache_wback_inv(start, size) _dma_cache_wback_inv(start, size) 564 #define dma_cache_wback(start, size) _dma_cache_wback(start, size) 565 #define dma_cache_inv(start, size) _dma_cache_inv(start, size) 566 567 #else /* Sane hardware */ 568 569 #define dma_cache_wback_inv(start,size) \ 570 do { (void) (start); (void) (size); } while (0) 571 #define dma_cache_wback(start,size) \ 572 do { (void) (start); (void) (size); } while (0) 573 #define dma_cache_inv(start,size) \ 574 do { (void) (start); (void) (size); } while (0) 575 576 #endif /* CONFIG_DMA_NONCOHERENT */ 577 578 /* 579 * Read a 32-bit register that requires a 64-bit read cycle on the bus. 580 * Avoid interrupt mucking, just adjust the address for 4-byte access. 581 * Assume the addresses are 8-byte aligned. 582 */ 583 #ifdef __MIPSEB__ 584 #define __CSR_32_ADJUST 4 585 #else 586 #define __CSR_32_ADJUST 0 587 #endif 588 589 #define csr_out32(v, a) (*(volatile u32 *)((unsigned long)(a) + __CSR_32_ADJUST) = (v)) 590 #define csr_in32(a) (*(volatile u32 *)((unsigned long)(a) + __CSR_32_ADJUST)) 591 592 /* 593 * Convert a physical pointer to a virtual kernel pointer for /dev/mem 594 * access 595 */ 596 #define xlate_dev_mem_ptr(p) __va(p) 597 598 /* 599 * Convert a virtual cached pointer to an uncached pointer 600 */ 601 #define xlate_dev_kmem_ptr(p) p 602 603 #endif /* _ASM_IO_H */ 604