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 #include <linux/irqflags.h> 19 20 #include <asm/addrspace.h> 21 #include <asm/bug.h> 22 #include <asm/byteorder.h> 23 #include <asm/cpu.h> 24 #include <asm/cpu-features.h> 25 #include <asm-generic/iomap.h> 26 #include <asm/page.h> 27 #include <asm/pgtable-bits.h> 28 #include <asm/processor.h> 29 #include <asm/string.h> 30 31 #include <ioremap.h> 32 #include <mangle-port.h> 33 34 /* 35 * Slowdown I/O port space accesses for antique hardware. 36 */ 37 #undef CONF_SLOWDOWN_IO 38 39 /* 40 * Raw operations are never swapped in software. OTOH values that raw 41 * operations are working on may or may not have been swapped by the bus 42 * hardware. An example use would be for flash memory that's used for 43 * execute in place. 44 */ 45 # define __raw_ioswabb(a, x) (x) 46 # define __raw_ioswabw(a, x) (x) 47 # define __raw_ioswabl(a, x) (x) 48 # define __raw_ioswabq(a, x) (x) 49 # define ____raw_ioswabq(a, x) (x) 50 51 /* ioswab[bwlq], __mem_ioswab[bwlq] are defined in mangle-port.h */ 52 53 #define IO_SPACE_LIMIT 0xffff 54 55 /* 56 * On MIPS I/O ports are memory mapped, so we access them using normal 57 * load/store instructions. mips_io_port_base is the virtual address to 58 * which all ports are being mapped. For sake of efficiency some code 59 * assumes that this is an address that can be loaded with a single lui 60 * instruction, so the lower 16 bits must be zero. Should be true on 61 * on any sane architecture; generic code does not use this assumption. 62 */ 63 extern const unsigned long mips_io_port_base; 64 65 /* 66 * Gcc will generate code to load the value of mips_io_port_base after each 67 * function call which may be fairly wasteful in some cases. So we don't 68 * play quite by the book. We tell gcc mips_io_port_base is a long variable 69 * which solves the code generation issue. Now we need to violate the 70 * aliasing rules a little to make initialization possible and finally we 71 * will need the barrier() to fight side effects of the aliasing chat. 72 * This trickery will eventually collapse under gcc's optimizer. Oh well. 73 */ 74 static inline void set_io_port_base(unsigned long base) 75 { 76 * (unsigned long *) &mips_io_port_base = base; 77 barrier(); 78 } 79 80 /* 81 * Thanks to James van Artsdalen for a better timing-fix than 82 * the two short jumps: using outb's to a nonexistent port seems 83 * to guarantee better timings even on fast machines. 84 * 85 * On the other hand, I'd like to be sure of a non-existent port: 86 * I feel a bit unsafe about using 0x80 (should be safe, though) 87 * 88 * Linus 89 * 90 */ 91 92 #define __SLOW_DOWN_IO \ 93 __asm__ __volatile__( \ 94 "sb\t$0,0x80(%0)" \ 95 : : "r" (mips_io_port_base)); 96 97 #ifdef CONF_SLOWDOWN_IO 98 #ifdef REALLY_SLOW_IO 99 #define SLOW_DOWN_IO { __SLOW_DOWN_IO; __SLOW_DOWN_IO; __SLOW_DOWN_IO; __SLOW_DOWN_IO; } 100 #else 101 #define SLOW_DOWN_IO __SLOW_DOWN_IO 102 #endif 103 #else 104 #define SLOW_DOWN_IO 105 #endif 106 107 /* 108 * virt_to_phys - map virtual addresses to physical 109 * @address: address to remap 110 * 111 * The returned physical address is the physical (CPU) mapping for 112 * the memory address given. It is only valid to use this function on 113 * addresses directly mapped or allocated via kmalloc. 114 * 115 * This function does not give bus mappings for DMA transfers. In 116 * almost all conceivable cases a device driver should not be using 117 * this function 118 */ 119 static inline unsigned long virt_to_phys(volatile const void *address) 120 { 121 return __pa(address); 122 } 123 124 /* 125 * phys_to_virt - map physical address to virtual 126 * @address: address to remap 127 * 128 * The returned virtual address is a current CPU mapping for 129 * the memory address given. It is only valid to use this function on 130 * addresses that have a kernel mapping 131 * 132 * This function does not handle bus mappings for DMA transfers. In 133 * almost all conceivable cases a device driver should not be using 134 * this function 135 */ 136 static inline void * phys_to_virt(unsigned long address) 137 { 138 return (void *)(address + PAGE_OFFSET - PHYS_OFFSET); 139 } 140 141 /* 142 * ISA I/O bus memory addresses are 1:1 with the physical address. 143 */ 144 static inline unsigned long isa_virt_to_bus(volatile void * address) 145 { 146 return (unsigned long)address - PAGE_OFFSET; 147 } 148 149 static inline void * isa_bus_to_virt(unsigned long address) 150 { 151 return (void *)(address + PAGE_OFFSET); 152 } 153 154 #define isa_page_to_bus page_to_phys 155 156 /* 157 * However PCI ones are not necessarily 1:1 and therefore these interfaces 158 * are forbidden in portable PCI drivers. 159 * 160 * Allow them for x86 for legacy drivers, though. 161 */ 162 #define virt_to_bus virt_to_phys 163 #define bus_to_virt phys_to_virt 164 165 /* 166 * Change "struct page" to physical address. 167 */ 168 #define page_to_phys(page) ((dma_addr_t)page_to_pfn(page) << PAGE_SHIFT) 169 170 extern void __iomem * __ioremap(phys_addr_t offset, phys_addr_t size, unsigned long flags); 171 extern void __iounmap(const volatile void __iomem *addr); 172 173 #ifndef CONFIG_PCI 174 struct pci_dev; 175 static inline void pci_iounmap(struct pci_dev *dev, void __iomem *addr) {} 176 #endif 177 178 static inline void __iomem * __ioremap_mode(phys_addr_t offset, unsigned long size, 179 unsigned long flags) 180 { 181 void __iomem *addr = plat_ioremap(offset, size, flags); 182 183 if (addr) 184 return addr; 185 186 #define __IS_LOW512(addr) (!((phys_addr_t)(addr) & (phys_addr_t) ~0x1fffffffULL)) 187 188 if (cpu_has_64bit_addresses) { 189 u64 base = UNCAC_BASE; 190 191 /* 192 * R10000 supports a 2 bit uncached attribute therefore 193 * UNCAC_BASE may not equal IO_BASE. 194 */ 195 if (flags == _CACHE_UNCACHED) 196 base = (u64) IO_BASE; 197 return (void __iomem *) (unsigned long) (base + offset); 198 } else if (__builtin_constant_p(offset) && 199 __builtin_constant_p(size) && __builtin_constant_p(flags)) { 200 phys_addr_t phys_addr, last_addr; 201 202 phys_addr = fixup_bigphys_addr(offset, size); 203 204 /* Don't allow wraparound or zero size. */ 205 last_addr = phys_addr + size - 1; 206 if (!size || last_addr < phys_addr) 207 return NULL; 208 209 /* 210 * Map uncached objects in the low 512MB of address 211 * space using KSEG1. 212 */ 213 if (__IS_LOW512(phys_addr) && __IS_LOW512(last_addr) && 214 flags == _CACHE_UNCACHED) 215 return (void __iomem *) 216 (unsigned long)CKSEG1ADDR(phys_addr); 217 } 218 219 return __ioremap(offset, size, flags); 220 221 #undef __IS_LOW512 222 } 223 224 /* 225 * ioremap - map bus memory into CPU space 226 * @offset: bus address of the memory 227 * @size: size of the resource to map 228 * 229 * ioremap performs a platform specific sequence of operations to 230 * make bus memory CPU accessible via the readb/readw/readl/writeb/ 231 * writew/writel functions and the other mmio helpers. The returned 232 * address is not guaranteed to be usable directly as a virtual 233 * address. 234 */ 235 #define ioremap(offset, size) \ 236 __ioremap_mode((offset), (size), _CACHE_UNCACHED) 237 238 /* 239 * ioremap_nocache - map bus memory into CPU space 240 * @offset: bus address of the memory 241 * @size: size of the resource to map 242 * 243 * ioremap_nocache performs a platform specific sequence of operations to 244 * make bus memory CPU accessible via the readb/readw/readl/writeb/ 245 * writew/writel functions and the other mmio helpers. The returned 246 * address is not guaranteed to be usable directly as a virtual 247 * address. 248 * 249 * This version of ioremap ensures that the memory is marked uncachable 250 * on the CPU as well as honouring existing caching rules from things like 251 * the PCI bus. Note that there are other caches and buffers on many 252 * busses. In particular driver authors should read up on PCI writes 253 * 254 * It's useful if some control registers are in such an area and 255 * write combining or read caching is not desirable: 256 */ 257 #define ioremap_nocache(offset, size) \ 258 __ioremap_mode((offset), (size), _CACHE_UNCACHED) 259 #define ioremap_uc ioremap_nocache 260 261 /* 262 * ioremap_cachable - map bus memory into CPU space 263 * @offset: bus address of the memory 264 * @size: size of the resource to map 265 * 266 * ioremap_nocache performs a platform specific sequence of operations to 267 * make bus memory CPU accessible via the readb/readw/readl/writeb/ 268 * writew/writel functions and the other mmio helpers. The returned 269 * address is not guaranteed to be usable directly as a virtual 270 * address. 271 * 272 * This version of ioremap ensures that the memory is marked cachable by 273 * the CPU. Also enables full write-combining. Useful for some 274 * memory-like regions on I/O busses. 275 */ 276 #define ioremap_cachable(offset, size) \ 277 __ioremap_mode((offset), (size), _page_cachable_default) 278 #define ioremap_cache ioremap_cachable 279 280 /* 281 * These two are MIPS specific ioremap variant. ioremap_cacheable_cow 282 * requests a cachable mapping, ioremap_uncached_accelerated requests a 283 * mapping using the uncached accelerated mode which isn't supported on 284 * all processors. 285 */ 286 #define ioremap_cacheable_cow(offset, size) \ 287 __ioremap_mode((offset), (size), _CACHE_CACHABLE_COW) 288 #define ioremap_uncached_accelerated(offset, size) \ 289 __ioremap_mode((offset), (size), _CACHE_UNCACHED_ACCELERATED) 290 291 static inline void iounmap(const volatile void __iomem *addr) 292 { 293 if (plat_iounmap(addr)) 294 return; 295 296 #define __IS_KSEG1(addr) (((unsigned long)(addr) & ~0x1fffffffUL) == CKSEG1) 297 298 if (cpu_has_64bit_addresses || 299 (__builtin_constant_p(addr) && __IS_KSEG1(addr))) 300 return; 301 302 __iounmap(addr); 303 304 #undef __IS_KSEG1 305 } 306 307 #if defined(CONFIG_CPU_CAVIUM_OCTEON) || defined(CONFIG_LOONGSON3_ENHANCEMENT) 308 #define war_io_reorder_wmb() wmb() 309 #else 310 #define war_io_reorder_wmb() barrier() 311 #endif 312 313 #define __BUILD_MEMORY_SINGLE(pfx, bwlq, type, irq) \ 314 \ 315 static inline void pfx##write##bwlq(type val, \ 316 volatile void __iomem *mem) \ 317 { \ 318 volatile type *__mem; \ 319 type __val; \ 320 \ 321 war_io_reorder_wmb(); \ 322 \ 323 __mem = (void *)__swizzle_addr_##bwlq((unsigned long)(mem)); \ 324 \ 325 __val = pfx##ioswab##bwlq(__mem, val); \ 326 \ 327 if (sizeof(type) != sizeof(u64) || sizeof(u64) == sizeof(long)) \ 328 *__mem = __val; \ 329 else if (cpu_has_64bits) { \ 330 unsigned long __flags; \ 331 type __tmp; \ 332 \ 333 if (irq) \ 334 local_irq_save(__flags); \ 335 __asm__ __volatile__( \ 336 ".set arch=r4000" "\t\t# __writeq""\n\t" \ 337 "dsll32 %L0, %L0, 0" "\n\t" \ 338 "dsrl32 %L0, %L0, 0" "\n\t" \ 339 "dsll32 %M0, %M0, 0" "\n\t" \ 340 "or %L0, %L0, %M0" "\n\t" \ 341 "sd %L0, %2" "\n\t" \ 342 ".set mips0" "\n" \ 343 : "=r" (__tmp) \ 344 : "0" (__val), "m" (*__mem)); \ 345 if (irq) \ 346 local_irq_restore(__flags); \ 347 } else \ 348 BUG(); \ 349 } \ 350 \ 351 static inline type pfx##read##bwlq(const volatile void __iomem *mem) \ 352 { \ 353 volatile type *__mem; \ 354 type __val; \ 355 \ 356 __mem = (void *)__swizzle_addr_##bwlq((unsigned long)(mem)); \ 357 \ 358 if (sizeof(type) != sizeof(u64) || sizeof(u64) == sizeof(long)) \ 359 __val = *__mem; \ 360 else if (cpu_has_64bits) { \ 361 unsigned long __flags; \ 362 \ 363 if (irq) \ 364 local_irq_save(__flags); \ 365 __asm__ __volatile__( \ 366 ".set arch=r4000" "\t\t# __readq" "\n\t" \ 367 "ld %L0, %1" "\n\t" \ 368 "dsra32 %M0, %L0, 0" "\n\t" \ 369 "sll %L0, %L0, 0" "\n\t" \ 370 ".set mips0" "\n" \ 371 : "=r" (__val) \ 372 : "m" (*__mem)); \ 373 if (irq) \ 374 local_irq_restore(__flags); \ 375 } else { \ 376 __val = 0; \ 377 BUG(); \ 378 } \ 379 \ 380 /* prevent prefetching of coherent DMA data prematurely */ \ 381 rmb(); \ 382 return pfx##ioswab##bwlq(__mem, __val); \ 383 } 384 385 #define __BUILD_IOPORT_SINGLE(pfx, bwlq, type, p, slow) \ 386 \ 387 static inline void pfx##out##bwlq##p(type val, unsigned long port) \ 388 { \ 389 volatile type *__addr; \ 390 type __val; \ 391 \ 392 war_io_reorder_wmb(); \ 393 \ 394 __addr = (void *)__swizzle_addr_##bwlq(mips_io_port_base + port); \ 395 \ 396 __val = pfx##ioswab##bwlq(__addr, val); \ 397 \ 398 /* Really, we want this to be atomic */ \ 399 BUILD_BUG_ON(sizeof(type) > sizeof(unsigned long)); \ 400 \ 401 *__addr = __val; \ 402 slow; \ 403 } \ 404 \ 405 static inline type pfx##in##bwlq##p(unsigned long port) \ 406 { \ 407 volatile type *__addr; \ 408 type __val; \ 409 \ 410 __addr = (void *)__swizzle_addr_##bwlq(mips_io_port_base + port); \ 411 \ 412 BUILD_BUG_ON(sizeof(type) > sizeof(unsigned long)); \ 413 \ 414 __val = *__addr; \ 415 slow; \ 416 \ 417 /* prevent prefetching of coherent DMA data prematurely */ \ 418 rmb(); \ 419 return pfx##ioswab##bwlq(__addr, __val); \ 420 } 421 422 #define __BUILD_MEMORY_PFX(bus, bwlq, type) \ 423 \ 424 __BUILD_MEMORY_SINGLE(bus, bwlq, type, 1) 425 426 #define BUILDIO_MEM(bwlq, type) \ 427 \ 428 __BUILD_MEMORY_PFX(__raw_, bwlq, type) \ 429 __BUILD_MEMORY_PFX(, bwlq, type) \ 430 __BUILD_MEMORY_PFX(__mem_, bwlq, type) \ 431 432 BUILDIO_MEM(b, u8) 433 BUILDIO_MEM(w, u16) 434 BUILDIO_MEM(l, u32) 435 BUILDIO_MEM(q, u64) 436 437 #define __BUILD_IOPORT_PFX(bus, bwlq, type) \ 438 __BUILD_IOPORT_SINGLE(bus, bwlq, type, ,) \ 439 __BUILD_IOPORT_SINGLE(bus, bwlq, type, _p, SLOW_DOWN_IO) 440 441 #define BUILDIO_IOPORT(bwlq, type) \ 442 __BUILD_IOPORT_PFX(, bwlq, type) \ 443 __BUILD_IOPORT_PFX(__mem_, bwlq, type) 444 445 BUILDIO_IOPORT(b, u8) 446 BUILDIO_IOPORT(w, u16) 447 BUILDIO_IOPORT(l, u32) 448 #ifdef CONFIG_64BIT 449 BUILDIO_IOPORT(q, u64) 450 #endif 451 452 #define __BUILDIO(bwlq, type) \ 453 \ 454 __BUILD_MEMORY_SINGLE(____raw_, bwlq, type, 0) 455 456 __BUILDIO(q, u64) 457 458 #define readb_relaxed readb 459 #define readw_relaxed readw 460 #define readl_relaxed readl 461 #define readq_relaxed readq 462 463 #define writeb_relaxed writeb 464 #define writew_relaxed writew 465 #define writel_relaxed writel 466 #define writeq_relaxed writeq 467 468 #define readb_be(addr) \ 469 __raw_readb((__force unsigned *)(addr)) 470 #define readw_be(addr) \ 471 be16_to_cpu(__raw_readw((__force unsigned *)(addr))) 472 #define readl_be(addr) \ 473 be32_to_cpu(__raw_readl((__force unsigned *)(addr))) 474 #define readq_be(addr) \ 475 be64_to_cpu(__raw_readq((__force unsigned *)(addr))) 476 477 #define writeb_be(val, addr) \ 478 __raw_writeb((val), (__force unsigned *)(addr)) 479 #define writew_be(val, addr) \ 480 __raw_writew(cpu_to_be16((val)), (__force unsigned *)(addr)) 481 #define writel_be(val, addr) \ 482 __raw_writel(cpu_to_be32((val)), (__force unsigned *)(addr)) 483 #define writeq_be(val, addr) \ 484 __raw_writeq(cpu_to_be64((val)), (__force unsigned *)(addr)) 485 486 /* 487 * Some code tests for these symbols 488 */ 489 #define readq readq 490 #define writeq writeq 491 492 #define __BUILD_MEMORY_STRING(bwlq, type) \ 493 \ 494 static inline void writes##bwlq(volatile void __iomem *mem, \ 495 const void *addr, unsigned int count) \ 496 { \ 497 const volatile type *__addr = addr; \ 498 \ 499 while (count--) { \ 500 __mem_write##bwlq(*__addr, mem); \ 501 __addr++; \ 502 } \ 503 } \ 504 \ 505 static inline void reads##bwlq(volatile void __iomem *mem, void *addr, \ 506 unsigned int count) \ 507 { \ 508 volatile type *__addr = addr; \ 509 \ 510 while (count--) { \ 511 *__addr = __mem_read##bwlq(mem); \ 512 __addr++; \ 513 } \ 514 } 515 516 #define __BUILD_IOPORT_STRING(bwlq, type) \ 517 \ 518 static inline void outs##bwlq(unsigned long port, const void *addr, \ 519 unsigned int count) \ 520 { \ 521 const volatile type *__addr = addr; \ 522 \ 523 while (count--) { \ 524 __mem_out##bwlq(*__addr, port); \ 525 __addr++; \ 526 } \ 527 } \ 528 \ 529 static inline void ins##bwlq(unsigned long port, void *addr, \ 530 unsigned int count) \ 531 { \ 532 volatile type *__addr = addr; \ 533 \ 534 while (count--) { \ 535 *__addr = __mem_in##bwlq(port); \ 536 __addr++; \ 537 } \ 538 } 539 540 #define BUILDSTRING(bwlq, type) \ 541 \ 542 __BUILD_MEMORY_STRING(bwlq, type) \ 543 __BUILD_IOPORT_STRING(bwlq, type) 544 545 BUILDSTRING(b, u8) 546 BUILDSTRING(w, u16) 547 BUILDSTRING(l, u32) 548 #ifdef CONFIG_64BIT 549 BUILDSTRING(q, u64) 550 #endif 551 552 553 #ifdef CONFIG_CPU_CAVIUM_OCTEON 554 #define mmiowb() wmb() 555 #else 556 /* Depends on MIPS II instruction set */ 557 #define mmiowb() asm volatile ("sync" ::: "memory") 558 #endif 559 560 static inline void memset_io(volatile void __iomem *addr, unsigned char val, int count) 561 { 562 memset((void __force *) addr, val, count); 563 } 564 static inline void memcpy_fromio(void *dst, const volatile void __iomem *src, int count) 565 { 566 memcpy(dst, (void __force *) src, count); 567 } 568 static inline void memcpy_toio(volatile void __iomem *dst, const void *src, int count) 569 { 570 memcpy((void __force *) dst, src, count); 571 } 572 573 /* 574 * The caches on some architectures aren't dma-coherent and have need to 575 * handle this in software. There are three types of operations that 576 * can be applied to dma buffers. 577 * 578 * - dma_cache_wback_inv(start, size) makes caches and coherent by 579 * writing the content of the caches back to memory, if necessary. 580 * The function also invalidates the affected part of the caches as 581 * necessary before DMA transfers from outside to memory. 582 * - dma_cache_wback(start, size) makes caches and coherent by 583 * writing the content of the caches back to memory, if necessary. 584 * The function also invalidates the affected part of the caches as 585 * necessary before DMA transfers from outside to memory. 586 * - dma_cache_inv(start, size) invalidates the affected parts of the 587 * caches. Dirty lines of the caches may be written back or simply 588 * be discarded. This operation is necessary before dma operations 589 * to the memory. 590 * 591 * This API used to be exported; it now is for arch code internal use only. 592 */ 593 #if defined(CONFIG_DMA_NONCOHERENT) || defined(CONFIG_DMA_MAYBE_COHERENT) 594 595 extern void (*_dma_cache_wback_inv)(unsigned long start, unsigned long size); 596 extern void (*_dma_cache_wback)(unsigned long start, unsigned long size); 597 extern void (*_dma_cache_inv)(unsigned long start, unsigned long size); 598 599 #define dma_cache_wback_inv(start, size) _dma_cache_wback_inv(start, size) 600 #define dma_cache_wback(start, size) _dma_cache_wback(start, size) 601 #define dma_cache_inv(start, size) _dma_cache_inv(start, size) 602 603 #else /* Sane hardware */ 604 605 #define dma_cache_wback_inv(start,size) \ 606 do { (void) (start); (void) (size); } while (0) 607 #define dma_cache_wback(start,size) \ 608 do { (void) (start); (void) (size); } while (0) 609 #define dma_cache_inv(start,size) \ 610 do { (void) (start); (void) (size); } while (0) 611 612 #endif /* CONFIG_DMA_NONCOHERENT || CONFIG_DMA_MAYBE_COHERENT */ 613 614 /* 615 * Read a 32-bit register that requires a 64-bit read cycle on the bus. 616 * Avoid interrupt mucking, just adjust the address for 4-byte access. 617 * Assume the addresses are 8-byte aligned. 618 */ 619 #ifdef __MIPSEB__ 620 #define __CSR_32_ADJUST 4 621 #else 622 #define __CSR_32_ADJUST 0 623 #endif 624 625 #define csr_out32(v, a) (*(volatile u32 *)((unsigned long)(a) + __CSR_32_ADJUST) = (v)) 626 #define csr_in32(a) (*(volatile u32 *)((unsigned long)(a) + __CSR_32_ADJUST)) 627 628 /* 629 * Convert a physical pointer to a virtual kernel pointer for /dev/mem 630 * access 631 */ 632 #define xlate_dev_mem_ptr(p) __va(p) 633 634 /* 635 * Convert a virtual cached pointer to an uncached pointer 636 */ 637 #define xlate_dev_kmem_ptr(p) p 638 639 void __ioread64_copy(void *to, const void __iomem *from, size_t count); 640 641 #endif /* _ASM_IO_H */ 642