| 1 | |
| 2 | |
| 3 | * arch/parisc/lib/io.c
|
| 4 | *
|
| 5 | * Copyright (c) Matthew Wilcox 2001 for Hewlett-Packard
|
| 6 | * Copyright (c) Randolph Chung 2001 <[email protected]>
|
| 7 | *
|
| 8 | * IO accessing functions which shouldn't be inlined because they're too big
|
| 9 |
|
| 10 |
|
| 11 | #include <linux/kernel.h>
|
| 12 | #include <linux/module.h>
|
| 13 | #include <asm/io.h>
|
| 14 |
|
| 15 | |
| 16 | * Read COUNT 8-bit bytes from port PORT into memory starting at
|
| 17 | * SRC.
|
| 18 |
|
| 19 | void insb (unsigned long port, void *dst, unsigned long count)
|
| 20 | {
|
| 21 | unsigned char *p;
|
| 22 |
|
| 23 | p = (unsigned char *)dst;
|
| 24 |
|
| 25 | while (((unsigned long)p) & 0x3) {
|
| 26 | if (!count)
|
| 27 | return;
|
| 28 | count--;
|
| 29 | *p = inb(port);
|
| 30 | p++;
|
| 31 | }
|
| 32 |
|
| 33 | while (count >= 4) {
|
| 34 | unsigned int w;
|
| 35 | count -= 4;
|
| 36 | w = inb(port) << 24;
|
| 37 | w |= inb(port) << 16;
|
| 38 | w |= inb(port) << 8;
|
| 39 | w |= inb(port);
|
| 40 | *(unsigned int *) p = w;
|
| 41 | p += 4;
|
| 42 | }
|
| 43 |
|
| 44 | while (count) {
|
| 45 | --count;
|
| 46 | *p = inb(port);
|
| 47 | p++;
|
| 48 | }
|
| 49 | }
|
| 50 |
|
| 51 |
|
| 52 | |
| 53 | * Read COUNT 16-bit words from port PORT into memory starting at
|
| 54 | * SRC. SRC must be at least short aligned. This is used by the
|
| 55 | * IDE driver to read disk sectors. Performance is important, but
|
| 56 | * the interfaces seems to be slow: just using the inlined version
|
| 57 | * of the inw() breaks things.
|
| 58 |
|
| 59 | void insw (unsigned long port, void *dst, unsigned long count)
|
| 60 | {
|
| 61 | unsigned int l = 0, l2;
|
| 62 | unsigned char *p;
|
| 63 |
|
| 64 | p = (unsigned char *)dst;
|
| 65 |
|
| 66 | if (!count)
|
| 67 | return;
|
| 68 |
|
| 69 | switch (((unsigned long)p) & 0x3)
|
| 70 | {
|
| 71 | case 0x00:
|
| 72 | while (count>=2) {
|
| 73 |
|
| 74 | count -= 2;
|
| 75 | l = cpu_to_le16(inw(port)) << 16;
|
| 76 | l |= cpu_to_le16(inw(port));
|
| 77 | *(unsigned int *)p = l;
|
| 78 | p += 4;
|
| 79 | }
|
| 80 | if (count) {
|
| 81 | *(unsigned short *)p = cpu_to_le16(inw(port));
|
| 82 | }
|
| 83 | break;
|
| 84 |
|
| 85 | case 0x02:
|
| 86 | *(unsigned short *)p = cpu_to_le16(inw(port));
|
| 87 | p += 2;
|
| 88 | count--;
|
| 89 | while (count>=2) {
|
| 90 |
|
| 91 | count -= 2;
|
| 92 | l = cpu_to_le16(inw(port)) << 16;
|
| 93 | l |= cpu_to_le16(inw(port));
|
| 94 | *(unsigned int *)p = l;
|
| 95 | p += 4;
|
| 96 | }
|
| 97 | if (count) {
|
| 98 | *(unsigned short *)p = cpu_to_le16(inw(port));
|
| 99 | }
|
| 100 | break;
|
| 101 |
|
| 102 | case 0x01:
|
| 103 | case 0x03:
|
| 104 | |
| 105 | * in this case, 16bit will have to do -- DE
|
| 106 | --count;
|
| 107 |
|
| 108 | l = cpu_to_le16(inw(port));
|
| 109 | *p = l >> 8;
|
| 110 | p++;
|
| 111 | while (count--)
|
| 112 | {
|
| 113 | l2 = cpu_to_le16(inw(port));
|
| 114 | *(unsigned short *)p = (l & 0xff) << 8 | (l2 >> 8);
|
| 115 | p += 2;
|
| 116 | l = l2;
|
| 117 | }
|
| 118 | *p = l & 0xff;
|
| 119 | break;
|
| 120 | }
|
| 121 | }
|
| 122 |
|
| 123 |
|
| 124 |
|
| 125 | |
| 126 | * Read COUNT 32-bit words from port PORT into memory starting at
|
| 127 | * SRC. Now works with any alignment in SRC. Performance is important,
|
| 128 | * but the interfaces seems to be slow: just using the inlined version
|
| 129 | * of the inl() breaks things.
|
| 130 |
|
| 131 | void insl (unsigned long port, void *dst, unsigned long count)
|
| 132 | {
|
| 133 | unsigned int l = 0, l2;
|
| 134 | unsigned char *p;
|
| 135 |
|
| 136 | p = (unsigned char *)dst;
|
| 137 |
|
| 138 | if (!count)
|
| 139 | return;
|
| 140 |
|
| 141 | switch (((unsigned long) dst) & 0x3)
|
| 142 | {
|
| 143 | case 0x00:
|
| 144 | while (count--)
|
| 145 | {
|
| 146 | *(unsigned int *)p = cpu_to_le32(inl(port));
|
| 147 | p += 4;
|
| 148 | }
|
| 149 | break;
|
| 150 |
|
| 151 | case 0x02:
|
| 152 | --count;
|
| 153 |
|
| 154 | l = cpu_to_le32(inl(port));
|
| 155 | *(unsigned short *)p = l >> 16;
|
| 156 | p += 2;
|
| 157 |
|
| 158 | while (count--)
|
| 159 | {
|
| 160 | l2 = cpu_to_le32(inl(port));
|
| 161 | *(unsigned int *)p = (l & 0xffff) << 16 | (l2 >> 16);
|
| 162 | p += 4;
|
| 163 | l = l2;
|
| 164 | }
|
| 165 | *(unsigned short *)p = l & 0xffff;
|
| 166 | break;
|
| 167 | case 0x01:
|
| 168 | --count;
|
| 169 |
|
| 170 | l = cpu_to_le32(inl(port));
|
| 171 | *(unsigned char *)p = l >> 24;
|
| 172 | p++;
|
| 173 | *(unsigned short *)p = (l >> 8) & 0xffff;
|
| 174 | p += 2;
|
| 175 | while (count--)
|
| 176 | {
|
| 177 | l2 = cpu_to_le32(inl(port));
|
| 178 | *(unsigned int *)p = (l & 0xff) << 24 | (l2 >> 8);
|
| 179 | p += 4;
|
| 180 | l = l2;
|
| 181 | }
|
| 182 | *p = l & 0xff;
|
| 183 | break;
|
| 184 | case 0x03:
|
| 185 | --count;
|
| 186 |
|
| 187 | l = cpu_to_le32(inl(port));
|
| 188 | *p = l >> 24;
|
| 189 | p++;
|
| 190 | while (count--)
|
| 191 | {
|
| 192 | l2 = cpu_to_le32(inl(port));
|
| 193 | *(unsigned int *)p = (l & 0xffffff) << 8 | l2 >> 24;
|
| 194 | p += 4;
|
| 195 | l = l2;
|
| 196 | }
|
| 197 | *(unsigned short *)p = (l >> 8) & 0xffff;
|
| 198 | p += 2;
|
| 199 | *p = l & 0xff;
|
| 200 | break;
|
| 201 | }
|
| 202 | }
|
| 203 |
|
| 204 |
|
| 205 | |
| 206 | * Like insb but in the opposite direction.
|
| 207 | * Don't worry as much about doing aligned memory transfers:
|
| 208 | * doing byte reads the "slow" way isn't nearly as slow as
|
| 209 | * doing byte writes the slow way (no r-m-w cycle).
|
| 210 |
|
| 211 | void outsb(unsigned long port, const void * src, unsigned long count)
|
| 212 | {
|
| 213 | const unsigned char *p;
|
| 214 |
|
| 215 | p = (const unsigned char *)src;
|
| 216 | while (count) {
|
| 217 | count--;
|
| 218 | outb(*p, port);
|
| 219 | p++;
|
| 220 | }
|
| 221 | }
|
| 222 |
|
| 223 | |
| 224 | * Like insw but in the opposite direction. This is used by the IDE
|
| 225 | * driver to write disk sectors. Performance is important, but the
|
| 226 | * interfaces seems to be slow: just using the inlined version of the
|
| 227 | * outw() breaks things.
|
| 228 |
|
| 229 | void outsw (unsigned long port, const void *src, unsigned long count)
|
| 230 | {
|
| 231 | unsigned int l = 0, l2;
|
| 232 | const unsigned char *p;
|
| 233 |
|
| 234 | p = (const unsigned char *)src;
|
| 235 |
|
| 236 | if (!count)
|
| 237 | return;
|
| 238 |
|
| 239 | switch (((unsigned long)p) & 0x3)
|
| 240 | {
|
| 241 | case 0x00:
|
| 242 | while (count>=2) {
|
| 243 | count -= 2;
|
| 244 | l = *(unsigned int *)p;
|
| 245 | p += 4;
|
| 246 | outw(le16_to_cpu(l >> 16), port);
|
| 247 | outw(le16_to_cpu(l & 0xffff), port);
|
| 248 | }
|
| 249 | if (count) {
|
| 250 | outw(le16_to_cpu(*(unsigned short*)p), port);
|
| 251 | }
|
| 252 | break;
|
| 253 |
|
| 254 | case 0x02:
|
| 255 |
|
| 256 | outw(le16_to_cpu(*(unsigned short*)p), port);
|
| 257 | p += 2;
|
| 258 | count--;
|
| 259 |
|
| 260 | while (count>=2) {
|
| 261 | count -= 2;
|
| 262 | l = *(unsigned int *)p;
|
| 263 | p += 4;
|
| 264 | outw(le16_to_cpu(l >> 16), port);
|
| 265 | outw(le16_to_cpu(l & 0xffff), port);
|
| 266 | }
|
| 267 | if (count) {
|
| 268 | outw(le16_to_cpu(*(unsigned short *)p), port);
|
| 269 | }
|
| 270 | break;
|
| 271 |
|
| 272 | case 0x01:
|
| 273 | |
| 274 | * in this case, 16bit will have to do -- DE
|
| 275 |
|
| 276 | l = *p << 8;
|
| 277 | p++;
|
| 278 | count--;
|
| 279 | while (count)
|
| 280 | {
|
| 281 | count--;
|
| 282 | l2 = *(unsigned short *)p;
|
| 283 | p += 2;
|
| 284 | outw(le16_to_cpu(l | l2 >> 8), port);
|
| 285 | l = l2 << 8;
|
| 286 | }
|
| 287 | l2 = *(unsigned char *)p;
|
| 288 | outw (le16_to_cpu(l | l2>>8), port);
|
| 289 | break;
|
| 290 |
|
| 291 | }
|
| 292 | }
|
| 293 |
|
| 294 |
|
| 295 | |
| 296 | * Like insl but in the opposite direction. This is used by the IDE
|
| 297 | * driver to write disk sectors. Works with any alignment in SRC.
|
| 298 | * Performance is important, but the interfaces seems to be slow:
|
| 299 | * just using the inlined version of the outl() breaks things.
|
| 300 |
|
| 301 | void outsl (unsigned long port, const void *src, unsigned long count)
|
| 302 | {
|
| 303 | unsigned int l = 0, l2;
|
| 304 | const unsigned char *p;
|
| 305 |
|
| 306 | p = (const unsigned char *)src;
|
| 307 |
|
| 308 | if (!count)
|
| 309 | return;
|
| 310 |
|
| 311 | switch (((unsigned long)p) & 0x3)
|
| 312 | {
|
| 313 | case 0x00:
|
| 314 | while (count--)
|
| 315 | {
|
| 316 | outl(le32_to_cpu(*(unsigned int *)p), port);
|
| 317 | p += 4;
|
| 318 | }
|
| 319 | break;
|
| 320 |
|
| 321 | case 0x02:
|
| 322 | --count;
|
| 323 |
|
| 324 | l = *(unsigned short *)p;
|
| 325 | p += 2;
|
| 326 |
|
| 327 | while (count--)
|
| 328 | {
|
| 329 | l2 = *(unsigned int *)p;
|
| 330 | p += 4;
|
| 331 | outl (le32_to_cpu(l << 16 | l2 >> 16), port);
|
| 332 | l = l2;
|
| 333 | }
|
| 334 | l2 = *(unsigned short *)p;
|
| 335 | outl (le32_to_cpu(l << 16 | l2), port);
|
| 336 | break;
|
| 337 | case 0x01:
|
| 338 | --count;
|
| 339 |
|
| 340 | l = *p << 24;
|
| 341 | p++;
|
| 342 | l |= *(unsigned short *)p << 8;
|
| 343 | p += 2;
|
| 344 |
|
| 345 | while (count--)
|
| 346 | {
|
| 347 | l2 = *(unsigned int *)p;
|
| 348 | p += 4;
|
| 349 | outl (le32_to_cpu(l | l2 >> 24), port);
|
| 350 | l = l2 << 8;
|
| 351 | }
|
| 352 | l2 = *p;
|
| 353 | outl (le32_to_cpu(l | l2), port);
|
| 354 | break;
|
| 355 | case 0x03:
|
| 356 | --count;
|
| 357 |
|
| 358 | l = *p << 24;
|
| 359 | p++;
|
| 360 |
|
| 361 | while (count--)
|
| 362 | {
|
| 363 | l2 = *(unsigned int *)p;
|
| 364 | p += 4;
|
| 365 | outl (le32_to_cpu(l | l2 >> 8), port);
|
| 366 | l = l2 << 24;
|
| 367 | }
|
| 368 | l2 = *(unsigned short *)p << 16;
|
| 369 | p += 2;
|
| 370 | l2 |= *p;
|
| 371 | outl (le32_to_cpu(l | l2), port);
|
| 372 | break;
|
| 373 | }
|
| 374 | }
|
| 375 |
|
| 376 | EXPORT_SYMBOL(insb);
|
| 377 | EXPORT_SYMBOL(insw);
|
| 378 | EXPORT_SYMBOL(insl);
|
| 379 | EXPORT_SYMBOL(outsb);
|
| 380 | EXPORT_SYMBOL(outsw);
|
| 381 | EXPORT_SYMBOL(outsl);
|