1 /* 2 * (C) Copyright IBM Corporation 2006 3 * All Rights Reserved. 4 * 5 * Permission is hereby granted, free of charge, to any person obtaining a 6 * copy of this software and associated documentation files (the "Software"), 7 * to deal in the Software without restriction, including without limitation 8 * on the rights to use, copy, modify, merge, publish, distribute, sub 9 * license, and/or sell copies of the Software, and to permit persons to whom 10 * the Software is furnished to do so, subject to the following conditions: 11 * 12 * The above copyright notice and this permission notice (including the next 13 * paragraph) shall be included in all copies or substantial portions of the 14 * Software. 15 * 16 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR 17 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, 18 * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL 19 * IBM AND/OR THEIR SUPPLIERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER 20 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING 21 * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER 22 * DEALINGS IN THE SOFTWARE. 23 */ 24 25 /** 26 * \file linux_sysfs.c 27 * Access PCI subsystem using Linux's sysfs interface. This interface is 28 * available starting somewhere in the late 2.5.x kernel phase, and is the 29 * prefered method on all 2.6.x kernels. 30 * 31 * \author Ian Romanick <[email protected]> 32 */ 33 34 #define _GNU_SOURCE 35 36 #include <stdlib.h> 37 #include <string.h> 38 #include <stdio.h> 39 #include <unistd.h> 40 #include <sys/types.h> 41 #include <sys/stat.h> 42 #include <fcntl.h> 43 #include <sys/mman.h> 44 #include <dirent.h> 45 #include <errno.h> 46 47 #include "pciaccess.h" 48 #include "pciaccess_private.h" 49 50 static int pci_device_linux_sysfs_read_rom( struct pci_device * dev, 51 void * buffer ); 52 53 static int pci_device_linux_sysfs_probe( struct pci_device * dev ); 54 55 static int pci_device_linux_sysfs_map_region( struct pci_device * dev, 56 unsigned region, int write_enable ); 57 58 static int pci_device_linux_sysfs_unmap_region( struct pci_device * dev, 59 unsigned region ); 60 61 static int pci_device_linux_sysfs_read( struct pci_device * dev, void * data, 62 pciaddr_t offset, pciaddr_t size, pciaddr_t * bytes_read ); 63 64 static int pci_device_linux_sysfs_write( struct pci_device * dev, 65 const void * data, pciaddr_t offset, pciaddr_t size, 66 pciaddr_t * bytes_wrtten ); 67 68 static const struct pci_system_methods linux_sysfs_methods = { 69 .destroy = NULL, 70 .destroy_device = NULL, 71 .read_rom = pci_device_linux_sysfs_read_rom, 72 .probe = pci_device_linux_sysfs_probe, 73 .map = pci_device_linux_sysfs_map_region, 74 .unmap = pci_device_linux_sysfs_unmap_region, 75 76 .read = pci_device_linux_sysfs_read, 77 .write = pci_device_linux_sysfs_write, 78 79 .fill_capabilities = pci_fill_capabilities_generic 80 }; 81 82 #define SYS_BUS_PCI "/sys/bus/pci/devices" 83 84 85 static void populate_entries( struct pci_system * pci_sys ); 86 87 88 /** 89 * Attempt to access PCI subsystem using Linux's sysfs interface. 90 */ 91 int 92 pci_system_linux_sysfs_create( void ) 93 { 94 int err = 0; 95 struct stat st; 96 97 98 /* If the directory "/sys/bus/pci/devices" exists, then the PCI subsystem 99 * can be accessed using this interface. 100 */ 101 102 if ( stat( SYS_BUS_PCI, & st ) == 0 ) { 103 pci_sys = calloc( 1, sizeof( struct pci_system ) ); 104 if ( pci_sys != NULL ) { 105 pci_sys->methods = & linux_sysfs_methods; 106 populate_entries( pci_sys ); 107 } 108 else { 109 err = ENOMEM; 110 } 111 } 112 else { 113 err = errno; 114 } 115 116 return err; 117 } 118 119 120 /** 121 * Filter out the names "." and ".." from the scanned sysfs entries. 122 * 123 * \param d Directory entry being processed by \c scandir. 124 * 125 * \return 126 * Zero if the entry name matches either "." or "..", non-zero otherwise. 127 * 128 * \sa scandir, populate_entries 129 */ 130 static int 131 scan_sys_pci_filter( const struct dirent * d ) 132 { 133 return !((strcmp( d->d_name, "." ) == 0) 134 || (strcmp( d->d_name, ".." ) == 0)); 135 } 136 137 138 void 139 populate_entries( struct pci_system * p ) 140 { 141 struct dirent ** devices; 142 int n; 143 int i; 144 145 146 n = scandir( SYS_BUS_PCI, & devices, scan_sys_pci_filter, alphasort ); 147 if ( n > 0 ) { 148 p->num_devices = n; 149 p->devices = calloc( n, sizeof( struct pci_device_private ) ); 150 151 152 for ( i = 0 ; i < n ; i++ ) { 153 unsigned dom, bus, dev, func; 154 155 156 sscanf( devices[ i ]->d_name, "%04x:%02x:%02x.%1u", 157 & dom, & bus, & dev, & func ); 158 159 p->devices[ i ].base.domain = dom; 160 p->devices[ i ].base.bus = bus; 161 p->devices[ i ].base.dev = dev; 162 p->devices[ i ].base.func = func; 163 } 164 } 165 } 166 167 168 static int 169 pci_device_linux_sysfs_probe( struct pci_device * dev ) 170 { 171 char name[256]; 172 uint8_t config[256]; 173 char resource[512]; 174 int fd; 175 pciaddr_t bytes; 176 unsigned i; 177 int err; 178 179 180 err = pci_device_linux_sysfs_read( dev, config, 0, 256, & bytes ); 181 if ( bytes >= 64 ) { 182 struct pci_device_private *priv = (struct pci_device_private *) dev; 183 184 dev->vendor_id = (uint16_t)config[0] + ((uint16_t)config[1] << 8); 185 dev->device_id = (uint16_t)config[2] + ((uint16_t)config[3] << 8); 186 dev->device_class = (uint32_t)config[9] + ((uint32_t)config[10] << 8) 187 + ((uint32_t)config[11] << 16); 188 dev->revision = config[8]; 189 dev->subvendor_id = (uint16_t)config[44] + ((uint16_t)config[45] << 8); 190 dev->subdevice_id = (uint16_t)config[46] + ((uint16_t)config[47] << 8); 191 dev->irq = config[60]; 192 193 priv->header_type = config[14]; 194 195 196 /* The PCI config registers can be used to obtain information 197 * about the memory and I/O regions for the device. However, 198 * doing so requires some tricky parsing (to correctly handle 199 * 64-bit memory regions) and requires writing to the config 200 * registers. Since we'd like to avoid having to deal with the 201 * parsing issues and non-root users can write to PCI config 202 * registers, we use a different file in the device's sysfs 203 * directory called "resource". 204 * 205 * The resource file contains all of the needed information in 206 * a format that is consistent across all platforms. Each BAR 207 * and the expansion ROM have a single line of data containing 208 * 3, 64-bit hex values: the first address in the region, 209 * the last address in the region, and the region's flags. 210 */ 211 snprintf( name, 255, "%s/%04x:%02x:%02x.%1u/resource", 212 SYS_BUS_PCI, 213 dev->domain, 214 dev->bus, 215 dev->dev, 216 dev->func ); 217 fd = open( name, O_RDONLY ); 218 if ( fd != -1 ) { 219 char * next; 220 pciaddr_t low_addr; 221 pciaddr_t high_addr; 222 pciaddr_t flags; 223 224 225 bytes = read( fd, resource, 512 ); 226 resource[511] = '\0'; 227 228 close( fd ); 229 230 next = resource; 231 for ( i = 0 ; i < 6 ; i++ ) { 232 233 dev->regions[i].base_addr = strtoull( next, & next, 16 ); 234 high_addr = strtoull( next, & next, 16 ); 235 flags = strtoull( next, & next, 16 ); 236 237 if ( dev->regions[i].base_addr != 0 ) { 238 dev->regions[i].size = (high_addr 239 - dev->regions[i].base_addr) + 1; 240 241 dev->regions[i].is_IO = (flags & 0x01); 242 dev->regions[i].is_64 = (flags & 0x04); 243 dev->regions[i].is_prefetchable = (flags & 0x08); 244 } 245 } 246 247 low_addr = strtoull( next, & next, 16 ); 248 high_addr = strtoull( next, & next, 16 ); 249 flags = strtoull( next, & next, 16 ); 250 if ( low_addr != 0 ) { 251 dev->rom_size = (high_addr - low_addr) + 1; 252 } 253 } 254 } 255 256 return err; 257 } 258 259 260 static int 261 pci_device_linux_sysfs_read_rom( struct pci_device * dev, void * buffer ) 262 { 263 char name[256]; 264 int fd; 265 struct stat st; 266 int err = 0; 267 size_t total_bytes; 268 269 270 snprintf( name, 255, "%s/%04x:%02x:%02x.%1u/rom", 271 SYS_BUS_PCI, 272 dev->domain, 273 dev->bus, 274 dev->dev, 275 dev->func ); 276 277 fd = open( name, O_RDWR ); 278 if ( fd == -1 ) { 279 return errno; 280 } 281 282 283 if ( fstat( fd, & st ) == -1 ) { 284 close( fd ); 285 return errno; 286 } 287 288 289 /* This is a quirky thing on Linux. Even though the ROM and the file 290 * for the ROM in sysfs are read-only, the string "1" must be written to 291 * the file to enable the ROM. After the data has been read, "0" must be 292 * written to the file to disable the ROM. 293 */ 294 write( fd, "1", 1 ); 295 lseek( fd, 0, SEEK_SET ); 296 297 for ( total_bytes = 0 ; total_bytes < st.st_size ; /* empty */ ) { 298 const int bytes = read( fd, (char *) buffer + total_bytes, 299 st.st_size - total_bytes ); 300 if ( bytes == -1 ) { 301 err = errno; 302 break; 303 } 304 else if ( bytes == 0 ) { 305 break; 306 } 307 308 total_bytes += bytes; 309 } 310 311 312 lseek( fd, 0, SEEK_SET ); 313 write( fd, "0", 1 ); 314 315 close( fd ); 316 return err; 317 } 318 319 320 static int 321 pci_device_linux_sysfs_read( struct pci_device * dev, void * data, 322 pciaddr_t offset, pciaddr_t size, 323 pciaddr_t * bytes_read ) 324 { 325 char name[256]; 326 pciaddr_t temp_size = size; 327 int err = 0; 328 int fd; 329 330 331 if ( bytes_read != NULL ) { 332 *bytes_read = 0; 333 } 334 335 /* Each device has a directory under sysfs. Within that directory there 336 * is a file named "config". This file used to access the PCI config 337 * space. It is used here to obtain most of the information about the 338 * device. 339 */ 340 snprintf( name, 255, "%s/%04x:%02x:%02x.%1u/config", 341 SYS_BUS_PCI, 342 dev->domain, 343 dev->bus, 344 dev->dev, 345 dev->func ); 346 347 fd = open( name, O_RDONLY ); 348 if ( fd == -1 ) { 349 return errno; 350 } 351 352 353 while ( temp_size > 0 ) { 354 const ssize_t bytes = pread64( fd, data, temp_size, offset ); 355 356 /* If zero bytes were read, then we assume it's the end of the 357 * config file. 358 */ 359 if ( bytes <= 0 ) { 360 err = errno; 361 break; 362 } 363 364 temp_size -= bytes; 365 offset += bytes; 366 data += bytes; 367 } 368 369 if ( bytes_read != NULL ) { 370 *bytes_read = size - temp_size; 371 } 372 373 close( fd ); 374 return err; 375 } 376 377 378 static int 379 pci_device_linux_sysfs_write( struct pci_device * dev, const void * data, 380 pciaddr_t offset, pciaddr_t size, 381 pciaddr_t * bytes_written ) 382 { 383 char name[256]; 384 pciaddr_t temp_size = size; 385 int err = 0; 386 int fd; 387 388 389 if ( bytes_written != NULL ) { 390 *bytes_written = 0; 391 } 392 393 /* Each device has a directory under sysfs. Within that directory there 394 * is a file named "config". This file used to access the PCI config 395 * space. It is used here to obtain most of the information about the 396 * device. 397 */ 398 snprintf( name, 255, "%s/%04x:%02x:%02x.%1u/config", 399 SYS_BUS_PCI, 400 dev->domain, 401 dev->bus, 402 dev->dev, 403 dev->func ); 404 405 fd = open( name, O_WRONLY ); 406 if ( fd == -1 ) { 407 return errno; 408 } 409 410 411 while ( temp_size > 0 ) { 412 const ssize_t bytes = pwrite64( fd, data, temp_size, offset ); 413 414 /* If zero bytes were written, then we assume it's the end of the 415 * config file. 416 */ 417 if ( bytes <= 0 ) { 418 err = errno; 419 break; 420 } 421 422 temp_size -= bytes; 423 offset += bytes; 424 data += bytes; 425 } 426 427 if ( bytes_written != NULL ) { 428 *bytes_written = size - temp_size; 429 } 430 431 close( fd ); 432 return err; 433 } 434 435 436 /** 437 * Map a memory region for a device using the Linux sysfs interface. 438 * 439 * \param dev Device whose memory region is to be mapped. 440 * \param region Region, on the range [0, 5], that is to be mapped. 441 * \param write_enable Map for writing (non-zero). 442 * 443 * \return 444 * Zero on success or an \c errno value on failure. 445 * 446 * \sa pci_device_map_region, pci_device_linux_sysfs_unmap_region 447 * 448 * \todo 449 * Some older 2.6.x kernels don't implement the resourceN files. On those 450 * systems /dev/mem must be used. On these systems it is also possible that 451 * \c mmap64 may need to be used. 452 */ 453 static int 454 pci_device_linux_sysfs_map_region( struct pci_device * dev, unsigned region, 455 int write_enable ) 456 { 457 char name[256]; 458 int fd; 459 int err = 0; 460 const int prot = (write_enable) ? (PROT_READ | PROT_WRITE) : PROT_READ; 461 462 463 snprintf( name, 255, "%s/%04x:%02x:%02x.%1u/resource%u", 464 SYS_BUS_PCI, 465 dev->domain, 466 dev->bus, 467 dev->dev, 468 dev->func, 469 region ); 470 471 fd = open( name, (write_enable) ? O_RDWR : O_RDONLY ); 472 if ( fd == -1 ) { 473 return errno; 474 } 475 476 477 dev->regions[ region ].memory = mmap( NULL, dev->regions[ region ].size, 478 prot, MAP_SHARED, fd, 0 ); 479 if ( dev->regions[ region ].memory == MAP_FAILED ) { 480 err = errno; 481 dev->regions[ region ].memory = NULL; 482 } 483 484 close( fd ); 485 return err; 486 } 487 488 489 /** 490 * Unmap the specified region using the Linux sysfs interface. 491 * 492 * \param dev Device whose memory region is to be mapped. 493 * \param region Region, on the range [0, 5], that is to be mapped. 494 * 495 * \return 496 * Zero on success or an \c errno value on failure. 497 * 498 * \sa pci_device_unmap_region, pci_device_linux_sysfs_map_region 499 * 500 * \todo 501 * Some older 2.6.x kernels don't implement the resourceN files. On those 502 * systems /dev/mem must be used. On these systems it is also possible that 503 * \c mmap64 may need to be used. 504 */ 505 static int 506 pci_device_linux_sysfs_unmap_region( struct pci_device * dev, unsigned region ) 507 { 508 int err = 0; 509 510 if ( munmap( dev->regions[ region ].memory, dev->regions[ region ].size ) 511 == -1 ) { 512 err = errno; 513 } 514 515 dev->regions[ region ].memory = NULL; 516 517 return err; 518 } 519