2 * (C) Copyright IBM Corporation 2006
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:
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
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.
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.
31 * \author Ian Romanick <idr@us.ibm.com>
40 #include <sys/types.h>
51 #include <sys/ioctl.h>
54 #include "pciaccess.h"
55 #include "pciaccess_private.h"
56 #include "linux_devmem.h"
58 static int pci_device_linux_sysfs_read_rom( struct pci_device * dev,
61 static int pci_device_linux_sysfs_probe( struct pci_device * dev );
63 static int pci_device_linux_sysfs_map_range(struct pci_device *dev,
64 struct pci_device_mapping *map);
66 static int pci_device_linux_sysfs_read( struct pci_device * dev, void * data,
67 pciaddr_t offset, pciaddr_t size, pciaddr_t * bytes_read );
69 static int pci_device_linux_sysfs_write( struct pci_device * dev,
70 const void * data, pciaddr_t offset, pciaddr_t size,
71 pciaddr_t * bytes_wrtten );
73 static const struct pci_system_methods linux_sysfs_methods = {
75 .destroy_device = NULL,
76 .read_rom = pci_device_linux_sysfs_read_rom,
77 .probe = pci_device_linux_sysfs_probe,
78 .map_range = pci_device_linux_sysfs_map_range,
79 .unmap_range = pci_device_generic_unmap_range,
81 .read = pci_device_linux_sysfs_read,
82 .write = pci_device_linux_sysfs_write,
84 .fill_capabilities = pci_fill_capabilities_generic
87 #define SYS_BUS_PCI "/sys/bus/pci/devices"
90 static int populate_entries(struct pci_system * pci_sys);
94 * Attempt to access PCI subsystem using Linux's sysfs interface.
97 pci_system_linux_sysfs_create( void )
103 /* If the directory "/sys/bus/pci/devices" exists, then the PCI subsystem
104 * can be accessed using this interface.
107 if ( stat( SYS_BUS_PCI, & st ) == 0 ) {
108 pci_sys = calloc( 1, sizeof( struct pci_system ) );
109 if ( pci_sys != NULL ) {
110 pci_sys->methods = & linux_sysfs_methods;
111 err = populate_entries(pci_sys);
122 pci_sys->mtrr_fd = open("/proc/mtrr", O_WRONLY);
130 * Filter out the names "." and ".." from the scanned sysfs entries.
132 * \param d Directory entry being processed by \c scandir.
135 * Zero if the entry name matches either "." or "..", non-zero otherwise.
137 * \sa scandir, populate_entries
140 scan_sys_pci_filter( const struct dirent * d )
142 return !((strcmp( d->d_name, "." ) == 0)
143 || (strcmp( d->d_name, ".." ) == 0));
148 populate_entries( struct pci_system * p )
150 struct dirent ** devices;
156 n = scandir( SYS_BUS_PCI, & devices, scan_sys_pci_filter, alphasort );
159 p->devices = calloc( n, sizeof( struct pci_device_private ) );
161 if (p->devices != NULL) {
162 for (i = 0 ; i < n ; i++) {
165 unsigned dom, bus, dev, func;
166 struct pci_device_private *device =
167 (struct pci_device_private *) &p->devices[i];
170 sscanf(devices[i]->d_name, "%04x:%02x:%02x.%1u",
171 & dom, & bus, & dev, & func);
173 device->base.domain = dom;
174 device->base.bus = bus;
175 device->base.dev = dev;
176 device->base.func = func;
179 err = pci_device_linux_sysfs_read(& device->base, config, 0,
181 if ((bytes == 48) && !err) {
182 device->base.vendor_id = (uint16_t)config[0]
183 + ((uint16_t)config[1] << 8);
184 device->base.device_id = (uint16_t)config[2]
185 + ((uint16_t)config[3] << 8);
186 device->base.device_class = (uint32_t)config[9]
187 + ((uint32_t)config[10] << 8)
188 + ((uint32_t)config[11] << 16);
189 device->base.revision = config[8];
190 device->base.subvendor_id = (uint16_t)config[44]
191 + ((uint16_t)config[45] << 8);
192 device->base.subdevice_id = (uint16_t)config[46]
193 + ((uint16_t)config[47] << 8);
216 pci_device_linux_sysfs_probe( struct pci_device * dev )
227 err = pci_device_linux_sysfs_read( dev, config, 0, 256, & bytes );
229 struct pci_device_private *priv = (struct pci_device_private *) dev;
231 dev->irq = config[60];
232 priv->header_type = config[14];
235 /* The PCI config registers can be used to obtain information
236 * about the memory and I/O regions for the device. However,
237 * doing so requires some tricky parsing (to correctly handle
238 * 64-bit memory regions) and requires writing to the config
239 * registers. Since we'd like to avoid having to deal with the
240 * parsing issues and non-root users can write to PCI config
241 * registers, we use a different file in the device's sysfs
242 * directory called "resource".
244 * The resource file contains all of the needed information in
245 * a format that is consistent across all platforms. Each BAR
246 * and the expansion ROM have a single line of data containing
247 * 3, 64-bit hex values: the first address in the region,
248 * the last address in the region, and the region's flags.
250 snprintf( name, 255, "%s/%04x:%02x:%02x.%1u/resource",
256 fd = open( name, O_RDONLY );
264 bytes = read( fd, resource, 512 );
265 resource[511] = '\0';
270 for ( i = 0 ; i < 6 ; i++ ) {
272 dev->regions[i].base_addr = strtoull( next, & next, 16 );
273 high_addr = strtoull( next, & next, 16 );
274 flags = strtoull( next, & next, 16 );
276 if ( dev->regions[i].base_addr != 0 ) {
277 dev->regions[i].size = (high_addr
278 - dev->regions[i].base_addr) + 1;
280 dev->regions[i].is_IO = (flags & 0x01);
281 dev->regions[i].is_64 = (flags & 0x04);
282 dev->regions[i].is_prefetchable = (flags & 0x08);
286 low_addr = strtoull( next, & next, 16 );
287 high_addr = strtoull( next, & next, 16 );
288 flags = strtoull( next, & next, 16 );
289 if ( low_addr != 0 ) {
290 priv->rom_base = low_addr;
291 dev->rom_size = (high_addr - low_addr) + 1;
301 pci_device_linux_sysfs_read_rom( struct pci_device * dev, void * buffer )
310 snprintf( name, 255, "%s/%04x:%02x:%02x.%1u/rom",
317 fd = open( name, O_RDWR );
319 /* If reading the ROM using sysfs fails, fall back to the old
320 * /dev/mem based interface.
322 return pci_device_linux_devmem_read_rom(dev, buffer);
326 if ( fstat( fd, & st ) == -1 ) {
332 /* This is a quirky thing on Linux. Even though the ROM and the file
333 * for the ROM in sysfs are read-only, the string "1" must be written to
334 * the file to enable the ROM. After the data has been read, "0" must be
335 * written to the file to disable the ROM.
338 lseek( fd, 0, SEEK_SET );
340 for ( total_bytes = 0 ; total_bytes < st.st_size ; /* empty */ ) {
341 const int bytes = read( fd, (char *) buffer + total_bytes,
342 st.st_size - total_bytes );
347 else if ( bytes == 0 ) {
351 total_bytes += bytes;
355 lseek( fd, 0, SEEK_SET );
364 pci_device_linux_sysfs_read( struct pci_device * dev, void * data,
365 pciaddr_t offset, pciaddr_t size,
366 pciaddr_t * bytes_read )
369 pciaddr_t temp_size = size;
374 if ( bytes_read != NULL ) {
378 /* Each device has a directory under sysfs. Within that directory there
379 * is a file named "config". This file used to access the PCI config
380 * space. It is used here to obtain most of the information about the
383 snprintf( name, 255, "%s/%04x:%02x:%02x.%1u/config",
390 fd = open( name, O_RDONLY );
396 while ( temp_size > 0 ) {
397 const ssize_t bytes = pread64( fd, data, temp_size, offset );
399 /* If zero bytes were read, then we assume it's the end of the
412 if ( bytes_read != NULL ) {
413 *bytes_read = size - temp_size;
422 pci_device_linux_sysfs_write( struct pci_device * dev, const void * data,
423 pciaddr_t offset, pciaddr_t size,
424 pciaddr_t * bytes_written )
427 pciaddr_t temp_size = size;
432 if ( bytes_written != NULL ) {
436 /* Each device has a directory under sysfs. Within that directory there
437 * is a file named "config". This file used to access the PCI config
438 * space. It is used here to obtain most of the information about the
441 snprintf( name, 255, "%s/%04x:%02x:%02x.%1u/config",
448 fd = open( name, O_WRONLY );
454 while ( temp_size > 0 ) {
455 const ssize_t bytes = pwrite64( fd, data, temp_size, offset );
457 /* If zero bytes were written, then we assume it's the end of the
470 if ( bytes_written != NULL ) {
471 *bytes_written = size - temp_size;
480 * Map a memory region for a device using the Linux sysfs interface.
482 * \param dev Device whose memory region is to be mapped.
483 * \param map Parameters of the mapping that is to be created.
486 * Zero on success or an \c errno value on failure.
488 * \sa pci_device_map_rrange, pci_device_linux_sysfs_unmap_range
491 * Some older 2.6.x kernels don't implement the resourceN files. On those
492 * systems /dev/mem must be used. On these systems it is also possible that
493 * \c mmap64 may need to be used.
496 pci_device_linux_sysfs_map_range(struct pci_device *dev,
497 struct pci_device_mapping *map)
502 const int prot = ((map->flags & PCI_DEV_MAP_FLAG_WRITABLE) != 0)
503 ? (PROT_READ | PROT_WRITE) : PROT_READ;
504 const int open_flags = ((map->flags & PCI_DEV_MAP_FLAG_WRITABLE) != 0)
506 const off_t offset = map->base - dev->regions[map->region].base_addr;
508 struct mtrr_sentry sentry = {
511 .type = MTRR_TYPE_UNCACHABLE
515 snprintf(name, 255, "%s/%04x:%02x:%02x.%1u/resource%u",
523 fd = open(name, open_flags);
529 map->memory = mmap(NULL, map->size, prot, MAP_SHARED, fd, offset);
530 if (map->memory == MAP_FAILED) {
538 if ((map->flags & PCI_DEV_MAP_FLAG_CACHABLE) != 0) {
539 sentry.type = MTRR_TYPE_WRBACK;
540 } else if ((map->flags & PCI_DEV_MAP_FLAG_WRITE_COMBINE) != 0) {
541 sentry.type = MTRR_TYPE_WRCOMB;
544 if (pci_sys->mtrr_fd != -1) {
545 if (ioctl(pci_sys->mtrr_fd, MTRRIOC_ADD_ENTRY, &sentry) < 0) {
546 /* FIXME: Should we report an error in this case?
548 fprintf(stderr, "error setting MTRR "
549 "(base = 0x%08lx, size = 0x%08x, type = %u) %s (%d)\n",
550 sentry.base, sentry.size, sentry.type,
551 strerror(errno), errno);