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_unmap_range(struct pci_device *dev,
67 struct pci_device_mapping *map);
69 static int pci_device_linux_sysfs_read( struct pci_device * dev, void * data,
70 pciaddr_t offset, pciaddr_t size, pciaddr_t * bytes_read );
72 static int pci_device_linux_sysfs_write( struct pci_device * dev,
73 const void * data, pciaddr_t offset, pciaddr_t size,
74 pciaddr_t * bytes_wrtten );
76 static const struct pci_system_methods linux_sysfs_methods = {
78 .destroy_device = NULL,
79 .read_rom = pci_device_linux_sysfs_read_rom,
80 .probe = pci_device_linux_sysfs_probe,
81 .map_range = pci_device_linux_sysfs_map_range,
82 .unmap_range = pci_device_linux_sysfs_unmap_range,
84 .read = pci_device_linux_sysfs_read,
85 .write = pci_device_linux_sysfs_write,
87 .fill_capabilities = pci_fill_capabilities_generic
90 #define SYS_BUS_PCI "/sys/bus/pci/devices"
93 static int populate_entries(struct pci_system * pci_sys);
97 * Attempt to access PCI subsystem using Linux's sysfs interface.
100 pci_system_linux_sysfs_create( void )
106 /* If the directory "/sys/bus/pci/devices" exists, then the PCI subsystem
107 * can be accessed using this interface.
110 if ( stat( SYS_BUS_PCI, & st ) == 0 ) {
111 pci_sys = calloc( 1, sizeof( struct pci_system ) );
112 if ( pci_sys != NULL ) {
113 pci_sys->methods = & linux_sysfs_methods;
114 err = populate_entries(pci_sys);
125 pci_sys->mtrr_fd = open("/proc/mtrr", O_WRONLY);
133 * Filter out the names "." and ".." from the scanned sysfs entries.
135 * \param d Directory entry being processed by \c scandir.
138 * Zero if the entry name matches either "." or "..", non-zero otherwise.
140 * \sa scandir, populate_entries
143 scan_sys_pci_filter( const struct dirent * d )
145 return !((strcmp( d->d_name, "." ) == 0)
146 || (strcmp( d->d_name, ".." ) == 0));
151 populate_entries( struct pci_system * p )
153 struct dirent ** devices;
159 n = scandir( SYS_BUS_PCI, & devices, scan_sys_pci_filter, alphasort );
162 p->devices = calloc( n, sizeof( struct pci_device_private ) );
164 if (p->devices != NULL) {
165 for (i = 0 ; i < n ; i++) {
168 unsigned dom, bus, dev, func;
169 struct pci_device_private *device =
170 (struct pci_device_private *) &p->devices[i];
173 sscanf(devices[i]->d_name, "%04x:%02x:%02x.%1u",
174 & dom, & bus, & dev, & func);
176 device->base.domain = dom;
177 device->base.bus = bus;
178 device->base.dev = dev;
179 device->base.func = func;
182 err = pci_device_linux_sysfs_read(& device->base, config, 0,
184 if ((bytes == 48) && !err) {
185 device->base.vendor_id = (uint16_t)config[0]
186 + ((uint16_t)config[1] << 8);
187 device->base.device_id = (uint16_t)config[2]
188 + ((uint16_t)config[3] << 8);
189 device->base.device_class = (uint32_t)config[9]
190 + ((uint32_t)config[10] << 8)
191 + ((uint32_t)config[11] << 16);
192 device->base.revision = config[8];
193 device->base.subvendor_id = (uint16_t)config[44]
194 + ((uint16_t)config[45] << 8);
195 device->base.subdevice_id = (uint16_t)config[46]
196 + ((uint16_t)config[47] << 8);
219 pci_device_linux_sysfs_probe( struct pci_device * dev )
230 err = pci_device_linux_sysfs_read( dev, config, 0, 256, & bytes );
232 struct pci_device_private *priv = (struct pci_device_private *) dev;
234 dev->irq = config[60];
235 priv->header_type = config[14];
238 /* The PCI config registers can be used to obtain information
239 * about the memory and I/O regions for the device. However,
240 * doing so requires some tricky parsing (to correctly handle
241 * 64-bit memory regions) and requires writing to the config
242 * registers. Since we'd like to avoid having to deal with the
243 * parsing issues and non-root users can write to PCI config
244 * registers, we use a different file in the device's sysfs
245 * directory called "resource".
247 * The resource file contains all of the needed information in
248 * a format that is consistent across all platforms. Each BAR
249 * and the expansion ROM have a single line of data containing
250 * 3, 64-bit hex values: the first address in the region,
251 * the last address in the region, and the region's flags.
253 snprintf( name, 255, "%s/%04x:%02x:%02x.%1u/resource",
259 fd = open( name, O_RDONLY );
267 bytes = read( fd, resource, 512 );
268 resource[511] = '\0';
273 for ( i = 0 ; i < 6 ; i++ ) {
275 dev->regions[i].base_addr = strtoull( next, & next, 16 );
276 high_addr = strtoull( next, & next, 16 );
277 flags = strtoull( next, & next, 16 );
279 if ( dev->regions[i].base_addr != 0 ) {
280 dev->regions[i].size = (high_addr
281 - dev->regions[i].base_addr) + 1;
283 dev->regions[i].is_IO = (flags & 0x01);
284 dev->regions[i].is_64 = (flags & 0x04);
285 dev->regions[i].is_prefetchable = (flags & 0x08);
289 low_addr = strtoull( next, & next, 16 );
290 high_addr = strtoull( next, & next, 16 );
291 flags = strtoull( next, & next, 16 );
292 if ( low_addr != 0 ) {
293 priv->rom_base = low_addr;
294 dev->rom_size = (high_addr - low_addr) + 1;
304 pci_device_linux_sysfs_read_rom( struct pci_device * dev, void * buffer )
314 snprintf( name, 255, "%s/%04x:%02x:%02x.%1u/rom",
321 fd = open( name, O_RDWR );
323 /* If reading the ROM using sysfs fails, fall back to the old
324 * /dev/mem based interface.
326 return pci_device_linux_devmem_read_rom(dev, buffer);
330 if ( fstat( fd, & st ) == -1 ) {
335 rom_size = st.st_size;
339 /* This is a quirky thing on Linux. Even though the ROM and the file
340 * for the ROM in sysfs are read-only, the string "1" must be written to
341 * the file to enable the ROM. After the data has been read, "0" must be
342 * written to the file to disable the ROM.
345 lseek( fd, 0, SEEK_SET );
347 for ( total_bytes = 0 ; total_bytes < rom_size ; /* empty */ ) {
348 const int bytes = read( fd, (char *) buffer + total_bytes,
349 rom_size - total_bytes );
354 else if ( bytes == 0 ) {
358 total_bytes += bytes;
362 lseek( fd, 0, SEEK_SET );
371 pci_device_linux_sysfs_read( struct pci_device * dev, void * data,
372 pciaddr_t offset, pciaddr_t size,
373 pciaddr_t * bytes_read )
376 pciaddr_t temp_size = size;
379 char *data_bytes = data;
381 if ( bytes_read != NULL ) {
385 /* Each device has a directory under sysfs. Within that directory there
386 * is a file named "config". This file used to access the PCI config
387 * space. It is used here to obtain most of the information about the
390 snprintf( name, 255, "%s/%04x:%02x:%02x.%1u/config",
397 fd = open( name, O_RDONLY );
403 while ( temp_size > 0 ) {
404 const ssize_t bytes = pread64( fd, data_bytes, temp_size, offset );
406 /* If zero bytes were read, then we assume it's the end of the
419 if ( bytes_read != NULL ) {
420 *bytes_read = size - temp_size;
429 pci_device_linux_sysfs_write( struct pci_device * dev, const void * data,
430 pciaddr_t offset, pciaddr_t size,
431 pciaddr_t * bytes_written )
434 pciaddr_t temp_size = size;
437 const char *data_bytes = data;
439 if ( bytes_written != NULL ) {
443 /* Each device has a directory under sysfs. Within that directory there
444 * is a file named "config". This file used to access the PCI config
445 * space. It is used here to obtain most of the information about the
448 snprintf( name, 255, "%s/%04x:%02x:%02x.%1u/config",
455 fd = open( name, O_WRONLY );
461 while ( temp_size > 0 ) {
462 const ssize_t bytes = pwrite64( fd, data_bytes, temp_size, offset );
464 /* If zero bytes were written, then we assume it's the end of the
477 if ( bytes_written != NULL ) {
478 *bytes_written = size - temp_size;
487 * Map a memory region for a device using the Linux sysfs interface.
489 * \param dev Device whose memory region is to be mapped.
490 * \param map Parameters of the mapping that is to be created.
493 * Zero on success or an \c errno value on failure.
495 * \sa pci_device_map_rrange, pci_device_linux_sysfs_unmap_range
498 * Some older 2.6.x kernels don't implement the resourceN files. On those
499 * systems /dev/mem must be used. On these systems it is also possible that
500 * \c mmap64 may need to be used.
503 pci_device_linux_sysfs_map_range(struct pci_device *dev,
504 struct pci_device_mapping *map)
509 const int prot = ((map->flags & PCI_DEV_MAP_FLAG_WRITABLE) != 0)
510 ? (PROT_READ | PROT_WRITE) : PROT_READ;
511 const int open_flags = ((map->flags & PCI_DEV_MAP_FLAG_WRITABLE) != 0)
513 const off_t offset = map->base - dev->regions[map->region].base_addr;
515 struct mtrr_sentry sentry = {
518 .type = MTRR_TYPE_UNCACHABLE
522 snprintf(name, 255, "%s/%04x:%02x:%02x.%1u/resource%u",
530 fd = open(name, open_flags);
536 map->memory = mmap(NULL, map->size, prot, MAP_SHARED, fd, offset);
537 if (map->memory == MAP_FAILED) {
545 if ((map->flags & PCI_DEV_MAP_FLAG_CACHABLE) != 0) {
546 sentry.type = MTRR_TYPE_WRBACK;
547 } else if ((map->flags & PCI_DEV_MAP_FLAG_WRITE_COMBINE) != 0) {
548 sentry.type = MTRR_TYPE_WRCOMB;
551 if (pci_sys->mtrr_fd != -1 && sentry.type != MTRR_TYPE_UNCACHABLE) {
552 if (ioctl(pci_sys->mtrr_fd, MTRRIOC_ADD_ENTRY, &sentry) < 0) {
553 /* FIXME: Should we report an error in this case?
555 fprintf(stderr, "error setting MTRR "
556 "(base = 0x%08lx, size = 0x%08x, type = %u) %s (%d)\n",
557 sentry.base, sentry.size, sentry.type,
558 strerror(errno), errno);
568 * Unmap a memory region for a device using the Linux sysfs interface.
570 * \param dev Device whose memory region is to be unmapped.
571 * \param map Parameters of the mapping that is to be destroyed.
574 * Zero on success or an \c errno value on failure.
576 * \sa pci_device_map_rrange, pci_device_linux_sysfs_map_range
579 * Some older 2.6.x kernels don't implement the resourceN files. On those
580 * systems /dev/mem must be used. On these systems it is also possible that
581 * \c mmap64 may need to be used.
584 pci_device_linux_sysfs_unmap_range(struct pci_device *dev,
585 struct pci_device_mapping *map)
589 struct mtrr_sentry sentry = {
592 .type = MTRR_TYPE_UNCACHABLE
596 err = pci_device_generic_unmap_range (dev, map);
601 if ((map->flags & PCI_DEV_MAP_FLAG_CACHABLE) != 0) {
602 sentry.type = MTRR_TYPE_WRBACK;
603 } else if ((map->flags & PCI_DEV_MAP_FLAG_WRITE_COMBINE) != 0) {
604 sentry.type = MTRR_TYPE_WRCOMB;
607 if (pci_sys->mtrr_fd != -1 && sentry.type != MTRR_TYPE_UNCACHABLE) {
608 if (ioctl(pci_sys->mtrr_fd, MTRRIOC_DEL_ENTRY, &sentry) < 0) {
609 /* FIXME: Should we report an error in this case?
611 fprintf(stderr, "error setting MTRR "
612 "(base = 0x%08lx, size = 0x%08x, type = %u) %s (%d)\n",
613 sentry.base, sentry.size, sentry.type,
614 strerror(errno), errno);