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>
47 #include "pciaccess.h"
48 #include "pciaccess_private.h"
49 #include "linux_devmem.h"
51 static int pci_device_linux_sysfs_read_rom( struct pci_device * dev,
54 static int pci_device_linux_sysfs_probe( struct pci_device * dev );
56 static int pci_device_linux_sysfs_map_region( struct pci_device * dev,
57 unsigned region, int write_enable );
59 static int pci_device_linux_sysfs_unmap_region( struct pci_device * dev,
62 static int pci_device_linux_sysfs_read( struct pci_device * dev, void * data,
63 pciaddr_t offset, pciaddr_t size, pciaddr_t * bytes_read );
65 static int pci_device_linux_sysfs_write( struct pci_device * dev,
66 const void * data, pciaddr_t offset, pciaddr_t size,
67 pciaddr_t * bytes_wrtten );
69 static const struct pci_system_methods linux_sysfs_methods = {
71 .destroy_device = NULL,
72 .read_rom = pci_device_linux_sysfs_read_rom,
73 .probe = pci_device_linux_sysfs_probe,
74 .map = pci_device_linux_sysfs_map_region,
75 .unmap = pci_device_linux_sysfs_unmap_region,
77 .read = pci_device_linux_sysfs_read,
78 .write = pci_device_linux_sysfs_write,
80 .fill_capabilities = pci_fill_capabilities_generic
83 #define SYS_BUS_PCI "/sys/bus/pci/devices"
86 static int populate_entries(struct pci_system * pci_sys);
90 * Attempt to access PCI subsystem using Linux's sysfs interface.
93 pci_system_linux_sysfs_create( void )
99 /* If the directory "/sys/bus/pci/devices" exists, then the PCI subsystem
100 * can be accessed using this interface.
103 if ( stat( SYS_BUS_PCI, & st ) == 0 ) {
104 pci_sys = calloc( 1, sizeof( struct pci_system ) );
105 if ( pci_sys != NULL ) {
106 pci_sys->methods = & linux_sysfs_methods;
107 err = populate_entries(pci_sys);
122 * Filter out the names "." and ".." from the scanned sysfs entries.
124 * \param d Directory entry being processed by \c scandir.
127 * Zero if the entry name matches either "." or "..", non-zero otherwise.
129 * \sa scandir, populate_entries
132 scan_sys_pci_filter( const struct dirent * d )
134 return !((strcmp( d->d_name, "." ) == 0)
135 || (strcmp( d->d_name, ".." ) == 0));
140 populate_entries( struct pci_system * p )
142 struct dirent ** devices;
148 n = scandir( SYS_BUS_PCI, & devices, scan_sys_pci_filter, alphasort );
151 p->devices = calloc( n, sizeof( struct pci_device_private ) );
153 if (p->devices != NULL) {
154 for (i = 0 ; i < n ; i++) {
157 unsigned dom, bus, dev, func;
158 struct pci_device_private *device =
159 (struct pci_device_private *) &p->devices[i];
162 sscanf(devices[i]->d_name, "%04x:%02x:%02x.%1u",
163 & dom, & bus, & dev, & func);
165 device->base.domain = dom;
166 device->base.bus = bus;
167 device->base.dev = dev;
168 device->base.func = func;
171 err = pci_device_linux_sysfs_read(& device->base, config, 0,
173 if ((bytes == 48) && !err) {
174 device->base.vendor_id = (uint16_t)config[0]
175 + ((uint16_t)config[1] << 8);
176 device->base.device_id = (uint16_t)config[2]
177 + ((uint16_t)config[3] << 8);
178 device->base.device_class = (uint32_t)config[9]
179 + ((uint32_t)config[10] << 8)
180 + ((uint32_t)config[11] << 16);
181 device->base.revision = config[8];
182 device->base.subvendor_id = (uint16_t)config[44]
183 + ((uint16_t)config[45] << 8);
184 device->base.subdevice_id = (uint16_t)config[46]
185 + ((uint16_t)config[47] << 8);
208 pci_device_linux_sysfs_probe( struct pci_device * dev )
219 err = pci_device_linux_sysfs_read( dev, config, 0, 256, & bytes );
221 struct pci_device_private *priv = (struct pci_device_private *) dev;
223 dev->irq = config[60];
224 priv->header_type = config[14];
227 /* The PCI config registers can be used to obtain information
228 * about the memory and I/O regions for the device. However,
229 * doing so requires some tricky parsing (to correctly handle
230 * 64-bit memory regions) and requires writing to the config
231 * registers. Since we'd like to avoid having to deal with the
232 * parsing issues and non-root users can write to PCI config
233 * registers, we use a different file in the device's sysfs
234 * directory called "resource".
236 * The resource file contains all of the needed information in
237 * a format that is consistent across all platforms. Each BAR
238 * and the expansion ROM have a single line of data containing
239 * 3, 64-bit hex values: the first address in the region,
240 * the last address in the region, and the region's flags.
242 snprintf( name, 255, "%s/%04x:%02x:%02x.%1u/resource",
248 fd = open( name, O_RDONLY );
256 bytes = read( fd, resource, 512 );
257 resource[511] = '\0';
262 for ( i = 0 ; i < 6 ; i++ ) {
264 dev->regions[i].base_addr = strtoull( next, & next, 16 );
265 high_addr = strtoull( next, & next, 16 );
266 flags = strtoull( next, & next, 16 );
268 if ( dev->regions[i].base_addr != 0 ) {
269 dev->regions[i].size = (high_addr
270 - dev->regions[i].base_addr) + 1;
272 dev->regions[i].is_IO = (flags & 0x01);
273 dev->regions[i].is_64 = (flags & 0x04);
274 dev->regions[i].is_prefetchable = (flags & 0x08);
278 low_addr = strtoull( next, & next, 16 );
279 high_addr = strtoull( next, & next, 16 );
280 flags = strtoull( next, & next, 16 );
281 if ( low_addr != 0 ) {
282 priv->rom_base = low_addr;
283 dev->rom_size = (high_addr - low_addr) + 1;
293 pci_device_linux_sysfs_read_rom( struct pci_device * dev, void * buffer )
302 snprintf( name, 255, "%s/%04x:%02x:%02x.%1u/rom",
309 fd = open( name, O_RDWR );
311 /* If reading the ROM using sysfs fails, fall back to the old
312 * /dev/mem based interface.
314 return pci_device_linux_devmem_read_rom(dev, buffer);
318 if ( fstat( fd, & st ) == -1 ) {
324 /* This is a quirky thing on Linux. Even though the ROM and the file
325 * for the ROM in sysfs are read-only, the string "1" must be written to
326 * the file to enable the ROM. After the data has been read, "0" must be
327 * written to the file to disable the ROM.
330 lseek( fd, 0, SEEK_SET );
332 for ( total_bytes = 0 ; total_bytes < st.st_size ; /* empty */ ) {
333 const int bytes = read( fd, (char *) buffer + total_bytes,
334 st.st_size - total_bytes );
339 else if ( bytes == 0 ) {
343 total_bytes += bytes;
347 lseek( fd, 0, SEEK_SET );
356 pci_device_linux_sysfs_read( struct pci_device * dev, void * data,
357 pciaddr_t offset, pciaddr_t size,
358 pciaddr_t * bytes_read )
361 pciaddr_t temp_size = size;
366 if ( bytes_read != NULL ) {
370 /* Each device has a directory under sysfs. Within that directory there
371 * is a file named "config". This file used to access the PCI config
372 * space. It is used here to obtain most of the information about the
375 snprintf( name, 255, "%s/%04x:%02x:%02x.%1u/config",
382 fd = open( name, O_RDONLY );
388 while ( temp_size > 0 ) {
389 const ssize_t bytes = pread64( fd, data, temp_size, offset );
391 /* If zero bytes were read, then we assume it's the end of the
404 if ( bytes_read != NULL ) {
405 *bytes_read = size - temp_size;
414 pci_device_linux_sysfs_write( struct pci_device * dev, const void * data,
415 pciaddr_t offset, pciaddr_t size,
416 pciaddr_t * bytes_written )
419 pciaddr_t temp_size = size;
424 if ( bytes_written != NULL ) {
428 /* Each device has a directory under sysfs. Within that directory there
429 * is a file named "config". This file used to access the PCI config
430 * space. It is used here to obtain most of the information about the
433 snprintf( name, 255, "%s/%04x:%02x:%02x.%1u/config",
440 fd = open( name, O_WRONLY );
446 while ( temp_size > 0 ) {
447 const ssize_t bytes = pwrite64( fd, data, temp_size, offset );
449 /* If zero bytes were written, then we assume it's the end of the
462 if ( bytes_written != NULL ) {
463 *bytes_written = size - temp_size;
472 * Map a memory region for a device using the Linux sysfs interface.
474 * \param dev Device whose memory region is to be mapped.
475 * \param region Region, on the range [0, 5], that is to be mapped.
476 * \param write_enable Map for writing (non-zero).
479 * Zero on success or an \c errno value on failure.
481 * \sa pci_device_map_region, pci_device_linux_sysfs_unmap_region
484 * Some older 2.6.x kernels don't implement the resourceN files. On those
485 * systems /dev/mem must be used. On these systems it is also possible that
486 * \c mmap64 may need to be used.
489 pci_device_linux_sysfs_map_region( struct pci_device * dev, unsigned region,
495 const int prot = (write_enable) ? (PROT_READ | PROT_WRITE) : PROT_READ;
498 snprintf( name, 255, "%s/%04x:%02x:%02x.%1u/resource%u",
506 fd = open( name, (write_enable) ? O_RDWR : O_RDONLY );
512 dev->regions[ region ].memory = mmap( NULL, dev->regions[ region ].size,
513 prot, MAP_SHARED, fd, 0 );
514 if ( dev->regions[ region ].memory == MAP_FAILED ) {
516 dev->regions[ region ].memory = NULL;
525 * Unmap the specified region using the Linux sysfs interface.
527 * \param dev Device whose memory region is to be mapped.
528 * \param region Region, on the range [0, 5], that is to be mapped.
531 * Zero on success or an \c errno value on failure.
533 * \sa pci_device_unmap_region, pci_device_linux_sysfs_map_region
536 * Some older 2.6.x kernels don't implement the resourceN files. On those
537 * systems /dev/mem must be used. On these systems it is also possible that
538 * \c mmap64 may need to be used.
541 pci_device_linux_sysfs_unmap_region( struct pci_device * dev, unsigned region )
545 if ( munmap( dev->regions[ region ].memory, dev->regions[ region ].size )
550 dev->regions[ region ].memory = NULL;