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 * preferred 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 void pci_device_linux_sysfs_enable(struct pci_device *dev);
60 static int pci_device_linux_sysfs_read_rom( struct pci_device * dev,
63 static int pci_device_linux_sysfs_probe( struct pci_device * dev );
65 static int pci_device_linux_sysfs_map_range(struct pci_device *dev,
66 struct pci_device_mapping *map);
68 static int pci_device_linux_sysfs_unmap_range(struct pci_device *dev,
69 struct pci_device_mapping *map);
71 static int pci_device_linux_sysfs_read( struct pci_device * dev, void * data,
72 pciaddr_t offset, pciaddr_t size, pciaddr_t * bytes_read );
74 static int pci_device_linux_sysfs_write( struct pci_device * dev,
75 const void * data, pciaddr_t offset, pciaddr_t size,
76 pciaddr_t * bytes_written );
78 static int pci_device_linux_sysfs_boot_vga( struct pci_device * dev );
79 static int pci_device_linux_sysfs_has_kernel_driver(struct pci_device *dev);
81 static const struct pci_system_methods linux_sysfs_methods = {
83 .destroy_device = NULL,
84 .read_rom = pci_device_linux_sysfs_read_rom,
85 .probe = pci_device_linux_sysfs_probe,
86 .map_range = pci_device_linux_sysfs_map_range,
87 .unmap_range = pci_device_linux_sysfs_unmap_range,
89 .read = pci_device_linux_sysfs_read,
90 .write = pci_device_linux_sysfs_write,
92 .fill_capabilities = pci_fill_capabilities_generic,
93 .enable = pci_device_linux_sysfs_enable,
94 .boot_vga = pci_device_linux_sysfs_boot_vga,
95 .has_kernel_driver = pci_device_linux_sysfs_has_kernel_driver,
98 #define SYS_BUS_PCI "/sys/bus/pci/devices"
101 static int populate_entries(struct pci_system * pci_sys);
105 * Attempt to access PCI subsystem using Linux's sysfs interface.
108 pci_system_linux_sysfs_create( void )
114 /* If the directory "/sys/bus/pci/devices" exists, then the PCI subsystem
115 * can be accessed using this interface.
118 if ( stat( SYS_BUS_PCI, & st ) == 0 ) {
119 pci_sys = calloc( 1, sizeof( struct pci_system ) );
120 if ( pci_sys != NULL ) {
121 pci_sys->methods = & linux_sysfs_methods;
123 pci_sys->mtrr_fd = open("/proc/mtrr", O_WRONLY);
125 err = populate_entries(pci_sys);
140 * Filter out the names "." and ".." from the scanned sysfs entries.
142 * \param d Directory entry being processed by \c scandir.
145 * Zero if the entry name matches either "." or "..", non-zero otherwise.
147 * \sa scandir, populate_entries
150 scan_sys_pci_filter( const struct dirent * d )
152 return !((strcmp( d->d_name, "." ) == 0)
153 || (strcmp( d->d_name, ".." ) == 0));
158 populate_entries( struct pci_system * p )
160 struct dirent ** devices;
166 n = scandir( SYS_BUS_PCI, & devices, scan_sys_pci_filter, alphasort );
169 p->devices = calloc( n, sizeof( struct pci_device_private ) );
171 if (p->devices != NULL) {
172 for (i = 0 ; i < n ; i++) {
175 unsigned dom, bus, dev, func;
176 struct pci_device_private *device =
177 (struct pci_device_private *) &p->devices[i];
180 sscanf(devices[i]->d_name, "%04x:%02x:%02x.%1u",
181 & dom, & bus, & dev, & func);
183 device->base.domain = dom;
184 device->base.bus = bus;
185 device->base.dev = dev;
186 device->base.func = func;
189 err = pci_device_linux_sysfs_read(& device->base, config, 0,
191 if ((bytes == 48) && !err) {
192 device->base.vendor_id = (uint16_t)config[0]
193 + ((uint16_t)config[1] << 8);
194 device->base.device_id = (uint16_t)config[2]
195 + ((uint16_t)config[3] << 8);
196 device->base.device_class = (uint32_t)config[9]
197 + ((uint32_t)config[10] << 8)
198 + ((uint32_t)config[11] << 16);
199 device->base.revision = config[8];
200 device->base.subvendor_id = (uint16_t)config[44]
201 + ((uint16_t)config[45] << 8);
202 device->base.subdevice_id = (uint16_t)config[46]
203 + ((uint16_t)config[47] << 8);
226 pci_device_linux_sysfs_probe( struct pci_device * dev )
237 err = pci_device_linux_sysfs_read( dev, config, 0, 256, & bytes );
239 struct pci_device_private *priv = (struct pci_device_private *) dev;
241 dev->irq = config[60];
242 priv->header_type = config[14];
245 /* The PCI config registers can be used to obtain information
246 * about the memory and I/O regions for the device. However,
247 * doing so requires some tricky parsing (to correctly handle
248 * 64-bit memory regions) and requires writing to the config
249 * registers. Since we'd like to avoid having to deal with the
250 * parsing issues and non-root users can write to PCI config
251 * registers, we use a different file in the device's sysfs
252 * directory called "resource".
254 * The resource file contains all of the needed information in
255 * a format that is consistent across all platforms. Each BAR
256 * and the expansion ROM have a single line of data containing
257 * 3, 64-bit hex values: the first address in the region,
258 * the last address in the region, and the region's flags.
260 snprintf( name, 255, "%s/%04x:%02x:%02x.%1u/resource",
266 fd = open( name, O_RDONLY );
274 bytes = read( fd, resource, 512 );
275 resource[511] = '\0';
280 for ( i = 0 ; i < 6 ; i++ ) {
282 dev->regions[i].base_addr = strtoull( next, & next, 16 );
283 high_addr = strtoull( next, & next, 16 );
284 flags = strtoull( next, & next, 16 );
286 if ( dev->regions[i].base_addr != 0 ) {
287 dev->regions[i].size = (high_addr
288 - dev->regions[i].base_addr) + 1;
290 dev->regions[i].is_IO = (flags & 0x01);
291 dev->regions[i].is_64 = (flags & 0x04);
292 dev->regions[i].is_prefetchable = (flags & 0x08);
296 low_addr = strtoull( next, & next, 16 );
297 high_addr = strtoull( next, & next, 16 );
298 flags = strtoull( next, & next, 16 );
299 if ( low_addr != 0 ) {
300 priv->rom_base = low_addr;
301 dev->rom_size = (high_addr - low_addr) + 1;
311 pci_device_linux_sysfs_read_rom( struct pci_device * dev, void * buffer )
321 snprintf( name, 255, "%s/%04x:%02x:%02x.%1u/rom",
328 fd = open( name, O_RDWR );
331 /* If reading the ROM using sysfs fails, fall back to the old
332 * /dev/mem based interface.
333 * disable this for newer kernels using configure
335 return pci_device_linux_devmem_read_rom(dev, buffer);
342 if ( fstat( fd, & st ) == -1 ) {
347 rom_size = st.st_size;
351 /* This is a quirky thing on Linux. Even though the ROM and the file
352 * for the ROM in sysfs are read-only, the string "1" must be written to
353 * the file to enable the ROM. After the data has been read, "0" must be
354 * written to the file to disable the ROM.
357 lseek( fd, 0, SEEK_SET );
359 for ( total_bytes = 0 ; total_bytes < rom_size ; /* empty */ ) {
360 const int bytes = read( fd, (char *) buffer + total_bytes,
361 rom_size - total_bytes );
366 else if ( bytes == 0 ) {
370 total_bytes += bytes;
374 lseek( fd, 0, SEEK_SET );
383 pci_device_linux_sysfs_read( struct pci_device * dev, void * data,
384 pciaddr_t offset, pciaddr_t size,
385 pciaddr_t * bytes_read )
388 pciaddr_t temp_size = size;
391 char *data_bytes = data;
393 if ( bytes_read != NULL ) {
397 /* Each device has a directory under sysfs. Within that directory there
398 * is a file named "config". This file used to access the PCI config
399 * space. It is used here to obtain most of the information about the
402 snprintf( name, 255, "%s/%04x:%02x:%02x.%1u/config",
409 fd = open( name, O_RDONLY );
415 while ( temp_size > 0 ) {
416 const ssize_t bytes = pread64( fd, data_bytes, temp_size, offset );
418 /* If zero bytes were read, then we assume it's the end of the
431 if ( bytes_read != NULL ) {
432 *bytes_read = size - temp_size;
441 pci_device_linux_sysfs_write( struct pci_device * dev, const void * data,
442 pciaddr_t offset, pciaddr_t size,
443 pciaddr_t * bytes_written )
446 pciaddr_t temp_size = size;
449 const char *data_bytes = data;
451 if ( bytes_written != NULL ) {
455 /* Each device has a directory under sysfs. Within that directory there
456 * is a file named "config". This file used to access the PCI config
457 * space. It is used here to obtain most of the information about the
460 snprintf( name, 255, "%s/%04x:%02x:%02x.%1u/config",
467 fd = open( name, O_WRONLY );
473 while ( temp_size > 0 ) {
474 const ssize_t bytes = pwrite64( fd, data_bytes, temp_size, offset );
476 /* If zero bytes were written, then we assume it's the end of the
489 if ( bytes_written != NULL ) {
490 *bytes_written = size - temp_size;
498 pci_device_linux_sysfs_map_range_wc(struct pci_device *dev,
499 struct pci_device_mapping *map)
503 const int prot = ((map->flags & PCI_DEV_MAP_FLAG_WRITABLE) != 0)
504 ? (PROT_READ | PROT_WRITE) : PROT_READ;
505 const int open_flags = ((map->flags & PCI_DEV_MAP_FLAG_WRITABLE) != 0)
507 const off_t offset = map->base - dev->regions[map->region].base_addr;
509 snprintf(name, 255, "%s/%04x:%02x:%02x.%1u/resource%u_wc",
516 fd = open(name, open_flags);
520 map->memory = mmap(NULL, map->size, prot, MAP_SHARED, fd, offset);
521 if (map->memory == MAP_FAILED) {
533 * Map a memory region for a device using the Linux sysfs interface.
535 * \param dev Device whose memory region is to be mapped.
536 * \param map Parameters of the mapping that is to be created.
539 * Zero on success or an \c errno value on failure.
541 * \sa pci_device_map_rrange, pci_device_linux_sysfs_unmap_range
544 * Some older 2.6.x kernels don't implement the resourceN files. On those
545 * systems /dev/mem must be used. On these systems it is also possible that
546 * \c mmap64 may need to be used.
549 pci_device_linux_sysfs_map_range(struct pci_device *dev,
550 struct pci_device_mapping *map)
555 const int prot = ((map->flags & PCI_DEV_MAP_FLAG_WRITABLE) != 0)
556 ? (PROT_READ | PROT_WRITE) : PROT_READ;
557 const int open_flags = ((map->flags & PCI_DEV_MAP_FLAG_WRITABLE) != 0)
559 const off_t offset = map->base - dev->regions[map->region].base_addr;
561 struct mtrr_sentry sentry = {
564 .type = MTRR_TYPE_UNCACHABLE
568 /* For WC mappings, try sysfs resourceN_wc file first */
569 if ((map->flags & PCI_DEV_MAP_FLAG_WRITE_COMBINE) &&
570 !pci_device_linux_sysfs_map_range_wc(dev, map))
573 snprintf(name, 255, "%s/%04x:%02x:%02x.%1u/resource%u",
581 fd = open(name, open_flags);
587 map->memory = mmap(NULL, map->size, prot, MAP_SHARED, fd, offset);
588 if (map->memory == MAP_FAILED) {
595 if ((map->flags & PCI_DEV_MAP_FLAG_CACHABLE) != 0) {
596 sentry.type = MTRR_TYPE_WRBACK;
597 } else if ((map->flags & PCI_DEV_MAP_FLAG_WRITE_COMBINE) != 0) {
598 sentry.type = MTRR_TYPE_WRCOMB;
601 if (pci_sys->mtrr_fd != -1 && sentry.type != MTRR_TYPE_UNCACHABLE) {
602 if (ioctl(pci_sys->mtrr_fd, MTRRIOC_ADD_ENTRY, &sentry) < 0) {
603 /* FIXME: Should we report an error in this case?
605 fprintf(stderr, "error setting MTRR "
606 "(base = 0x%08lx, size = 0x%08x, type = %u) %s (%d)\n",
607 sentry.base, sentry.size, sentry.type,
608 strerror(errno), errno);
611 /* KLUDGE ALERT -- rewrite the PTEs to turn off the CD and WT bits */
612 mprotect (map->memory, map->size, PROT_NONE);
613 err = mprotect (map->memory, map->size, PROT_READ|PROT_WRITE);
616 fprintf(stderr, "mprotect(PROT_READ | PROT_WRITE) failed: %s\n",
618 fprintf(stderr, "remapping without mprotect performance kludge.\n");
620 munmap(map->memory, map->size);
621 map->memory = mmap(NULL, map->size, prot, MAP_SHARED, fd, offset);
622 if (map->memory == MAP_FAILED) {
637 * Unmap a memory region for a device using the Linux sysfs interface.
639 * \param dev Device whose memory region is to be unmapped.
640 * \param map Parameters of the mapping that is to be destroyed.
643 * Zero on success or an \c errno value on failure.
645 * \sa pci_device_map_rrange, pci_device_linux_sysfs_map_range
648 * Some older 2.6.x kernels don't implement the resourceN files. On those
649 * systems /dev/mem must be used. On these systems it is also possible that
650 * \c mmap64 may need to be used.
653 pci_device_linux_sysfs_unmap_range(struct pci_device *dev,
654 struct pci_device_mapping *map)
658 struct mtrr_sentry sentry = {
661 .type = MTRR_TYPE_UNCACHABLE
665 err = pci_device_generic_unmap_range (dev, map);
670 if ((map->flags & PCI_DEV_MAP_FLAG_CACHABLE) != 0) {
671 sentry.type = MTRR_TYPE_WRBACK;
672 } else if ((map->flags & PCI_DEV_MAP_FLAG_WRITE_COMBINE) != 0) {
673 sentry.type = MTRR_TYPE_WRCOMB;
676 if (pci_sys->mtrr_fd != -1 && sentry.type != MTRR_TYPE_UNCACHABLE) {
677 if (ioctl(pci_sys->mtrr_fd, MTRRIOC_DEL_ENTRY, &sentry) < 0) {
678 /* FIXME: Should we report an error in this case?
680 fprintf(stderr, "error setting MTRR "
681 "(base = 0x%08lx, size = 0x%08x, type = %u) %s (%d)\n",
682 sentry.base, sentry.size, sentry.type,
683 strerror(errno), errno);
692 static void pci_device_linux_sysfs_enable(struct pci_device *dev)
697 snprintf( name, 255, "%s/%04x:%02x:%02x.%1u/enable",
704 fd = open( name, O_RDWR );
712 static int pci_device_linux_sysfs_boot_vga(struct pci_device *dev)
719 snprintf( name, 255, "%s/%04x:%02x:%02x.%1u/boot_vga",
726 fd = open( name, O_RDONLY );
730 bytes_read = read(fd, reply, 1);
740 static int pci_device_linux_sysfs_has_kernel_driver(struct pci_device *dev)
746 snprintf( name, 255, "%s/%04x:%02x:%02x.%1u/driver",
753 ret = stat(name, &dummy);