* The PCI Library -- Configuration Access via /sys/bus/pci
*
* Copyright (c) 2003 Matthew Wilcox <matthew@wil.cx>
- * Copyright (c) 1997--2008 Martin Mares <mj@ucw.cz>
+ * Copyright (c) 1997--2023 Martin Mares <mj@ucw.cz>
*
- * Can be freely distributed and used under the terms of the GNU GPL.
+ * Can be freely distributed and used under the terms of the GNU GPL v2+.
+ *
+ * SPDX-License-Identifier: GPL-2.0-or-later
*/
#define _GNU_SOURCE
#include <sys/types.h>
#include "internal.h"
-#include "pread.h"
static void
sysfs_config(struct pci_access *a)
sysfs_init(struct pci_access *a)
{
a->fd = -1;
+ a->fd_vpd = -1;
}
static void
-sysfs_cleanup(struct pci_access *a)
+sysfs_flush_cache(struct pci_access *a)
{
if (a->fd >= 0)
{
close(a->fd);
a->fd = -1;
}
+ if (a->fd_vpd >= 0)
+ {
+ close(a->fd_vpd);
+ a->fd_vpd = -1;
+ }
+ a->cached_dev = NULL;
+}
+
+static void
+sysfs_cleanup(struct pci_access *a)
+{
+ sysfs_flush_cache(a);
}
#define OBJNAMELEN 1024
d->access->error("File name too long");
}
+#define OBJBUFSIZE 1024
+
static int
-sysfs_get_value(struct pci_dev *d, char *object)
+sysfs_get_string(struct pci_dev *d, char *object, char *buf, int mandatory)
{
struct pci_access *a = d->access;
int fd, n;
- char namebuf[OBJNAMELEN], buf[256];
+ char namebuf[OBJNAMELEN];
+ void (*warn)(char *msg, ...) = (mandatory ? a->error : a->warning);
sysfs_obj_name(d, object, namebuf);
fd = open(namebuf, O_RDONLY);
if (fd < 0)
- a->error("Cannot open %s: %s", namebuf, strerror(errno));
- n = read(fd, buf, sizeof(buf));
+ {
+ if (mandatory || errno != ENOENT)
+ warn("Cannot open %s: %s", namebuf, strerror(errno));
+ return 0;
+ }
+ n = read(fd, buf, OBJBUFSIZE);
+ int read_errno = errno;
close(fd);
if (n < 0)
- a->error("Error reading %s: %s", namebuf, strerror(errno));
- if (n >= (int) sizeof(buf))
- a->error("Value in %s too long", namebuf);
+ {
+ warn("Error reading %s: %s", namebuf, strerror(read_errno));
+ return 0;
+ }
+ if (n >= OBJBUFSIZE)
+ {
+ warn("Value in %s too long", namebuf);
+ return 0;
+ }
buf[n] = 0;
- return strtol(buf, NULL, 0);
+ return 1;
+}
+
+static char *
+sysfs_deref_link(struct pci_dev *d, char *link_name)
+{
+ char path[2*OBJNAMELEN], rel_path[OBJNAMELEN];
+
+ sysfs_obj_name(d, link_name, path);
+ memset(rel_path, 0, sizeof(rel_path));
+
+ if (readlink(path, rel_path, sizeof(rel_path)) < 0)
+ return NULL;
+
+ sysfs_obj_name(d, "", path);
+ strcat(path, rel_path);
+
+ // Returns a pointer to malloc'ed memory
+ return realpath(path, NULL);
+}
+
+static int
+sysfs_get_value(struct pci_dev *d, char *object, int mandatory)
+{
+ char buf[OBJBUFSIZE];
+
+ if (sysfs_get_string(d, object, buf, mandatory))
+ return strtol(buf, NULL, 0);
+ else
+ return -1;
}
static void
{
struct pci_access *a = d->access;
char namebuf[OBJNAMELEN], buf[256];
+ struct { pciaddr_t flags, base_addr, size; } lines[10];
+ int have_bar_bases, have_rom_base, have_bridge_bases;
FILE *file;
int i;
+ have_bar_bases = have_rom_base = have_bridge_bases = 0;
sysfs_obj_name(d, "resource", namebuf);
file = fopen(namebuf, "r");
if (!file)
a->error("Cannot open %s: %s", namebuf, strerror(errno));
- for (i = 0; i < 7; i++)
+ for (i = 0; i < 7+6+4+1; i++)
{
- unsigned long long start, end, size;
+ unsigned long long start, end, size, flags;
if (!fgets(buf, sizeof(buf), file))
break;
- if (sscanf(buf, "%llx %llx", &start, &end) != 2)
+ if (sscanf(buf, "%llx %llx %llx", &start, &end, &flags) != 3)
a->error("Syntax error in %s", namebuf);
- if (start)
+ if (end > start)
size = end - start + 1;
else
size = 0;
if (i < 6)
{
- d->base_addr[i] = start;
+ d->flags[i] = flags;
+ flags &= PCI_ADDR_FLAG_MASK;
+ d->base_addr[i] = start | flags;
d->size[i] = size;
+ have_bar_bases = 1;
}
- else
+ else if (i == 6)
{
- d->rom_base_addr = start;
+ d->rom_flags = flags;
+ flags &= PCI_ADDR_FLAG_MASK;
+ d->rom_base_addr = start | flags;
d->rom_size = size;
+ have_rom_base = 1;
}
+ else if (i < 7+6+4)
+ {
+ /*
+ * If kernel was compiled without CONFIG_PCI_IOV option then after
+ * the ROM line for configured bridge device (that which had set
+ * subordinary bus number to non-zero value) are four additional lines
+ * which describe resources behind bridge. For PCI-to-PCI bridges they
+ * are: IO, MEM, PREFMEM and empty. For CardBus bridges they are: IO0,
+ * IO1, MEM0 and MEM1. For unconfigured bridges and other devices
+ * there is no additional line after the ROM line. If kernel was
+ * compiled with CONFIG_PCI_IOV option then after the ROM line and
+ * before the first bridge resource line are six additional lines
+ * which describe IOV resources. Read all remaining lines in resource
+ * file and based on the number of remaining lines (0, 4, 6, 10) parse
+ * resources behind bridge.
+ */
+ lines[i-7].flags = flags;
+ lines[i-7].base_addr = start;
+ lines[i-7].size = size;
+ }
+ }
+ if (i == 7+4 || i == 7+6+4)
+ {
+ int offset = (i == 7+6+4) ? 6 : 0;
+ for (i = 0; i < 4; i++)
+ {
+ d->bridge_flags[i] = lines[offset+i].flags;
+ d->bridge_base_addr[i] = lines[offset+i].base_addr;
+ d->bridge_size[i] = lines[offset+i].size;
+ }
+ have_bridge_bases = 1;
}
fclose(file);
+ if (!have_bar_bases)
+ clear_fill(d, PCI_FILL_BASES | PCI_FILL_SIZES | PCI_FILL_IO_FLAGS);
+ if (!have_rom_base)
+ clear_fill(d, PCI_FILL_ROM_BASE);
+ if (!have_bridge_bases)
+ clear_fill(d, PCI_FILL_BRIDGE_BASES);
}
static void sysfs_scan(struct pci_access *a)
d = pci_alloc_dev(a);
if (sscanf(entry->d_name, "%x:%x:%x.%d", &dom, &bus, &dev, &func) < 4)
a->error("sysfs_scan: Couldn't parse entry name %s", entry->d_name);
+
+ /* Ensure kernel provided domain that fits in a signed integer */
+ if (dom > 0x7fffffff)
+ a->error("sysfs_scan: Invalid domain %x", dom);
+
d->domain = dom;
d->bus = bus;
d->dev = dev;
d->func = func;
- if (!a->buscentric)
- {
- sysfs_get_resources(d);
- d->irq = sysfs_get_value(d, "irq");
- /*
- * We could read these faster from the config registers, but we want to give
- * the kernel a chance to fix up ID's and especially classes of broken devices.
- */
- d->vendor_id = sysfs_get_value(d, "vendor");
- d->device_id = sysfs_get_value(d, "device");
- d->device_class = sysfs_get_value(d, "class") >> 8;
- d->known_fields = PCI_FILL_IDENT | PCI_FILL_CLASS | PCI_FILL_IRQ | PCI_FILL_BASES | PCI_FILL_ROM_BASE | PCI_FILL_SIZES;
- }
pci_link_dev(a, d);
}
closedir(dir);
}
static void
-sysfs_fill_slots(struct pci_dev *d)
+sysfs_fill_slots(struct pci_access *a)
{
- struct pci_access *a = d->access;
char dirname[1024];
DIR *dir;
struct dirent *entry;
a->error("Directory name too long");
dir = opendir(dirname);
if (!dir)
- a->error("Cannot open %s", dirname);
- while ((entry = readdir(dir)))
+ return;
+
+ while (entry = readdir(dir))
{
char namebuf[OBJNAMELEN], buf[16];
FILE *file;
unsigned int dom, bus, dev;
- struct pci_dev *pd;
- int n = snprintf(namebuf, OBJNAMELEN, "%s/%s/%s", dirname, entry->d_name, "address");
+ int res = 0;
+ struct pci_dev *d;
/* ".", ".." or a special non-device perhaps */
if (entry->d_name[0] == '.')
continue;
+ n = snprintf(namebuf, OBJNAMELEN, "%s/%s/%s", dirname, entry->d_name, "address");
if (n < 0 || n >= OBJNAMELEN)
- d->access->error("File name too long");
+ a->error("File name too long");
file = fopen(namebuf, "r");
+ /*
+ * Old versions of Linux had a fakephp which didn't have an 'address'
+ * file. There's no useful information to be gleaned from these
+ * devices, pretend they're not there.
+ */
if (!file)
- a->error("Cannot open %s: %s", namebuf, strerror(errno));
- if (!fgets(buf, sizeof(buf), file))
- break;
- if (sscanf(buf, "%x:%x:%x", &dom, &bus, &dev) < 3)
- a->error("sysfs_scan: Couldn't parse entry address %s", buf);
- for (pd = a->devices; pd; pd = pd->next)
+ continue;
+
+ if (!fgets(buf, sizeof(buf), file) || (res = sscanf(buf, "%x:%x:%x", &dom, &bus, &dev)) < 3)
{
- if (dom == pd->domain && bus == pd->bus && dev == pd->dev && !pd->phy_slot)
- {
- pd->phy_slot = pci_malloc(a, strlen(entry->d_name) + 1);
- strcpy(pd->phy_slot, entry->d_name);
- }
- pd->known_fields |= PCI_FILL_PHYS_SLOT;
+ /*
+ * In some cases, the slot is not tied to a specific device before
+ * a card gets inserted. This happens for example on IBM pSeries
+ * and we need not warn about it.
+ */
+ if (res != 2)
+ a->warning("sysfs_fill_slots: Couldn't parse entry address %s", buf);
+ }
+ else
+ {
+ for (d = a->devices; d; d = d->next)
+ if (dom == (unsigned)d->domain && bus == d->bus && dev == d->dev && !d->phy_slot)
+ d->phy_slot = pci_set_property(d, PCI_FILL_PHYS_SLOT, entry->d_name);
}
fclose(file);
}
closedir(dir);
}
-static int
-sysfs_fill_info(struct pci_dev *d, int flags)
+static void
+sysfs_fill_info(struct pci_dev *d, unsigned int flags)
{
- int ret;
+ int value, want_class, want_class_ext;
- ret = pci_generic_fill_info(d, flags);
- if (flags & PCI_FILL_PHYS_SLOT && !(d->known_fields & PCI_FILL_PHYS_SLOT))
- sysfs_fill_slots(d);
+ if (!d->access->buscentric)
+ {
+ /*
+ * These fields can be read from the config registers, but we want to show
+ * the kernel's view, which has regions and IRQs remapped and other fields
+ * (most importantly classes) possibly fixed if the device is known broken.
+ */
+ if (want_fill(d, flags, PCI_FILL_IDENT))
+ {
+ d->vendor_id = sysfs_get_value(d, "vendor", 1);
+ d->device_id = sysfs_get_value(d, "device", 1);
+ }
+ want_class = want_fill(d, flags, PCI_FILL_CLASS);
+ want_class_ext = want_fill(d, flags, PCI_FILL_CLASS_EXT);
+ if (want_class || want_class_ext)
+ {
+ value = sysfs_get_value(d, "class", 1);
+ if (want_class)
+ d->device_class = value >> 8;
+ if (want_class_ext)
+ {
+ d->prog_if = value & 0xff;
+ value = sysfs_get_value(d, "revision", 0);
+ if (value < 0)
+ value = pci_read_byte(d, PCI_REVISION_ID);
+ if (value >= 0)
+ d->rev_id = value;
+ }
+ }
+ if (want_fill(d, flags, PCI_FILL_SUBSYS))
+ {
+ value = sysfs_get_value(d, "subsystem_vendor", 0);
+ if (value >= 0)
+ {
+ d->subsys_vendor_id = value;
+ value = sysfs_get_value(d, "subsystem_device", 0);
+ if (value >= 0)
+ d->subsys_id = value;
+ }
+ else
+ clear_fill(d, PCI_FILL_SUBSYS);
+ }
+ if (want_fill(d, flags, PCI_FILL_IRQ))
+ d->irq = sysfs_get_value(d, "irq", 1);
+ if (want_fill(d, flags, PCI_FILL_BASES | PCI_FILL_ROM_BASE | PCI_FILL_SIZES | PCI_FILL_IO_FLAGS | PCI_FILL_BRIDGE_BASES))
+ sysfs_get_resources(d);
+ if (want_fill(d, flags, PCI_FILL_PARENT))
+ {
+ unsigned int domain, bus, dev, func;
+ char *path_abs, *path_canon, *name;
+ char path_rel[OBJNAMELEN];
+ struct pci_dev *parent;
+
+ /* Construct sysfs path for parent device */
+ sysfs_obj_name(d, "..", path_rel);
+ path_abs = realpath(path_rel, NULL);
+ name = path_abs ? strrchr(path_abs, '/') : NULL;
+ name = name ? name+1 : name;
+ parent = NULL;
+
+ if (name && sscanf(name, "%x:%x:%x.%d", &domain, &bus, &dev, &func) == 4 && domain <= 0x7fffffff)
+ for (parent = d->access->devices; parent; parent = parent->next)
+ if (parent->domain == (int)domain && parent->bus == bus && parent->dev == dev && parent->func == func)
+ break;
+
+ if (parent)
+ {
+ /* Check if parsed BDF address from parent sysfs device is really expected PCI device */
+ sysfs_obj_name(parent, ".", path_rel);
+ path_canon = realpath(path_rel, NULL);
+ if (!path_canon || strcmp(path_canon, path_abs) != 0)
+ parent = NULL;
+
+ if (path_canon)
+ free(path_canon);
+ }
+
+ if (parent)
+ d->parent = parent;
+ else
+ clear_fill(d, PCI_FILL_PARENT);
+
+ if (path_abs)
+ free(path_abs);
+ }
+ }
+
+ if (want_fill(d, flags, PCI_FILL_PHYS_SLOT))
+ {
+ struct pci_dev *pd;
+ sysfs_fill_slots(d->access);
+ for (pd = d->access->devices; pd; pd = pd->next)
+ pd->known_fields |= PCI_FILL_PHYS_SLOT;
+ }
+
+ if (want_fill(d, flags, PCI_FILL_MODULE_ALIAS))
+ {
+ char buf[OBJBUFSIZE];
+ if (sysfs_get_string(d, "modalias", buf, 0))
+ d->module_alias = pci_set_property(d, PCI_FILL_MODULE_ALIAS, buf);
+ }
+
+ if (want_fill(d, flags, PCI_FILL_LABEL))
+ {
+ char buf[OBJBUFSIZE];
+ if (sysfs_get_string(d, "label", buf, 0))
+ d->label = pci_set_property(d, PCI_FILL_LABEL, buf);
+ }
+
+ if (want_fill(d, flags, PCI_FILL_NUMA_NODE))
+ d->numa_node = sysfs_get_value(d, "numa_node", 0);
+
+ if (want_fill(d, flags, PCI_FILL_IOMMU_GROUP))
+ {
+ char *group_link = sysfs_deref_link(d, "iommu_group");
+ if (group_link)
+ {
+ pci_set_property(d, PCI_FILL_IOMMU_GROUP, basename(group_link));
+ free(group_link);
+ }
+ }
+
+ if (want_fill(d, flags, PCI_FILL_DT_NODE))
+ {
+ char *node = sysfs_deref_link(d, "of_node");
+ if (node)
+ {
+ pci_set_property(d, PCI_FILL_DT_NODE, node);
+ free(node);
+ }
+ }
+
+ if (want_fill(d, flags, PCI_FILL_DRIVER))
+ {
+ char *driver_path = sysfs_deref_link(d, "driver");
+ if (driver_path)
+ {
+ char *driver = strrchr(driver_path, '/');
+ driver = driver ? driver+1 : driver_path;
+ pci_set_property(d, PCI_FILL_DRIVER, driver);
+ free(driver_path);
+ }
+ else
+ clear_fill(d, PCI_FILL_DRIVER);
+ }
- return ret;
+ pci_generic_fill_info(d, flags);
}
/* Intent of the sysfs_setup() caller */
struct pci_access *a = d->access;
char namebuf[OBJNAMELEN];
- if (a->cached_dev != d || intent == SETUP_WRITE_CONFIG && !a->fd_rw)
+ if (a->cached_dev != d || (intent == SETUP_WRITE_CONFIG && !a->fd_rw))
+ {
+ sysfs_flush_cache(a);
+ a->cached_dev = d;
+ }
+
+ if (intent == SETUP_READ_VPD)
+ {
+ if (a->fd_vpd < 0)
+ {
+ sysfs_obj_name(d, "vpd", namebuf);
+ a->fd_vpd = open(namebuf, O_RDONLY);
+ /* No warning on error; vpd may be absent or accessible only to root */
+ }
+ return a->fd_vpd;
+ }
+
+ if (a->fd < 0)
{
- if (a->fd >= 0)
- close(a->fd);
sysfs_obj_name(d, "config", namebuf);
a->fd_rw = a->writeable || intent == SETUP_WRITE_CONFIG;
a->fd = open(namebuf, a->fd_rw ? O_RDWR : O_RDONLY);
if (a->fd < 0)
a->warning("Cannot open %s", namebuf);
- a->fd_pos = 0;
}
-
- if (a->cached_dev != d)
- {
- if (a->fd_vpd >= 0)
- close(a->fd_vpd);
- sysfs_obj_name(d, "vpd", namebuf);
- a->fd_vpd = open(namebuf, O_RDONLY);
- /* No warning on error; vpd may be absent or accessible only to root */
- }
-
- a->cached_dev = d;
- return intent == SETUP_READ_VPD ? a->fd_vpd : a->fd;
+ return a->fd;
}
static int sysfs_read(struct pci_dev *d, int pos, byte *buf, int len)
if (fd < 0)
return 0;
- res = do_read(d, fd, buf, len, pos);
+ res = pread(fd, buf, len, pos);
if (res < 0)
{
d->access->warning("sysfs_read: read failed: %s", strerror(errno));
if (fd < 0)
return 0;
- res = do_write(d, fd, buf, len, pos);
+ res = pwrite(fd, buf, len, pos);
if (res < 0)
{
d->access->warning("sysfs_write: write failed: %s", strerror(errno));
return 1;
}
-#ifdef PCI_HAVE_DO_READ
-
-/* pread() is not available and do_read() only works for a single fd, so we
- * cannot implement read_vpd properly. */
-static int sysfs_read_vpd(struct pci_dev *d, int pos, byte *buf, int len)
-{
- return 0;
-}
-
-#else /* !PCI_HAVE_DO_READ */
-
static int sysfs_read_vpd(struct pci_dev *d, int pos, byte *buf, int len)
{
int fd = sysfs_setup(d, SETUP_READ_VPD);
return 1;
}
-#endif /* PCI_HAVE_DO_READ */
-
static void sysfs_cleanup_dev(struct pci_dev *d)
{
struct pci_access *a = d->access;
if (a->cached_dev == d)
- {
- a->cached_dev = NULL;
- close(a->fd);
- a->fd = -1;
- }
+ sysfs_flush_cache(a);
}
struct pci_methods pm_linux_sysfs = {
- "linux-sysfs",
- "The sys filesystem on Linux",
- sysfs_config,
- sysfs_detect,
- sysfs_init,
- sysfs_cleanup,
- sysfs_scan,
- sysfs_fill_info,
- sysfs_read,
- sysfs_write,
- sysfs_read_vpd,
- NULL, /* init_dev */
- sysfs_cleanup_dev
+ .name = "linux-sysfs",
+ .help = "The sys filesystem on Linux",
+ .config = sysfs_config,
+ .detect = sysfs_detect,
+ .init = sysfs_init,
+ .cleanup = sysfs_cleanup,
+ .scan = sysfs_scan,
+ .fill_info = sysfs_fill_info,
+ .read = sysfs_read,
+ .write = sysfs_write,
+ .read_vpd = sysfs_read_vpd,
+ .cleanup_dev = sysfs_cleanup_dev,
};