Skip to contentSkip to conversation

PCI: create device, function symlinks in /sys/bus/pci/slots/N/

PATCH Superseded Alex Chiang 5 participants 8 messages Aug 18, 2008 – Aug 19, 2008 PCI

v1 of 2 Superseded Latest is v2 →

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 PCI: create device, function symlinks in /sys/bus/pci/slots/N/

Create convenience symlinks in sysfs, linking slots to devices and functions, and vice versa. These links make it easier for users to figure out which devices actually live in what slots.

The device symlink points to the device's first function.

For example:

sapphire:/sys/bus/pci/slots # ls
1 10 2 3 4 5 6 7 8 9

sapphire:/sys/bus/pci/slots # ls -l 3
total 0
-r--r--r-- 1 root root 65536 Aug 18 14:10 address

lrwxrwxrwx 1 root root     0 Aug 18 14:10 device -> ../../../../devices/pci0000:23/0000:23:01.0
lrwxrwxrwx 1 root root     0 Aug 18 14:10 function0 -> ../../../../devices/pci0000:23/0000:23:01.0
lrwxrwxrwx 1 root root     0 Aug 18 14:10 function1 -> ../../../../devices/pci0000:23/0000:23:01.1

sapphire:/sys/bus/pci/slots # ls -l 3/device/slot
lrwxrwxrwx 1 root root 0 Aug 18 14:13 3/device/slot -> ../../../bus/pci/slots/3

The original form of this patch was written by Matthew Wilcox, but did not have links from the sysfs slots/ directory pointing back at devices and functions.

---
 slot.c |   58 ++++++++++++++++++++++++++++++++++++++++++++++++++++++++++
 1 file changed, 58 insertions(+)
drivers/pci/slot.c +58 -0
index 7e5b85c..76095ac 100644
@@ -47,6 +47,60 @@ static ssize_t address_read_file(struct pci_slot *slot, char *buf)
 				slot->number);
 }
 
+static void remove_sysfs_files(struct pci_slot *slot)
+{
+	char func[10];
+	struct list_head *tmp;
+
+	list_for_each(tmp, &slot->bus->devices) {
+		struct pci_dev *dev = pci_dev_b(tmp);
+		if (PCI_SLOT(dev->devfn) != slot->number)
+			continue;
+		sysfs_remove_link(&dev->dev.kobj, "slot");
+
+		snprintf(func, 10, "function%d", PCI_FUNC(dev->devfn));
+		sysfs_remove_link(&slot->kobj, func);
+	}
+	sysfs_remove_link(&slot->kobj, "device");
+}
+
+static int create_sysfs_files(struct pci_slot *slot)
+{
+	int result, lastdev = -1;
+	char func[10];
+	struct list_head *tmp;
+
+	list_for_each(tmp, &slot->bus->devices) {
+		struct pci_dev *dev = pci_dev_b(tmp);
+		if (PCI_SLOT(dev->devfn) != slot->number)
+			continue;
+
+		result = sysfs_create_link(&dev->dev.kobj, &slot->kobj, "slot");
+		if (result)
+			goto fail;
+
+		if (PCI_SLOT(dev->devfn) != lastdev) {
+			lastdev = PCI_SLOT(dev->devfn);
+			result = sysfs_create_link(&slot->kobj,
+						   &dev->dev.kobj,
+						   "device");
+			if (result)
+				goto fail;
+		}
+
+		snprintf(func, 10, "function%d", PCI_FUNC(dev->devfn));
+		result = sysfs_create_link(&slot->kobj, &dev->dev.kobj, func);
+		if (result)
+			goto fail;
+	}
+
+	return 0;
+
+fail:
+	remove_sysfs_files(slot);
+	return result;
+}
+
 static void pci_slot_release(struct kobject *kobj)
 {
 	struct pci_slot *slot = to_pci_slot(kobj);
@@ -54,6 +108,8 @@ static void pci_slot_release(struct kobject *kobj)
 	pr_debug("%s: releasing pci_slot on %x:%d\n", __func__,
 		 slot->bus->number, slot->number);
 
+	remove_sysfs_files(slot);
+
 	list_del(&slot->list);
 
 	kfree(slot);
@@ -150,6 +206,8 @@ placeholder:
 	INIT_LIST_HEAD(&slot->list);
 	list_add(&slot->list, &parent->slots);
 
+	create_sysfs_files(slot);
+
 	/* Don't care if debug printk has a -1 for slot_nr */
 	pr_debug("%s: created pci_slot on %04x:%02x:%02x\n",
 		 __func__, pci_domain_nr(parent), parent->number, slot_nr);
#1 lore raw
Quoted 18 lines · Alex Chiang

Create convenience symlinks in sysfs, linking slots to devices and functions, and vice versa. These links make it easier for users to figure out which devices actually live in what slots.

The device symlink points to the device's first function.

For example:

sapphire:/sys/bus/pci/slots # ls
1 10 2 3 4 5 6 7 8 9

sapphire:/sys/bus/pci/slots # ls -l 3
total 0
-r--r--r-- 1 root root 65536 Aug 18 14:10 address

lrwxrwxrwx 1 root root     0 Aug 18 14:10 device -> ../../../../devices/pci0000:23/0000:23:01.0
lrwxrwxrwx 1 root root     0 Aug 18 14:10 function0 -> ../../../../devices/pci0000:23/0000:23:01.0
lrwxrwxrwx 1 root root     0 Aug 18 14:10 function1 -> ../../../../devices/pci0000:23/0000:23:01.1
Quoted 6 lines

sapphire:/sys/bus/pci/slots # ls -l 3/device/slot
lrwxrwxrwx 1 root root 0 Aug 18 14:13 3/device/slot -> ../../../bus/pci/slots/3

The original form of this patch was written by Matthew Wilcox, but did not have links from the sysfs slots/ directory pointing back at devices and functions.

As you are adding new sysfs files/symlinks, please document them in Documentation/ABI.

thanks,

greg k-h

#2 lore raw
In reply to Alex Chiang
Quoted 11 lines · Alex Chiang

Create convenience symlinks in sysfs, linking slots to devices and functions, and vice versa. These links make it easier for users to figure out which devices actually live in what slots.

The device symlink points to the device's first function.

For example:

sapphire:/sys/bus/pci/slots # ls
1 10 2 3 4 5 6 7 8 9

Quoted 7 lines

sapphire:/sys/bus/pci/slots # ls -l 3
total 0
-r--r--r-- 1 root root 65536 Aug 18 14:10 address

lrwxrwxrwx 1 root root     0 Aug 18 14:10 device -> ../../../../devices/pci0000:23/0000:23:01.0
lrwxrwxrwx 1 root root     0 Aug 18 14:10 function0 -> ../../../../devices/pci0000:23/0000:23:01.0
lrwxrwxrwx 1 root root     0 Aug 18 14:10 function1 -> ../../../../devices/pci0000:23/0000:23:01.1

What's the point with "device" and "function0"? They appear to be the same thing by definition.

-hpa

#3 lore raw

* H. Peter Anvin <[email protected]>:

Quoted 19 lines
Quoted 11 lines · Alex Chiang

Create convenience symlinks in sysfs, linking slots to devices and functions, and vice versa. These links make it easier for users to figure out which devices actually live in what slots.

The device symlink points to the device's first function.

For example:

sapphire:/sys/bus/pci/slots # ls
1 10 2 3 4 5 6 7 8 9

Quoted 6 lines

sapphire:/sys/bus/pci/slots # ls -l 3
total 0
-r--r--r-- 1 root root 65536 Aug 18 14:10 address

lrwxrwxrwx 1 root root     0 Aug 18 14:10 device -> ../../../../devices/pci0000:23/0000:23:01.0
lrwxrwxrwx 1 root root     0 Aug 18 14:10 function0 -> ../../../../devices/pci0000:23/0000:23:01.0
lrwxrwxrwx 1 root root     0 Aug 18 14:10 function1 -> ../../../../devices/pci0000:23/0000:23:01.1
Quoted 2 lines

What's the point with "device" and "function0"? They appear to be the same thing by definition.

Good point, I'll remove 'device'.

Thanks.

/ac

#4 lore raw
In reply to Alex Chiang
Quoted 3 lines · Alex Chiang

The original form of this patch was written by Matthew Wilcox, but did not have links from the sysfs slots/ directory pointing back at devices and functions.

I think the reason I didn't bother was that you could already get this information from the 'address' file. ie:

$ ls -l /sys/bus/pci/devices/`cat /sys/bus/pci/slots/3/address`*

But I don't think we had a way to go from a device to the slot it's in, without searching through all the slots for matching address.

Quoted 17 lines
drivers/pci/slot.c +17 -0
+static void remove_sysfs_files(struct pci_slot *slot)
+{
+	char func[10];
+	struct list_head *tmp;
+
+	list_for_each(tmp, &slot->bus->devices) {
+		struct pci_dev *dev = pci_dev_b(tmp);
+		if (PCI_SLOT(dev->devfn) != slot->number)
+			continue;
+		sysfs_remove_link(&dev->dev.kobj, "slot");
+
+		snprintf(func, 10, "function%d", PCI_FUNC(dev->devfn));
+		sysfs_remove_link(&slot->kobj, func);
+	}
+	sysfs_remove_link(&slot->kobj, "device");
+}
+

drivers/pci/slot.c static ssize_t address_read_file(struct pci_slot *slot, char *buf) · line 76

Quoted 10 lines
drivers/pci/slot.c +10 -0
+static int create_sysfs_files(struct pci_slot *slot)
+{
+	int result, lastdev = -1;
+	char func[10];
+	struct list_head *tmp;
+
+	list_for_each(tmp, &slot->bus->devices) {
+		struct pci_dev *dev = pci_dev_b(tmp);
+		if (PCI_SLOT(dev->devfn) != slot->number)
+			continue;

Why not use pci_get_slot()?

Quoted 51 lines
drivers/pci/slot.c +30 -1
+		result = sysfs_create_link(&dev->dev.kobj, &slot->kobj, "slot");
+		if (result)
+			goto fail;
+
+		if (PCI_SLOT(dev->devfn) != lastdev) {
+			lastdev = PCI_SLOT(dev->devfn);
+			result = sysfs_create_link(&slot->kobj,
+						   &dev->dev.kobj,
+						   "device");
+			if (result)
+				goto fail;
+		}
+
+		snprintf(func, 10, "function%d", PCI_FUNC(dev->devfn));
+		result = sysfs_create_link(&slot->kobj, &dev->dev.kobj, func);
+		if (result)
+			goto fail;
+	}
+
+	return 0;
+
+fail:
+	remove_sysfs_files(slot);
+	return result;
+}
+
 static void pci_slot_release(struct kobject *kobj)
 {
 	struct pci_slot *slot = to_pci_slot(kobj);
@@ -54,6 +108,8 @@ static void pci_slot_release(struct kobject *kobj)
 	pr_debug("%s: releasing pci_slot on %x:%d\n", __func__,
 		 slot->bus->number, slot->number);
 
+	remove_sysfs_files(slot);
+
 	list_del(&slot->list);
 
 	kfree(slot);
@@ -150,6 +206,8 @@ placeholder:
 	INIT_LIST_HEAD(&slot->list);
 	list_add(&slot->list, &parent->slots);
 
+	create_sysfs_files(slot);
+
 	/* Don't care if debug printk has a -1 for slot_nr */
 	pr_debug("%s: created pci_slot on %04x:%02x:%02x\n",
 		 __func__, pci_domain_nr(parent), parent->number, slot_nr);
--

To unsubscribe from this list: send the line "unsubscribe linux-pci" in the body of a message to [email protected]
More majordomo info at http://vger.kernel.org/majordomo-info.html

Signature · 4 lines
Intel are signing my paycheques ... these opinions are still mine
"Bill, look, we understand that you're interested in selling us this
operating system, but compare it to ours.  We can't possibly take such
a retrograde step."
#5 lore raw

* Matthew Wilcox <[email protected]>:

Quoted 5 lines · Alex Chiang
Quoted 3 lines

The original form of this patch was written by Matthew Wilcox, but did not have links from the sysfs slots/ directory pointing back at devices and functions.

Quoted 7 lines

I think the reason I didn't bother was that you could already get this information from the 'address' file. ie:

$ ls -l /sys/bus/pci/devices/`cat /sys/bus/pci/slots/3/address`*

But I don't think we had a way to go from a device to the slot it's in, without searching through all the slots for matching address.

Hm, ok. So I guess the tradeoff here is convenience vs. memory.

If others are opposed to a 'functionN' backlink, I don't have very strong feelings, but I thought it was useful.

Quoted 21 lines
Quoted 21 lines
drivers/pci/slot.c +21 -0
+static void remove_sysfs_files(struct pci_slot *slot)
+{
+	char func[10];
+	struct list_head *tmp;
+
+	list_for_each(tmp, &slot->bus->devices) {
+		struct pci_dev *dev = pci_dev_b(tmp);
+		if (PCI_SLOT(dev->devfn) != slot->number)
+			continue;
+		sysfs_remove_link(&dev->dev.kobj, "slot");
+
+		snprintf(func, 10, "function%d", PCI_FUNC(dev->devfn));
+		sysfs_remove_link(&slot->kobj, func);
+	}
+	sysfs_remove_link(&slot->kobj, "device");
+}
+
+static int create_sysfs_files(struct pci_slot *slot)
+{
+	int result, lastdev = -1;
+	char func[10];
Quoted 8 lines

drivers/pci/slot.c static ssize_t address_read_file(struct pci_slot *slot, char *buf) · line 58

Quoted 6 lines
drivers/pci/slot.c +6 -0
+	struct list_head *tmp;
+
+	list_for_each(tmp, &slot->bus->devices) {
+		struct pci_dev *dev = pci_dev_b(tmp);
+		if (PCI_SLOT(dev->devfn) != slot->number)
+			continue;

Why not use pci_get_slot()?

This will deadlock, because we're already holding pci_bus_sem as a writer, taken during pci_create_slot():

	down_write(&pci_bus_sem);

Also, it doesn't really help us get rid of a loop, since slot->number doesn't encode the entire devfn; it only has the device number. So we would still have to do something like this:

	for (i = 0; i < 8; i++) {
		/* XXX: deadlock! */
		dev = pci_get_slot(slot->bus, PCI_DEVFN(slot->number, i));
		if (!dev)
			break;

Of course, it is entirely possible that I misconstrued what you were trying to suggest, so if I missed your point, please let me know. :)

Thanks.

/ac

Quoted 59 lines
Quoted 52 lines
drivers/pci/slot.c +30 -1
+		result = sysfs_create_link(&dev->dev.kobj, &slot->kobj, "slot");
+		if (result)
+			goto fail;
+
+		if (PCI_SLOT(dev->devfn) != lastdev) {
+			lastdev = PCI_SLOT(dev->devfn);
+			result = sysfs_create_link(&slot->kobj,
+						   &dev->dev.kobj,
+						   "device");
+			if (result)
+				goto fail;
+		}
+
+		snprintf(func, 10, "function%d", PCI_FUNC(dev->devfn));
+		result = sysfs_create_link(&slot->kobj, &dev->dev.kobj, func);
+		if (result)
+			goto fail;
+	}
+
+	return 0;
+
+fail:
+	remove_sysfs_files(slot);
+	return result;
+}
+
 static void pci_slot_release(struct kobject *kobj)
 {
 	struct pci_slot *slot = to_pci_slot(kobj);
@@ -54,6 +108,8 @@ static void pci_slot_release(struct kobject *kobj)
 	pr_debug("%s: releasing pci_slot on %x:%d\n", __func__,
 		 slot->bus->number, slot->number);
 
+	remove_sysfs_files(slot);
+
 	list_del(&slot->list);
 
 	kfree(slot);
@@ -150,6 +206,8 @@ placeholder:
 	INIT_LIST_HEAD(&slot->list);
 	list_add(&slot->list, &parent->slots);
 
+	create_sysfs_files(slot);
+
 	/* Don't care if debug printk has a -1 for slot_nr */
 	pr_debug("%s: created pci_slot on %04x:%02x:%02x\n",
 		 __func__, pci_domain_nr(parent), parent->number, slot_nr);
--

To unsubscribe from this list: send the line "unsubscribe linux-pci" in the body of a message to [email protected]
More majordomo info at http://vger.kernel.org/majordomo-info.html

Signature · 4 lines
Intel are signing my paycheques ... these opinions are still mine
"Bill, look, we understand that you're interested in selling us this
operating system, but compare it to ours.  We can't possibly take such
a retrograde step."
#6 lore raw
In reply to Alex Chiang
Quoted 3 lines · Alex Chiang

Create convenience symlinks in sysfs, linking slots to devices and functions, and vice versa. These links make it easier for users to figure out which devices actually live in what slots.

This looks it would be quite useful but the symlinks are not adjusted during hot-add and hot-remove operations (e.g. stale symlinks persist after hot-remove and new symlinks are not created in response to hot-add) so it could confuse more than help on systems with PCI hotplug support.

Gary

Signature · 6 lines
Gary Hade
System x Enablement
IBM Linux Technology Center
503-578-4503  IBM T/L: 775-4503
[email protected]
http://www.ibm.com/linux/ltc
#7 lore raw

* Gary Hade <[email protected]>:

Quoted 11 lines · Alex Chiang
Quoted 4 lines

Create convenience symlinks in sysfs, linking slots to devices and functions, and vice versa. These links make it easier for users to figure out which devices actually live in what slots.

This looks it would be quite useful but the symlinks are not adjusted during hot-add and hot-remove operations (e.g. stale symlinks persist after hot-remove and new symlinks are not created in response to hot-add) so it could confuse more than help on systems with PCI hotplug support.

Hi Gary,

Yes, you're right. Willy's original version of the patch had symmetric functions in pci_release_dev/pci_scan_device, which I didn't include because I didn't understand why we needed that code.

Now I understand. :)

Thanks for pointing it out. I'll work on this for v2.

/ac