Skip to contentSkip to conversation

What is struct pci_driver.owner for?

DISCUSSION Roland Dreier 5 participants 8 messages Oct 18, 2005 – Oct 19, 2005 InfiniBand

lore
Share & tools
lore
#0 lore raw

I just noticed that at some point, struct pci_driver grew a .owner member. However, only a handful of drivers set it:

    $ grep -r -A10 pci_driver drivers/ | grep owner
    drivers/block/sx8.c-    .owner          = THIS_MODULE,
    drivers/ieee1394/pcilynx.c-     .owner =           THIS_MODULE,
    drivers/net/spider_net.c-       .owner          = THIS_MODULE,
    drivers/video/imsttfb.c-        .owner          = THIS_MODULE,
    drivers/video/kyro/fbdev.c-     .owner          = THIS_MODULE,
    drivers/video/tridentfb.c-      .owner  = THIS_MODULE,

Should all drivers be setting .owner = THIS_MODULE? Is this a good kernel janitors task?

- R.

#1 lore raw
Quoted 11 lines · Roland Dreier

I just noticed that at some point, struct pci_driver grew a .owner member. However, only a handful of drivers set it:

    $ grep -r -A10 pci_driver drivers/ | grep owner
    drivers/block/sx8.c-    .owner          = THIS_MODULE,
    drivers/ieee1394/pcilynx.c-     .owner =           THIS_MODULE,
    drivers/net/spider_net.c-       .owner          = THIS_MODULE,
    drivers/video/imsttfb.c-        .owner          = THIS_MODULE,
    drivers/video/kyro/fbdev.c-     .owner          = THIS_MODULE,
    drivers/video/tridentfb.c-      .owner  = THIS_MODULE,
Quoted 2 lines

Should all drivers be setting .owner = THIS_MODULE? Is this a good kernel janitors task?

In theory its for module refcounting. With so many PCI drivers and so few pci_driver::owner users, it makes me wonder how needed it is.

If it is needed (I've done no analysis, even though I am sx8.c author), then it should be applied uniformly.

Jeff

#2 lore raw
Quoted 12 lines · Jeff Garzik
Quoted 11 lines · Roland Dreier

I just noticed that at some point, struct pci_driver grew a .owner member. However, only a handful of drivers set it:

   $ grep -r -A10 pci_driver drivers/ | grep owner
   drivers/block/sx8.c-    .owner          = THIS_MODULE,
   drivers/ieee1394/pcilynx.c-     .owner =           THIS_MODULE,
   drivers/net/spider_net.c-       .owner          = THIS_MODULE,
   drivers/video/imsttfb.c-        .owner          = THIS_MODULE,
   drivers/video/kyro/fbdev.c-     .owner          = THIS_MODULE,
   drivers/video/tridentfb.c-      .owner  = THIS_MODULE,
Quoted 5 lines
Quoted 2 lines

Should all drivers be setting .owner = THIS_MODULE? Is this a good kernel janitors task?

In theory its for module refcounting. With so many PCI drivers and so few pci_driver::owner users, it makes me wonder how needed it is.

It might in the future be needed for refcounting, I originally added it when I thought it was needed.

But what it really does today is create the symlink from the driver to the module that is contained in it, in sysfs. Which is very invaluable for people who want to know these things (installer programs, etc.)

For example:

$ tree /sys/bus/pci/drivers/uhci_hcd/
/sys/bus/pci/drivers/uhci_hcd/
|-- 0000:00:1d.0 -> ../../../../devices/pci0000:00/0000:00:1d.0
|-- 0000:00:1d.1 -> ../../../../devices/pci0000:00/0000:00:1d.1
|-- 0000:00:1d.2 -> ../../../../devices/pci0000:00/0000:00:1d.2
|-- 0000:00:1d.3 -> ../../../../devices/pci0000:00/0000:00:1d.3
|-- bind
|-- module -> ../../../../module/uhci_hcd
|-- new_id
`-- unbind

That "module" symlink is created only if the .owner field is set. That's why people are going through and adding it to all of the drivers in the system.

Hope this helps,

greg k-h

#3 lore raw
Quoted 13 lines · Greg KH
Quoted 12 lines · Jeff Garzik
Quoted 11 lines · Roland Dreier

I just noticed that at some point, struct pci_driver grew a .owner member. However, only a handful of drivers set it:

   $ grep -r -A10 pci_driver drivers/ | grep owner
   drivers/block/sx8.c-    .owner          = THIS_MODULE,
   drivers/ieee1394/pcilynx.c-     .owner =           THIS_MODULE,
   drivers/net/spider_net.c-       .owner          = THIS_MODULE,
   drivers/video/imsttfb.c-        .owner          = THIS_MODULE,
   drivers/video/kyro/fbdev.c-     .owner          = THIS_MODULE,
   drivers/video/tridentfb.c-      .owner  = THIS_MODULE,
Quoted 8 lines
Quoted 5 lines
Quoted 2 lines

Should all drivers be setting .owner = THIS_MODULE? Is this a good kernel janitors task?

In theory its for module refcounting. With so many PCI drivers and so few pci_driver::owner users, it makes me wonder how needed it is.

It might in the future be needed for refcounting, I originally added it when I thought it was needed.

Note that both pci_driver and pci_driver.driver both have an "owner" field. One should go - there's no point needlessly duplicating stuff like this.

Signature · 3 lines
Russell King
 Linux kernel    2.6 ARM Linux   - http://www.arm.linux.org.uk/
 maintainer of:  2.6 Serial core
#4 lore raw
In reply to Greg KH

Greg> But what it really does today is create the symlink from the Greg> driver to the module that is contained in it, in sysfs. Greg> Which is very invaluable for people who want to know these Greg> things (installer programs, etc.)

Greg> That "module" symlink is created only if the .owner field is Greg> set. That's why people are going through and adding it to Greg> all of the drivers in the system.

OK, I'll make my own small contribution and push this for 2.6.15:

drivers/infiniband/hw/mthca/mthca_main.c +1 -0
@@ -1196,6 +1196,7 @@ MODULE_DEVICE_TABLE(pci, mthca_pci_table
 
 static struct pci_driver mthca_driver = {
 	.name		= DRV_NAME,
+	.owner		= THIS_MODULE,
 	.id_table	= mthca_pci_table,
 	.probe		= mthca_init_one,
 	.remove		= __devexit_p(mthca_remove_one)
#5 lore raw

Roland Dreier napsal(a):

Quoted 5 lines

Greg> But what it really does today is create the symlink from the Greg> driver to the module that is contained in it, in sysfs. Greg> Which is very invaluable for people who want to know these Greg> things (installer programs, etc.)

Quoted 7 lines

Greg> That "module" symlink is created only if the .owner field is Greg> set. That's why people are going through and adding it to Greg> all of the drivers in the system.

OK, I'll make my own small contribution and push this for 2.6.15:

Ok, but read Documentation/SubmittingPatches. At least Signed-off-by is missing.

drivers/infiniband/hw/mthca/mthca_main.c MODULE_DEVICE_TABLE(pci, mthca_pci_table · line 1202

Quoted 13 lines
drivers/infiniband/hw/mthca/mthca_main.c +1 -0
@@ -1196,6 +1196,7 @@ MODULE_DEVICE_TABLE(pci, mthca_pci_table
 
 static struct pci_driver mthca_driver = {
 	.name		= DRV_NAME,
+	.owner		= THIS_MODULE,
 	.id_table	= mthca_pci_table,
 	.probe		= mthca_init_one,
 	.remove		= __devexit_p(mthca_remove_one)

regards,

Signature · 3 lines
Jiri Slaby         www.fi.muni.cz/~xslaby
~\-/~      [email protected]      ~\-/~
B67499670407CE62ACC8 22A032CC55C339D47A7E
#6 lore raw

Jiri> Ok, but read Documentation/SubmittingPatches. At least Jiri> Signed-off-by is missing.

Thanks, but that wasn't really a patch for anyone to accept. It was just a note about what I just put into my git tree, which I'll ask Linus to pull from directly when 2.6.15 opens. Of course in the real commit message I put a nice description and Signed-off-by: line.

- R.

#7 lore raw

Roland Dreier napsal(a):

Quoted 9 lines

Jiri> Ok, but read Documentation/SubmittingPatches. At least Jiri> Signed-off-by is missing.

Thanks, but that wasn't really a patch for anyone to accept. It was just a note about what I just put into my git tree, which I'll ask Linus to pull from directly when 2.6.15 opens. Of course in the real commit message I put a nice description and Signed-off-by: line.

Oh, sorry.

Signature · 3 lines
Jiri Slaby         www.fi.muni.cz/~xslaby
~\-/~      [email protected]      ~\-/~
B67499670407CE62ACC8 22A032CC55C339D47A7E