What is struct pci_driver.owner for?
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.
Quoted 11 lines · Roland Dreier
I just noticed that at some point, struct
pci_drivergrew 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
Quoted 12 lines · Jeff Garzik
Quoted 11 lines · Roland Dreier
I just noticed that at some point, struct
pci_drivergrew 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::ownerusers, 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
Quoted 13 lines · Greg KH
Quoted 12 lines · Jeff Garzik
Quoted 11 lines · Roland Dreier
I just noticed that at some point, struct
pci_drivergrew 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::ownerusers, 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
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_tablestatic 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)
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_tablestatic 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
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.
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