On Tue, 5 Feb 2008 15:44:13 -0800, Greg KH wrote:
quoted text
> On Tue, Feb 05, 2008 at 09:16:42PM +0100, Jean Delvare wrote:
> > This is 2.6.24, CONFIG_SYSFS_DEPRECATED=y:
> > 
> > # ls -l /sys/class/i2c-adapter/i2c-0
> > total 0
> > lrwxrwxrwx 1 root root    0 f?v  5 18:07 device -> ../../../devices/pci0000:00/0000:00:01.0/0000:01:00.0
> > lrwxrwxrwx 1 root root    0 f?v  5 18:17 i2c-dev:i2c-0 -> ../../../class/i2c-dev/i2c-0
> > -r--r--r-- 1 root root 4096 f?v  5 18:07 name
> > drwxr-xr-x 2 root root    0 f?v  5 18:17 power
> > lrwxrwxrwx 1 root root    0 f?v  5 18:07 subsystem -> ../../../class/i2c-adapter
> > -rw-r--r-- 1 root root 4096 f?v  5  2008 uevent
> > 
> > 2.6.24 rebuilt without CONFIG_SYSFS_DEPRECATED:
> > 
> > # ls -l /sys/class/i2c-adapter/i2c-0/
> > total 0
> > lrwxrwxrwx 1 root root    0 f?v  5 18:42 device -> ../../../../../../devices/pci0000:00/0000:00:01.0/0000:01:00.0
> > drwxr-xr-x 3 root root    0 f?v  5 18:42 i2c-0
> > -r--r--r-- 1 root root 4096 f?v  5 18:31 name
> > drwxr-xr-x 2 root root    0 f?v  5 18:42 power
> > lrwxrwxrwx 1 root root    0 f?v  5 18:31 subsystem -> ../../../../../../class/i2c-adapter
> > -rw-r--r-- 1 root root 4096 f?v  5  2008 uevent
> > 
> > The latter corresponds to what older kernels had ("i2c-0"). This means
> > that enabling CONFIG_SYSFS_DEPRECATED causes the i2c-dev class device
> > names to change.
> 
> Yes.  Well, no, not really.  The class device names are the same, it's
> just that it is a symlink in the DEPRECATED=Y name and a real directory
> in the =N case.
>
> > Isn't it supposed to be exactly the other way around, i.e. enabling
> > CONFIG_SYSFS_DEPRECATED should preserve the names as they were in
> > older kernels?
> 
> The real name is the same (look above, it's really called "i2c-0" in
> both cases, just that the symlink in the =Y case has to have a different
> name to handle the fact that there could be duplicate names in this
> format.
> 
> Does that help?

Not really. Why is =Y supposed to be a factor for duplicate names? The
exact same collision can happen in the =N case. This is exactly the
problem Markus and myself tried to address with
109f0e93b6b728f03c1eb4af02bc25d71b646c59. The i2c-dev class names its
class devices i2c-%d, and the firmware class as well (for i2c devices).
If 109f0e93b6b728f03c1eb4af02bc25d71b646c59 is going to be reverted,
then we will see the same collision again, but only for DEPRECATED=N.
That's rather confusing.

To add to the confusion, it seems that, in some cases, the potential
name collision is addressed by inserting a directory (named after the
class name) between the parent device and the class device. For example:

/sys/bus/pci/devices/0000:00:11.0/i2c-adapter/i2c-0

0000:00:11.0 is the PCI device, i2c-0 is the class device and
i2c-adapter is the class name.

So why isn't it done in all cases? This would solve the problem, as the
colliding class devices would finally get their own namespace:

/sys/bus/pci/devices/0000:00:11.0/i2c-adapter/i2c-0/i2c-dev/i2c-0
/sys/bus/pci/devices/0000:00:11.0/i2c-adapter/i2c-0/firmware/i2c-0

quoted text
> It's really to keep older programs still work properly.  The =N case is
> what you should pay attention to for all modern distros.

Unfortunately the =N case is the one that will break when you revert
109f0e93b6b728f03c1eb4af02bc25d71b646c59.

-- 
Jean Delvare
--
unsubscribe notice
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/