psys: Do not skip registering ipu_psys_bus for kernels >= 6.10

[jwrdegoede]

The new auxbus code-paths were not registering the ipu_psys_bus used for
psys->dev.bus = &ipu_psys_bus, this was causing udev to not be able to
properly enumerate the device. For example running:

udevadm test -a add /dev/ipu-psys0

would fail with the following error:

Failed to clone sd_device object: No such file or directory

This udev issue in turn was causing issues with udev rules to set
permissions on /dev/ipu-psys0 not working

Fix this by:

1. Rename ipu6_psys_bus to ipu_psys_bus, there is no need for separate
   names for this struct with different kernels and having the same name
   allows code sharing.

2. Switching back from using module_auxiliary_driver() to using
   module_init() / module_exit() codes for registering the drivers.

3. Move the now always used ipu_psys_init() and ipu_psys_exit()
   out of any ifdefs blocks, so that these functions now always
   (un)register the chrdev region and the ipu_psys_bus.

4. Remove the now duplicate chrdev region handling from the auxbus
   ipu6_psys_probe() and ipu6_psys_remove() functions (now handled
   by ipu_psys_init() and ipu_psys_exit()).

[hao-yao]

Hi Hans @jwrdegoede , I'm looking into this PR and rebased internally. Will update it later.

[jwrdegoede]

Hi Hans @jwrdegoede , I'm looking into this PR and rebased internally. Will update it later.

Note you can reproduce the issue (with a new enough kernel using the in kernel isys) by doing:

udevadm info -e > udevdb.txt

and then look for /dev/ipu-psys0 in udevdb.txt without this pull-request there will be no udevdb entry for that.

[bingbucao]

Here is the udev rule to change the owner on Chrome:
KERNELS=="ipu-psys[0-9]", OWNER="arc-camera", GROUP="arc-camera", MODE="0600"
psys->dev.bus = &ipu6_psys_bus can help to allow udev to apply the rule.
Is there some specific reason we need to bus_register()?

[hao-yao]

Hi Hans @jwrdegoede , I resolved a conflict and rebased your PR to my fork here which have synced with intel/ipu6-drivers/master, could you help to update to this PR? Thank you.

[hao-yao]

Hi Bingbu @bingbucao , seems the PSYS driver called bus_register() before kernel v6.10 but not after v6.10. As we are using the struct bus_type ipu_isys_bus, I think it's better to call bus_register() on it although we can change udev rule condition to let it work.

@jwrdegoede Hans, is there any other issues could we met if we don't call bus_register()?

[jwrdegoede]

@jwrdegoede Hans, is there any other issues could we met if we don't call bus_register()?

ATM the device's bus member is set to the ipu6-bus, setting a device's bus-member to a non registered bus is not supported and will result in undefined behavior. I'm surprised things work as well as they do without this patch.

You could try setting the device's bus member to NULL, I think that will just result in the device being attached to the virtual bus.

[bingbucao]

@jwrdegoede Hans, is there any other issues could we met if we don't call bus_register()?

ATM the device's bus member is set to the ipu6-bus, setting a device's bus-member to a non registered bus is not supported and will result in undefined behavior. I'm surprised things work as well as they do without this patch.

You could try setting the device's bus member to NULL, I think that will just result in the device being attached to the virtual bus.

ACK. We will land this PR.