@kib : Only refcount_release() can be called w/o checking the returned result, so I didn't add an attribute for that function. The others should be fine from what I can see.

Update assert message, suggested by @kib .

Minor refactor.

I think the return value of refcount_release() should also be checked, but I see many users in the kernel which don't check it.

Use already existing refcount_acquire_if_not_zero(). Suggested by @mjg .

Update as per @mjg 's suggestion

Ping is this patch still relevant?

Make sure the return value of ifa_ref() is checked.

Make sure return value is checked.

Updated patch with full context.

@bz: Try modifying the script to create 65536 /dev/tun devices.

Silly question; can you explain that test case? How do 255 multicast addresses assigned to 255 interfaces make you run out of ifindex space?

Found more issues related - fix while at it.

Update as per kib's suggestion.

Patch with full context.

We have this patched queued internally for testing and review. You'd like to have it in before 12 is out?

I think "GBase" was the common before this change aswell.

Roberto, can you plug the failing device w/o your patch, setting:

I think your patch is wrong.

I see IFM_40G_ER4 is missing from the table. There is only IFM_40G_LR4 currently.

I think the reason HID picks up the device first will be solved by:
https://reviews.freebsd.org/D16735

Can you explain the change with a few words?

I'm just wondering if the name TOUCHSCREEN is too generic.