[LTP] [PATCH v3] userfaultfd: Minor fixes

[Cyril Hrubis]

Hi!
> As a separate effort: Cyril raised a question if (as a separate effort) should
> all test have some kconfig check (e.g. CONFIG_USERFAULTFD=y)
> 
> IMHO it's a question if we are ok with wide spread of needs_kconfig even if it's
> not necessary (i.e. here SAFE_USERFAULTFD() check. In that case the benefit
> would be 1) documentation of dependencies for testers in the test catalog 2)
> check earlier that before running test code. And of course disadvantage to
> really drag kconfig dependency, but we have accepted that already.
> 
> Ideally we'd decide on some policy (which may also includes .needs_drivers).
> @all I planned to reword my original policy patch [1] to Li's suggestion +
> Cyril's kconfig changes [2], but that still does not have "use kconfig to
> document dependencies".

I think that we can get rid of needs_drivers if we add more runtime
checks.

We have needs_drivers in:

- kvm tests -> CONFIG_KVM + runtime check for kvm module
- zram tests -> CONFIG_ZRAM + runtime check for zram module
- squasfs tests -> CONFIG_SQUASHFS + runtime check for squashfs module
- setxattr02 -> CONFIG_BLK_DEV_RAM + runtime check for brd module
- ioctl08 -> has both filesystems and needs_drivers set to btrfs
             likely we do not need the needs_drivers there
- ioctl_loop tests -> CONFIG_BLK_DEV_LOOP + runtime check for loop module
- madvise11 -> CONFIG_HWPOISON_INJECT + runtime check for hwpoison_inject module
- quotactl tests -> CONFIG_QUOTA_V2 + runtime check for quota_v2 module
- uinput tests -> CONFIG_BLK_DEV_LOOP + CONFIG_TUN + CONFIG_INPUT_UINPUT
- can tests -> CONFIG_CAN_VCAN + CONFIG_CAN_RAW + CONFIG_CAN_BCM


If we agree that we want to switch from needs_drivers to needs_kconfig +
runtime checks I will add mappings from congfigs to module names into
tst_kconfig and convert the tests..

-- 
Cyril Hrubis
chrubis@suse.cz