Comment # 47 on bug 907393 from Luis Rodriguez

(In reply to Werner Fink from comment #40)
> (In reply to Robert Milasan from comment #39)
> 
> The only explanation is that the kernel module takes this time.

Which kernel module?

>  As patch
> 
>   1087-infinit-timeout-for-kmod-loaded-modules.patch
> 
> has been added due bug #889297 to avoid that a running kmod/modprobe is
> killed and cause other troubles.  The patch also includes a kernel command
> line option
> 
>      udev.killkmod=1
> 
> which should allow udevd to kill a running kmod/modprobe.

Does that work around work?

> But be aware that
> this might cause other trouble (compare with bug #889297).

Indeed, its exactly why it was advised against the timeout to kill the driver,
as the error paths could cause more issues than not.

> The question
> rises what is better, a longer boot time to long running hardware and kernel
> driver  	initialization or an unstable system due a killed kmod/modprobe.

The upstream agreed upon solution was to device async probe support in the
kernel as that is what systemd wanted, and then let system use it. Kernel
support for async probe support is underway, I'm ironing final kinks with
Dmitry before posting a final patch set. If we want to test that set early I am
comfortable with a patch set we can enable and if system wants it can opt in
for it early on OpenSUSE. A module parameter would just need to be passed onto
every module loaded through kmod.

> IMHO and due bug #889297 this is a kernel issue.

Its a long standing collateral design issue on systemd caused by folks making
the assumption we probed asynchronously on modules. Upstream-wise we decided
then to implement support for this but by no means can one then consider this a
kernel issue as async probe is just about to get merged and its a new feature.
Systemd design needs to be considered while we get this support upstream on the
kernel.

> The kernel guys have  ordered this patch know they have to fix the affected driver.

Which driver?

If not a driver can you please explain exactly how the patch introduced is
causing an issue if the patch was supposed to just to allow no timeout when
loading modules with kmod?