Hello Joel,

On Fri, Sep 29, 2023 at 11:51:43AM -0700, Joel Becker wrote:

On Thu, Sep 28, 2023 at 07:44:58AM -0700, Breno Leitao wrote:

quoted

Should I create a configfs_register_item() to solve this problem?

It's an express philosophy of configfs that all lifetimes are controlled
by userspace, which is why we don't have such a facility.  If hch wants
to change this, I defer to his judgement.  But I don't think it is
necessary.

I am happy the suggestion you gave, so, no need to change this
philosophy.

In this fashion, each console created on the command line will get a
name of `cmdline0`, `cmdline1`, etc.  They will not be part of the
configfs tree.  If the user comes along later and says `mkdir
/sys/kernel/config/netconsole/cmdline1`, the existing `cmdline1` console
will be attached to the configfs tree.  The user is then free to disable
and reconfigure the device.

I am happy with this approach.

There would, of course, be some other corner cases to handle.  Do we
allow dynamic names that look like command-line names if no command-line
parameter exists? 

I'd say so, I think `cmdline` targets shouldn't be special if there are
not netconsole targets defined at boot time.

Does `rmdir /sys/kernel/config/netconsole/cmdline0`
actually delete the command-line console entry, or does it return
-EBUSY?  And so on.

I've looked at the code and tested, it `cmdline0`could be removed as any
other target entry. On top of that, the user can create a new `cmdline0`
entry, and, it will be set with the default values for the attributes.

I am planning to send a patch based on what you suggested, and we can
continue the discuss there.

Thanks for the _very_ detailed feedback!
Breno