Re: Is there a bug in hotplug.functions w/ ${TYPE}.usermap?

[Stamatis Mitrofanis <ewstam@softhome.net>]

Heath Elwayne Petersen wrote:

>I have been trying to use linux-hotplug to run a script when a device is 
>recognized.  However, I have been running into a problem.  It seams that 
>there may be an error in the hotplug.functions load_drivers function.
>...
>If this is the case, the error is that the find command does not return a 
>false return code when it fails to find the file.  Therefore, the modprobe will always 
>be tried, and the following code, which attempts to execute 
>/etc/hotplug/$TYPE/$MODULE will never be executed.
>
To be honest, I find that load_driver part of the code ugly alltogether. 
 More on that below...

>OUTPUT OF: diff hotplug.functions.orig hotplug.functions
>---- Cut Here ----
>128c128,129
><           elif find $MODULE_DIR -name $MODULE.o >/dev/null 2>&1 &&
>---
>
>>          elif find $MODULE_DIR -name $MODULE.o 2>&1 \
>>                      | grep -q . >/dev/null 2>&1 &&
>>
>---- Cut Here ----
>
Firstly, that `find' line is checking that belongs in modprobe. It's 
modprobe's job to see if a (abstract) "driver" is available. modprobe 
has aliases, null/none drivers etc. so it's high-level.

If we're looking for a "module file" we sould be using insmod. The 
_command_ insmod. Even if it's the same executable as modprobe, it has 
different semantics.

load_modules is ugly because:
- code is not very unreadable (at least one error went unnoticed)
- it has many story-telling FIXMEs in it for very long (so, it may have 
a bad architecture)

It's the agent's job to map requests to drivers and modutil's job to 
load kernel modules. Using a load_driver function between these two is 
bad because it complicates things.

Why is the requirement for either *modules or the function *_map_modules 
from load_driver when the mapping can be done by the agent (possibly by 
calling its helper program)? The agent can simply call modprobe which is 
definitely high-level-enough. modprobe itself checks whether a driver is 
already loaded.

For agents that don't map drivers or don't work with kernel drivers at 
all (net, printer, xfree86) load_modules is of no use.

For driver-specific processing it is (much) better to use scripts as 
/etc/hotplug/usb/xxx.agent . They should be called in place of (and not 
in addition to) modprobe so that there can be greater flexibility. (the 
agent script delegates to a driver script)

I am working with David Brownell for a series of patches that will ease 
things in this area. I had already solved the problem by the time the 
previous message was posted!


_______________________________________________
Linux-hotplug-devel mailing list  http://linux-hotplug.sourceforge.net
Linux-hotplug-devel@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/linux-hotplug-devel