Re: VME: devices not removed after commit 050c3d52cc7

[Paul Gortmaker <paul.gortmaker@windriver.com>]

[Adding Martyn to Cc]

[VME: devices not removed after commit 050c3d52cc7] On 13/01/2017 (Fri 11:03) Stefano Babic wrote:

> Hi,
> 
> I have updated a custom VME device driver (mainly based on vme_user.c)
> to 4.9 (previously it was for 3.14-).
> 
> I see that VME device drivers cannot be loaded and unloaded due to this
> commit:
> 
> commit 050c3d52cc7810d9d17b8cd231708609af6876ae
> Author: Paul Gortmaker <paul.gortmaker@windriver.com>
> Date:   Sun Jul 3 14:05:56 2016 -0400
> 
>     vme: make core vme support explicitly non-modular

I've gone back and looked at this, and vme_user.c and I'm not yet 100%
convinced this is the right conclusion.  But perhaps, and I've put
Martyn on the Cc, in the hopes that he can clarify as well, if needed.

> 
> 
> In fact, this drops the remove function, that scans all devices attached
> to the bus and call their remove function.

So I guess my confusion here is between removal of a VME device, vs. the
removal of a complete VME bus.  The above commit you reference was based
on the premise that removal of a VME bus is not supported.  Which is not
to say that a VME device removal is not supported.

> 
> That means that "remove" entry points in VME device driver (let see in
> drivers/staging/vme/devices/vme_user.c) are now dead code and the
> required cleanup code is not called at all (devices and class are not
> removed). Reloading the same driver cause errors due to the missing
> cleanup by unloading.  This does not let build VME device drivers as
> module, as it is supposed to be done.

Again, I don't think this analysis is 100% right, but I can't be sure
because your driver is out of tree and I don't know what it does
precisely.  Looking at vme_user.c example, it has its own .remove
function that should be executed at module unload, and that would do all
the cleanup (see vme_user_remove).

> 
> Paul, what do you mind ?

For sure, we can restore the .remove and vme_bus_remove portions of that
commit if it is a real regression against a correct use of the
infrastructure, but I'm still not clear how you'd be triggering the
vme_bus_remove unless the vme device driver was going up into its
parent's bus struct directly.  Maybe Martyn can spot where I've
misunderstood the bus vs. device separation here.

Paul.
--

> 
> Best regards,
> Stefano Babic
> 
> 
> -- 
> =====================================================================
> DENX Software Engineering GmbH,      Managing Director: Wolfgang Denk
> HRB 165235 Munich, Office: Kirchenstr.5, D-82194 Groebenzell, Germany
> Phone: +49-8142-66989-53 Fax: +49-8142-66989-80 Email: sbabic@denx.de
> =====================================================================