On Tuesday 17 August 2004 02:13, Alan Cox wrote:

On Tue, Aug 17, 2004 at 01:35:11AM +0200, Bartlomiej Zolnierkiewicz wrote:

quoted

Alan, 'ide_key' protection misses driver specfiic /proc/entries.

Ok need to fix that

quoted

It is also still racy for some drivers because ide_register_hw() ->
init_hwif_data() sets hwif->key to zero - you must set hwif->hold to 1.

ide_register_hw holds ide_setting_sem. I think that should be ok ?

ide_setting_sem doesn't help situation when hwif is unregistered and some 
other driver is loaded later and takes this hwif using ide_register_hw().

Highly unlikely but possible.

quoted

Can't we solve the problem in simpler way by covering affected /proc
handlers with ide_setting_sem?

You'd need to refcount the ide objects because the 'data' value is long
lived. I did consider this but it seemed more complex. It also leaves end
users able to open an ide file and prevent unloading although I'm not
sure it is a big issue.

Yep. :/