Patch - bug in adding/removing modules

Patch - bug in adding/removing modules

[Matthew Tippett]

[-- Attachment #1: Type: text/plain, Size: 1098 bytes --]

Although I was only looking at the battery module, I believe I have 
fixed a potential oops regarding inserting and removing modules.

The attached patch fixes a problem that I have seen, a quick look at 
other modules seems to have the same problem.

The description of the problem is ...

There is a macro called acpi_device_dir which returns a pointer to the a 
battery instance directory in /proc/acpi/battery.  When removing a 
module, the device is removed from the proc heirachy, but the entry in 
the device table is never set to null.  So consequently when the device 
is reinserted the add_fs call already sees the proc directory entry and 
consequently uses the /proc/acpi/battery and along the way corrupts some 
memory.

So the patch NULLs out the proc dir entry in the device structure in a 
manner symmetric with the creation and deletion of the proc directory.

Regards,

Matthew

-- 
Matthew Tippett - matthew.tippett-rieW9WUcm8FFJ04o6PK0Fg@public.gmane.org - (416) 435-4118
Technology Forum - http://www.technology-forum.org/
Commercial Open Source - http://www.commercialos.org/

[-- Attachment #2: battery.diff --]
[-- Type: text/plain, Size: 373 bytes --]

--- battery.c.orig	2002-12-15 14:14:51.000000000 -0500
+++ battery.c	2002-12-15 14:11:29.000000000 -0500
@@ -658,8 +658,10 @@
 {
 	ACPI_FUNCTION_TRACE("acpi_battery_remove_fs");
 
-	if (acpi_device_dir(device)) 
+	if (acpi_device_dir(device)) {
 		remove_proc_entry(acpi_device_bid(device), acpi_battery_dir);
+		acpi_device_dir(device) = NULL;
+	}
 
 	return_VALUE(0);
 }