Fix error handling in sysfs registration

PATCH In review Matthew Dobson 2 participants 2 messages Mar 18, 2003 – Mar 25, 2003 Driver core

Share & tools

Matthew Dobson

author

2003-03-18 01:43 UTC #0 lore raw

PATCH 1/1 Fix error handling in sysfs registration

The cpu, memblk, and node driver/device registration should be a little more clean in the way it handles registration failures. Or at least *consistent* amongst the topology elements. Right now, failures are either silent, obscure, or leave things in an inconsistent state.

Cheers!

-Matt

drivers/base/cpu.c +5 -2

diff -Nur --exclude-from=/usr/src/.dontdiff linux-2.5.64-vanilla/drivers/base/cpu.c linux-2.5.64-sysfs_cleanup/drivers/base/cpu.c
@@ -48,6 +48,9 @@
 
 int __init cpu_dev_init(void)
 {
-	devclass_register(&cpu_devclass);
-	return driver_register(&cpu_driver);
+	int error;
+	if (!(error = devclass_register(&cpu_devclass)))
+		if (error = driver_register(&cpu_driver))
+			devclass_unregister(&cpu_devclass);
+	return error;
 }

drivers/base/memblk.c +6 -3

diff -Nur --exclude-from=/usr/src/.dontdiff linux-2.5.64-vanilla/drivers/base/memblk.c linux-2.5.64-sysfs_cleanup/drivers/base/memblk.c
@@ -47,9 +47,12 @@
 }
 
 
-static int __init register_memblk_type(void)
+int __init register_memblk_type(void)
 {
-	int error = devclass_register(&memblk_devclass);
-	return error ? error : driver_register(&memblk_driver);
+	int error;
+	if (!(error = devclass_register(&memblk_devclass)))
+		if (error = driver_register(&memblk_driver))
+			devclass_unregister(&memblk_devclass);
+	return error;
 }
 postcore_initcall(register_memblk_type);

drivers/base/node.c +6 -3

diff -Nur --exclude-from=/usr/src/.dontdiff linux-2.5.64-vanilla/drivers/base/node.c linux-2.5.64-sysfs_cleanup/drivers/base/node.c
@@ -89,9 +89,12 @@
 }
 
 
-static int __init register_node_type(void)
+int __init register_node_type(void)
 {
-	int error = devclass_register(&node_devclass);
-	return error ? error : driver_register(&node_driver);
+	int error;
+	if (!(error = devclass_register(&node_devclass)))
+		if (error = driver_register(&node_driver))
+			devclass_unregister(&node_devclass);
+	return error;
 }
 postcore_initcall(register_node_type);

Patrick Mochel

2003-03-25 21:58 UTC #1 lore raw

Sorry about the delay, I've been traveling..

Quoted 4 lines

The cpu, memblk, and node driver/device registration should be a little more clean in the way it handles registration failures. Or at least *consistent* amongst the topology elements. Right now, failures are either silent, obscure, or leave things in an inconsistent state.

Thanks. I've applied this, and will make sure it gets merged (finally).

-pat