speedtch usbatm.h,1.4,1.5 usbatm2.c,1.13,1.14

[Roman Kagan]

  Hi Duncan,

On Wed, Feb 02, 2005 at 11:23:50PM +0300, Roman Kagan wrote:
> Still, to complete the experiment with the asynchronous heavy_init
> kthread shutdown (and to keep you busy reviewing :) I've made up the
> patch below against current CVS.

Another reminder of how easy it is to get refcounting wrong :(

Here's an incremental patch fixing three memleaks, two of them being
present in the synchronous version too: 1) extra refcount increment and
2) need to call subdriver's ->unbind if ->bind succeeded but there was a
failure later on in usbatm_usb_probe, since usbatm_usb_disconnect won't
be called in this case.

I'll be sending a complete async patch vs current CVS in a minute.

Cheers,
  Roman.


diff -u usbatm.c usbatm.c
--- usbatm.c	2 Feb 2005 20:22:27 -0000
+++ usbatm.c	7 Feb 2005 02:54:25 -0000
@@ -1068,7 +1068,7 @@
 
 	memset(instance, 0, sizeof(*instance));
 
-	kref_init(&instance->refcount);	/* one for USB */
+	kref_init(&instance->refcount);	/* dropped in usbatm_usb_disconnect */
 
 	init_MUTEX(&instance->serialize);
 
@@ -1188,11 +1188,15 @@
 	else
 		error = usbatm_atm_init(instance);
 
-	if (error)
+	if (error) {
+		if (instance->driver->unbind)
+			instance->driver->unbind(instance, intf);
+
+		usbatm_put_sub_instance(instance);
 		goto fail;
+	}
 
 	usb_get_dev(dev);
-	usbatm_get_instance(instance);	/* dropped in usbatm_usb_disconnect */
 	usb_set_intfdata(intf, instance);
 
 	return 0;