| Message ID | 20190314072713.20666-1-kjlu@umn.edu (mailing list archive) |
|---|---|
| State | Superseded |
| Headers | show |
| Series | usb: u132-hcd: fix potential NULL pointer dereference | expand |
On Thu, Mar 14, 2019 at 02:27:11AM -0500, Kangjie Lu wrote: > In case create_singlethread_workqueue fails, the fix notifies > callers the error to avoid potential NULL pointer dereferences. > > Signed-off-by: Kangjie Lu <kjlu@umn.edu> > --- > drivers/usb/host/u132-hcd.c | 2 ++ > 1 file changed, 2 insertions(+) > > diff --git a/drivers/usb/host/u132-hcd.c b/drivers/usb/host/u132-hcd.c > index 934584f0a20a..6d5b532b03f8 100644 > --- a/drivers/usb/host/u132-hcd.c > +++ b/drivers/usb/host/u132-hcd.c > @@ -3203,6 +3203,8 @@ static int __init u132_hcd_init(void) > return -ENODEV; > printk(KERN_INFO "driver %s\n", hcd_name); > workqueue = create_singlethread_workqueue("u132"); > + if (unlikely(!workqueue)) You only ever use unlikely/likely if you can actually measure the difference with and without it. For stuff like this, it is not needed at all, and in fact, the compiler and CPU already know this type of thing, so it is going to be faster without it. And are you sure you properly unwound from anything that was created/initialized above these lines? thanks, greg k-h
On 3/19/19 8:56 AM, Greg Kroah-Hartman wrote: > On Thu, Mar 14, 2019 at 02:27:11AM -0500, Kangjie Lu wrote: >> In case create_singlethread_workqueue fails, the fix notifies >> callers the error to avoid potential NULL pointer dereferences. >> >> Signed-off-by: Kangjie Lu <kjlu@umn.edu> >> --- >> drivers/usb/host/u132-hcd.c | 2 ++ >> 1 file changed, 2 insertions(+) >> >> diff --git a/drivers/usb/host/u132-hcd.c b/drivers/usb/host/u132-hcd.c >> index 934584f0a20a..6d5b532b03f8 100644 >> --- a/drivers/usb/host/u132-hcd.c >> +++ b/drivers/usb/host/u132-hcd.c >> @@ -3203,6 +3203,8 @@ static int __init u132_hcd_init(void) >> return -ENODEV; >> printk(KERN_INFO "driver %s\n", hcd_name); >> workqueue = create_singlethread_workqueue("u132"); >> + if (unlikely(!workqueue)) > You only ever use unlikely/likely if you can actually measure the > difference with and without it. For stuff like this, it is not needed > at all, and in fact, the compiler and CPU already know this type of > thing, so it is going to be faster without it. > > And are you sure you properly unwound from anything that was > created/initialized above these lines? I didn't see what resources require release in this case. Given the existing error paths for usb_disabled() and platform_driver_register(), returning -ENOMEM looks like a consistent error handling. Let me know if it isn't. > > thanks, > > greg k-h
diff --git a/drivers/usb/host/u132-hcd.c b/drivers/usb/host/u132-hcd.c index 934584f0a20a..6d5b532b03f8 100644 --- a/drivers/usb/host/u132-hcd.c +++ b/drivers/usb/host/u132-hcd.c @@ -3203,6 +3203,8 @@ static int __init u132_hcd_init(void) return -ENODEV; printk(KERN_INFO "driver %s\n", hcd_name); workqueue = create_singlethread_workqueue("u132"); + if (unlikely(!workqueue)) + return -ENOMEM; retval = platform_driver_register(&u132_platform_driver); return retval; }
In case create_singlethread_workqueue fails, the fix notifies callers the error to avoid potential NULL pointer dereferences. Signed-off-by: Kangjie Lu <kjlu@umn.edu> --- drivers/usb/host/u132-hcd.c | 2 ++ 1 file changed, 2 insertions(+)