From hap9@epoch.ncsc.mil Wed Mar 22 13:26:14 2006 From: Ryan Wilson Subject: driver core: driver_bind attribute returns incorrect value To: gregkh@suse.de Cc: linux-kernel@vger.kernel.org Date: Wed, 22 Mar 2006 16:26:25 -0500 Message-Id: <1143062785.22254.15.camel@moss-tarheels.epoch.ncsc.mil> The manual driver <-> device binding attribute in sysfs doesn't return the correct value on failure or success of driver_probe_device. driver_probe_device returns 1 on success (the driver accepted the device) or 0 on probe failure (when the driver didn't accept the device but no real error occured). However, the attribute can't just return 0 or 1, it must return the number of bytes consumed from buf or an error value. Returning 0 indicates to userspace that nothing was written (even though the kernel has tried to do the bind/probe and failed). Returning 1 indicates that only one character was accepted in which case userspace will re-try the write with a partial string. A more correct version of driver_bind would return count (to indicate the entire string was consumed) when driver_probe_device returns 1 and -ENODEV when driver_probe_device returns 0. This patch makes that change. Signed-off-by: Ryan Wilson Signed-off-by: Greg Kroah-Hartman --- drivers/base/bus.c | 5 +++++ 1 file changed, 5 insertions(+) --- gregkh-2.6.orig/drivers/base/bus.c +++ gregkh-2.6/drivers/base/bus.c @@ -524,6 +524,11 @@ static int bus_rescan_devices_helper(str device_attach(dev); if (dev->parent) up(&dev->parent->sem); + + if (err > 0) /* success */ + err = count; + else if (err == 0) /* driver didn't accept device */ + err = -ENODEV; } return 0; }