Hi Bernhard,
On 29 June 2016 at 01:43, Bernhard Nortmann bernhard.nortmann@web.de wrote:
This patch modifies the usb_device_info() function to enumerate active USB hubs by iterating UCLASS_USB_HUB directly.
The previous code used UCLASS_USB nodes instead and retrieved their first child node, expecting it to be a hub. However, it did not protect against retrieving (and dereferencing) a NULL pointer this way.
Depending on the available USB hardware, this might happen easily. For example the USB controller on sun7i-a20 has top-level OHCI nodes that won't have any children as long as no USB1-only peripheral is attached. ("usb tree" will only show EHCI nodes.)
The failure can also be demonstrated with U-Boot's sandbox architecture, by simply enabling the inactive "usb_2" node in test.dts. This creates a similar situation, where the existing usb_device_info() implementation would crash (segfault) when issuing a "usb info" command.
Signed-off-by: Bernhard Nortmann bernhard.nortmann@web.de
cmd/usb.c | 15 +++++---------- 1 file changed, 5 insertions(+), 10 deletions(-)
Reviewed-by: Simon Glass sjg@chromium.org
My only concern is whether this changes the ordering (since the devices' devnum is printed out). But the existing ordering may not be any better anyway, and if we had a problem we could sort it before outputting the info, so it seems OK to me.
Thanks for fixing the bug.
- Simon