> On 01/09/2012 10:08 PM, Vincent Palatin wrote:
> > On Mon, Jan 9, 2012 at 12:57, Marek Vasut <marek.va...@gmail.com> wrote:
> >>> Fix the crash when running several times usb_init() with a USB ethernet
> >>> device plugged.
> >>>
> >>> Signed-off-by: Vincent Palatin <vpala...@chromium.org>
> >>> Tested-by: Wolfgang Grandegger <w...@denx.de>
> >>> ---
> >>>
> >>> drivers/usb/eth/usb_ether.c | 7 +++++--
> >>> include/net.h | 1 +
> >>> net/eth.c | 29 +++++++++++++++++++++++++++++
> >>> 3 files changed, 35 insertions(+), 2 deletions(-)
> >>>
> >>> diff --git a/drivers/usb/eth/usb_ether.c b/drivers/usb/eth/usb_ether.c
> >>> index 6565ea5..73a0790 100644
> >>> --- a/drivers/usb/eth/usb_ether.c
> >>> +++ b/drivers/usb/eth/usb_ether.c
> >>> @@ -127,8 +127,11 @@ int usb_host_eth_scan(int mode)
> >>>
> >>> old_async = usb_disable_asynch(1); /* asynch transfer not allowed
> >>> */
> >>>
> >>> - for (i = 0; i < USB_MAX_ETH_DEV; i++)
> >>> - memset(&usb_eth[i], 0, sizeof(usb_eth[i]));
> >>> + /* unregister a previously detected device */
> >>> + for (i = 0; i < usb_max_eth_dev; i++)
> >>> + eth_unregister(&usb_eth[i].eth_dev);
> >>> +
> >>> + memset(usb_eth, 0, sizeof(usb_eth));
> >>>
> >>> for (i = 0; prob_dev[i].probe; i++) {
> >>>
> >>> if (prob_dev[i].before_probe)
> >>>
> >>> diff --git a/include/net.h b/include/net.h
> >>> index e4d42c2..1707a7f 100644
> >>> --- a/include/net.h
> >>> +++ b/include/net.h
> >>> @@ -96,6 +96,7 @@ struct eth_device {
> >>>
> >>> extern int eth_initialize(bd_t *bis); /* Initialize network
> >>> subsystem
> >>
> >> */
> >>
> >>> extern int eth_register(struct eth_device* dev);/* Register network
> >>> device
> >>>
> >>> */ +extern int eth_unregister(struct eth_device *dev);/* Remove network
> >>> device */ extern void eth_try_another(int first_restart); /* Change
> >>> the device */ extern void eth_set_current(void); /* set
> >>> nterface to
> >>
> >> ethcur
> >>
> >>> var */ extern struct eth_device *eth_get_dev(void); /* get the
> >>> current device MAC */ diff --git a/net/eth.c b/net/eth.c
> >>> index b4b9b43..3fb5fb6 100644
> >>> --- a/net/eth.c
> >>> +++ b/net/eth.c
> >>> @@ -224,6 +224,35 @@ int eth_register(struct eth_device *dev)
> >>>
> >>> return 0;
> >>>
> >>> }
> >>>
> >>> +int eth_unregister(struct eth_device *dev)
> >>> +{
> >>> + struct eth_device *cur;
> >>> +
> >>> + /* No device */
> >>> + if (!eth_devices)
> >>> + return -1;
> >>> +
> >>> + for (cur = eth_devices; cur->next != eth_devices && cur->next !=
> >>> dev; + cur = cur->next)
> >>> + ;
> >>> +
> >>> + /* Device not found */
> >>> + if (cur->next != dev)
> >>> + return -1;
> >>> +
> >>> + cur->next = dev->next;
> >>> +
> >>> + if (eth_devices == dev)
> >>> + eth_devices = dev->next == eth_devices ? NULL :
> >>> dev->next; +
> >>> + if (eth_current == dev) {
> >>> + eth_current = eth_devices;
> >>> + eth_current_changed();
> >>> + }
> >>> +
> >>> + return 0;
> >>> +}
> >>> +
> >>>
> >>> int eth_initialize(bd_t *bis)
> >>> {
> >>>
> >>> int num_devices = 0;
> >>
> >> Looks sane. On what hardware did this happen (just for the record)?
> >
> > I got it on x86 and ARM (tegra2) platforms (both with an EHCI
> > controller). I did most of my testing with ASIX-based USB-ethernet
> > dongle, but the issue seems pretty generic.
>
> It happens with *any* hardware calling eth_register() more than once
> because "struct eth_device *dev*" is memset 0 before calling
> eth_register() resulting in an endless loop when scanning the devices.
Yes, I'll try to remember to test this in the evening.
M
>
> Viele Grüße,
>
> Wolfgang
_______________________________________________
U-Boot mailing list
U-Boot@lists.denx.de
http://lists.denx.de/mailman/listinfo/u-boot
