Hi All,
I finally got a few spare hours to do some U-Boot hacking and to my dismay found the build for my board (eNET) had broken. It builds OK, but crashed during Ethernet initialisation. I tried winding back to a last known good commit without much luck.
I realised I had upgraded to Ubuntu 10.10 lately, and thought maybe a tool-chain change was the culprit (was running gcc-4.4.1 and binutils 2.19 from source, now running Ubuntu gcc 4.4.5 and Binutils 2.20)
But, I think that commit ecee9324d73555e744593f3e0d387bec4c566f55 may be 'not quite right'. My board uses the rtl8139 driver, and the following patch gets my board booting again:
diff --git a/drivers/net/rtl8139.c b/drivers/net/rtl8139.c index db8a727..3646148 100644 --- a/drivers/net/rtl8139.c +++ b/drivers/net/rtl8139.c @@ -229,6 +229,7 @@ int rtl8139_initialize(bd_t *bis) dev->halt = rtl_disable; dev->send = rtl_transmit; dev->recv = rtl_poll; + dev->write_hwaddr = NULL; #ifdef CONFIG_MCAST_TFTP dev->mcast = rtl_bcast_addr; #endif
Just before this code we see that dev is malloc'd:
dev = (struct eth_device *)malloc(sizeof *dev);
So there is no guarantee that dev is NULL'd.
Has anyone else run into similar problems?
I saw discussion a little while ago regarding implementing a version of malloc that returns cleared memory - did this gain any traction?
Regards,
Graeme