| Message ID | 1314362614-27958-1-git-send-email-inguin@gmx.de |
|---|---|
| State | Changes Requested |
| Delegated to: | Joe Hershberger |
| Headers | show |
On Friday, August 26, 2011 08:43:34 Ingo van Lil wrote: > @@ -261,11 +261,6 @@ int eth_initialize(bd_t *bis) > > memcpy(dev->enetaddr, env_enetaddr, 6); > } > - if (dev->write_hwaddr && > - !eth_mac_skip(eth_number) && > - is_valid_ether_addr(dev->enetaddr)) { > - dev->write_hwaddr(dev); > - } removing this breaks the original reason write_hwaddr was added in the first place: people want the MAC programmed into the hardware even if they dont use the network. we should probably add an explicit comment above this block so people dont try and "fix it" in the future. > @@ -347,8 +342,14 @@ int eth_init(bd_t *bis) > do { > uchar env_enetaddr[6]; > > - if (eth_getenv_enetaddr_by_index(eth_number, env_enetaddr)) > + if (eth_getenv_enetaddr_by_index(eth_number, env_enetaddr)) { > memcpy(dev->enetaddr, env_enetaddr, 6); > + if (dev->write_hwaddr && > + !eth_mac_skip(eth_number) && > + is_valid_ether_addr(dev->enetaddr)) { > + dev->write_hwaddr(dev); > + } > + } this changes the current API between the common layer and the drivers. i'm not saying that's a bad thing, just that we need to: (1) update the documentation (see doc/README.{drivers.eth,enetaddr}) (2) update the drivers which manually call their write_hwaddr() in their init() func to no longer do that -mike
diff --git a/net/eth.c b/net/eth.c index 6082c90..2b43ae7 100644 --- a/net/eth.c +++ b/net/eth.c @@ -261,11 +261,6 @@ int eth_initialize(bd_t *bis) memcpy(dev->enetaddr, env_enetaddr, 6); } - if (dev->write_hwaddr && - !eth_mac_skip(eth_number) && - is_valid_ether_addr(dev->enetaddr)) { - dev->write_hwaddr(dev); - } eth_number++; dev = dev->next; @@ -347,8 +342,14 @@ int eth_init(bd_t *bis) do { uchar env_enetaddr[6]; - if (eth_getenv_enetaddr_by_index(eth_number, env_enetaddr)) + if (eth_getenv_enetaddr_by_index(eth_number, env_enetaddr)) { memcpy(dev->enetaddr, env_enetaddr, 6); + if (dev->write_hwaddr && + !eth_mac_skip(eth_number) && + is_valid_ether_addr(dev->enetaddr)) { + dev->write_hwaddr(dev); + } + } ++eth_number; dev = dev->next;
Currently the ethernet MAC address is read from the 'ethaddr' environment variable into the dev->enetaddr field each time the network hardware is initialized, but it is written to the actual hardware only once at board startup. When 'ethaddr' is set or changed after startup the device can no longer receive packets because the hardware filter is still programmed to the old MAC. This patch moves the writing of the hardware address from eth_initialize (board startup) to eth_init (just before actually using the network). Signed-off-by: Ingo van Lil <inguin@gmx.de> --- net/eth.c | 13 +++++++------ 1 files changed, 7 insertions(+), 6 deletions(-)