[U-Boot] [PATCH] sunxi: Add support for eth1addr
Currently we will already fill ethaddr with a fixed unique address based on the SoCs serial (from the sid) to make sure that boards which use the integrated emac / gmac get a fixed mac rather then a random one.
On some boards (observed on 2 tablets using sdio rtl8703as wifi chips) the wifi does not come with a fixed mac either, so also set eth1addr, so that dts files can set an ethernet1 alias to get mac-address and local-mac-address filled for dt nodes describing the wifi controller.
Signed-off-by: Hans de Goede hdegoede@redhat.com --- board/sunxi/board.c | 12 ++++++++++++ 1 file changed, 12 insertions(+)
diff --git a/board/sunxi/board.c b/board/sunxi/board.c index b5a50f4..41d796c 100644 --- a/board/sunxi/board.c +++ b/board/sunxi/board.c @@ -641,6 +641,18 @@ int misc_init_r(void) eth_setenv_enetaddr("ethaddr", mac_addr); }
+ if (!getenv("eth1addr")) { + /* Non OUI / registered MAC address */ + mac_addr[0] = 0x12; + mac_addr[1] = (sid[0] >> 0) & 0xff; + mac_addr[2] = (sid[3] >> 24) & 0xff; + mac_addr[3] = (sid[3] >> 16) & 0xff; + mac_addr[4] = (sid[3] >> 8) & 0xff; + mac_addr[5] = (sid[3] >> 0) & 0xff; + + eth_setenv_enetaddr("eth1addr", mac_addr); + } + if (!getenv("serial#")) { snprintf(serial_string, sizeof(serial_string), "%08x%08x", sid[0], sid[3]);
On Sun, 2016-06-26 at 13:54 +0200, Hans de Goede wrote:
Currently we will already fill ethaddr with a fixed unique address based on the SoCs serial (from the sid) to make sure that boards which use the integrated emac / gmac get a fixed mac rather then a random one.
On some boards (observed on 2 tablets using sdio rtl8703as wifi chips) the wifi does not come with a fixed mac either, so also set eth1addr, so that dts files can set an ethernet1 alias to get mac-address and local-mac-address filled for dt nodes describing the wifi controller.
This does it unconditionally, won't having eth1addr show up for boards which only have one network device (WIFI or otherwise) be potentially confusing for users? i.e. lacking it would be a sign that the online guide you are following might not exactly be relevant to your board, or people seeing it and then wasting time trying to figure out how to use the second device. Of secondary concern (since I think it is far less liklely) would be confusing some software somewhere.
Also what is the impact on boards which have a second network device which does have a proper MAC address, either one which is part of the board or e.g. on a USB dongle? I think we don't want to override any burnt in MAC addresses.
Ian.
Signed-off-by: Hans de Goede hdegoede@redhat.com
board/sunxi/board.c | 12 ++++++++++++ 1 file changed, 12 insertions(+)
diff --git a/board/sunxi/board.c b/board/sunxi/board.c index b5a50f4..41d796c 100644 --- a/board/sunxi/board.c +++ b/board/sunxi/board.c @@ -641,6 +641,18 @@ int misc_init_r(void) eth_setenv_enetaddr("ethaddr", mac_addr); }
if (!getenv("eth1addr")) {/* Non OUI / registered MAC address */mac_addr[0] = 0x12;mac_addr[1] = (sid[0] >> 0) & 0xff;mac_addr[2] = (sid[3] >> 24) & 0xff;mac_addr[3] = (sid[3] >> 16) & 0xff;mac_addr[4] = (sid[3] >> 8) & 0xff;mac_addr[5] = (sid[3] >> 0) & 0xff;eth_setenv_enetaddr("eth1addr", mac_addr);}if (!getenv("serial#")) { snprintf(serial_string, sizeof(serial_string), "%08x%08x", sid[0], sid[3]);
Hi,
On 30-06-16 12:50, Ian Campbell wrote:
On Sun, 2016-06-26 at 13:54 +0200, Hans de Goede wrote:
Currently we will already fill ethaddr with a fixed unique address based on the SoCs serial (from the sid) to make sure that boards which use the integrated emac / gmac get a fixed mac rather then a random one.
On some boards (observed on 2 tablets using sdio rtl8703as wifi chips) the wifi does not come with a fixed mac either, so also set eth1addr, so that dts files can set an ethernet1 alias to get mac-address and local-mac-address filled for dt nodes describing the wifi controller.
This does it unconditionally, won't having eth1addr show up for boards which only have one network device (WIFI or otherwise) be potentially confusing for users? i.e. lacking it would be a sign that the online guide you are following might not exactly be relevant to your board, or people seeing it and then wasting time trying to figure out how to use the second device. Of secondary concern (since I think it is far less liklely) would be confusing some software somewhere.
Also what is the impact on boards which have a second network device which does have a proper MAC address, either one which is part of the board or e.g. on a USB dongle? I think we don't want to override any burnt in MAC addresses.
This just sets eth1addr in the u-boot env, it will only actually do something to the devicetree if there is an ethernet1 alias in there (which there should not be for e.g an usb dongle).
So this can only cause any issues if people add a ethernet1 alias when they should not.
As for overriding a burnt-in mac-address, if the ethernet1 alias is present then u-boot will only modify a pre-existing "mac-address" dt property, if none is present it will not touch it. It will unconditional modify / add a "local-mac-address" property. Drivers should only honor "local-mac-address" if there is no burnt-in mac-address, see e.g. :
https://github.com/jwrdegoede/rtl8723bs/commit/be5e87ea71677120ede4572132089...
Now if users have a driver which honors mac-address to override the burnt-in mac, and the add an alias ethernet1 and they add an "mac-address" property to the dt node the ethernet1 alias points to, then they get exactly what they ask for.
Regards,
Hans
Ian.
Signed-off-by: Hans de Goede hdegoede@redhat.com
board/sunxi/board.c | 12 ++++++++++++ 1 file changed, 12 insertions(+)
diff --git a/board/sunxi/board.c b/board/sunxi/board.c index b5a50f4..41d796c 100644 --- a/board/sunxi/board.c +++ b/board/sunxi/board.c @@ -641,6 +641,18 @@ int misc_init_r(void) eth_setenv_enetaddr("ethaddr", mac_addr); }
if (!getenv("eth1addr")) {/* Non OUI / registered MAC address */mac_addr[0] = 0x12;mac_addr[1] = (sid[0] >> 0) & 0xff;mac_addr[2] = (sid[3] >> 24) & 0xff;mac_addr[3] = (sid[3] >> 16) & 0xff;mac_addr[4] = (sid[3] >> 8) & 0xff;mac_addr[5] = (sid[3] >> 0) & 0xff;eth_setenv_enetaddr("eth1addr", mac_addr);}- if (!getenv("serial#")) { snprintf(serial_string,
sizeof(serial_string), "%08x%08x", sid[0], sid[3]);
On Thu, 2016-06-30 at 13:15 +0200, Hans de Goede wrote:
Hi,
On 30-06-16 12:50, Ian Campbell wrote:
On Sun, 2016-06-26 at 13:54 +0200, Hans de Goede wrote:
Currently we will already fill ethaddr with a fixed unique address based on the SoCs serial (from the sid) to make sure that boards which use the integrated emac / gmac get a fixed mac rather then a random one.
On some boards (observed on 2 tablets using sdio rtl8703as wifi chips) the wifi does not come with a fixed mac either, so also set eth1addr, so that dts files can set an ethernet1 alias to get mac-address and local-mac-address filled for dt nodes describing the wifi controller.
This does it unconditionally, won't having eth1addr show up for boards which only have one network device (WIFI or otherwise) be potentially confusing for users? i.e. lacking it would be a sign that the online guide you are following might not exactly be relevant to your board, or people seeing it and then wasting time trying to figure out how to use the second device. Of secondary concern (since I think it is far less liklely) would be confusing some software somewhere.
[...]
This just sets eth1addr in the u-boot env,
It's this which I worried might confuse people, people who notice eth1addr (perhaps due to tab completion on "printenv eth"?) will wonder where the eth1 device is and/or why it is not working for them.
The rest of what you say regarding how this goes on to interact with Linux makes sense to me, although even then having an eth1addr which does't correspond to the MAC used by the device under Linux (because it is burnt in and therefore eth1addr is ignored) still seems potentially confusing to me.
I see at least some platforms use a CONFIG_HAS_ETH1 to control this. It could also perhaps be gated on the DTB used by u-boot itself by looking at the mac-address for the ethernet1 alias (and ethernet2 too I suppose).
Ian.
Hi,
On 30-06-16 15:52, Ian Campbell wrote:
On Thu, 2016-06-30 at 13:15 +0200, Hans de Goede wrote:
Hi,
On 30-06-16 12:50, Ian Campbell wrote:
On Sun, 2016-06-26 at 13:54 +0200, Hans de Goede wrote:
Currently we will already fill ethaddr with a fixed unique address based on the SoCs serial (from the sid) to make sure that boards which use the integrated emac / gmac get a fixed mac rather then a random one.
On some boards (observed on 2 tablets using sdio rtl8703as wifi chips) the wifi does not come with a fixed mac either, so also set eth1addr, so that dts files can set an ethernet1 alias to get mac-address and local-mac-address filled for dt nodes describing the wifi controller.
This does it unconditionally, won't having eth1addr show up for boards which only have one network device (WIFI or otherwise) be potentially confusing for users? i.e. lacking it would be a sign that the online guide you are following might not exactly be relevant to your board, or people seeing it and then wasting time trying to figure out how to use the second device. Of secondary concern (since I think it is far less liklely) would be confusing some software somewhere.
[...]
This just sets eth1addr in the u-boot env,
It's this which I worried might confuse people, people who notice eth1addr (perhaps due to tab completion on "printenv eth"?) will wonder where the eth1 device is and/or why it is not working for them.
People who use the u-boot cmdline at all really are experienced users so TBH I'm not all that worried about this.
The rest of what you say regarding how this goes on to interact with Linux makes sense to me, although even then having an eth1addr which does't correspond to the MAC used by the device under Linux (because it is burnt in and therefore eth1addr is ignored) still seems potentially confusing to me.
I see at least some platforms use a CONFIG_HAS_ETH1 to control this. It could also perhaps be gated on the DTB used by u-boot itself by looking at the mac-address for the ethernet1 alias (and ethernet2 too I suppose).
Adding a Kconfig option just to have a cleaner u-boot env seems a bit like overkill to me, and parsing the dtb even more so.
Regards,
Hans
On Thu, 2016-06-30 at 17:07 +0200, Hans de Goede wrote:
Hi,
On 30-06-16 15:52, Ian Campbell wrote:
On Thu, 2016-06-30 at 13:15 +0200, Hans de Goede wrote:
Hi,
On 30-06-16 12:50, Ian Campbell wrote:
On Sun, 2016-06-26 at 13:54 +0200, Hans de Goede wrote:
Currently we will already fill ethaddr with a fixed unique address based on the SoCs serial (from the sid) to make sure that boards which use the integrated emac / gmac get a fixed mac rather then a random one.
On some boards (observed on 2 tablets using sdio rtl8703as wifi chips) the wifi does not come with a fixed mac either, so also set eth1addr, so that dts files can set an ethernet1 alias to get mac- address and local-mac-address filled for dt nodes describing the wifi controller.
This does it unconditionally, won't having eth1addr show up for boards which only have one network device (WIFI or otherwise) be potentially confusing for users? i.e. lacking it would be a sign that the online guide you are following might not exactly be relevant to your board, or people seeing it and then wasting time trying to figure out how to use the second device. Of secondary concern (since I think it is far less liklely) would be confusing some software somewhere.
[...]
This just sets eth1addr in the u-boot env,
It's this which I worried might confuse people, people who notice eth1addr (perhaps due to tab completion on "printenv eth"?) will wonder where the eth1 device is and/or why it is not working for them.
People who use the u-boot cmdline at all really are experienced users so TBH I'm not all that worried about this.
I know that I personally once wasted quite a bit of time (on arndale, but still) being mislead/confused by different eth*addr variables (arndale has another possible prefix too, usbaddr IIRC, which makes it doubly confusing) and what applies to what and when. So IME having irrelevant envvars like that floating around in the default env really does lead to confusion even for people who (supposedly) know what they are doing.
Even naive users will find random guides online which don't quite apply to their particular board (especially likely with the vast number of sunxi variants in existence) and get lead down the wrong path (which should be "fail early since the envvar discussed doesn't exist)
Ian.
participants (2)
-
Hans de Goede -
Ian Campbell