| Message ID | 1386999720-23460-3-git-send-email-ynvich@gmail.com |
|---|---|
| State | Superseded |
| Delegated to: | Marek Vasut |
| Headers | show |
On Saturday, December 14, 2013 at 06:41:58 AM, Sergei Ianovich wrote: > I always used tftp in my test, so the first dm9000 on LP-8x4x was > always properly initialized. However, if the boot doesn't include > network related commands, linux will not find a valid MAC and will > complain. > > No longer. > > Signed-off-by: Sergei Ianovich <ynvich@gmail.com> > --- > board/icpdas/lp8x4x/lp8x4x.c | 8 ++++++++ > 1 file changed, 8 insertions(+) > > diff --git a/board/icpdas/lp8x4x/lp8x4x.c b/board/icpdas/lp8x4x/lp8x4x.c > index 1b68ef3..8396caa 100644 > --- a/board/icpdas/lp8x4x/lp8x4x.c > +++ b/board/icpdas/lp8x4x/lp8x4x.c > @@ -112,10 +112,18 @@ void usb_board_stop(void) > #ifdef CONFIG_DRIVER_DM9000 > void lp8x4x_eth1_mac_init(void) > { > + u8 ethaddr[8]; > u8 eth1addr[8]; > int i; > u8 reg; > > + eth_getenv_enetaddr_by_index("eth", 0, ethaddr); > + if (is_valid_ether_addr(ethaddr)) { > + for (i = 0, reg = 0x10; i < 6; i++, reg++) { > + writeb(reg, (u8 *)(DM9000_IO)); > + writeb(ethaddr[i], (u8 *)(DM9000_DATA)); > + } > + } > eth_getenv_enetaddr_by_index("eth", 1, eth1addr); > if (!is_valid_ether_addr(eth1addr)) > return; Please pass the ethernet address via DT, will that not work for you ? Best regards, Marek Vasut
On Sat, 2013-12-14 at 13:30 +0100, Marek Vasut wrote: > On Saturday, December 14, 2013 at 06:41:58 AM, Sergei Ianovich wrote: > > I always used tftp in my test, so the first dm9000 on LP-8x4x was > > always properly initialized. However, if the boot doesn't include > > network related commands, linux will not find a valid MAC and will > > complain. > > > Please pass the ethernet address via DT, will that not work for you ? This will require a custom dtb object for every device recompiled every time kernel is upgraded. The current way is to set once per device in U-Boot environment variable, which doesn't need to change if kernel changes. I strongly believe the current way is easier.
On Saturday, December 14, 2013 at 04:39:00 PM, Sergei Ianovich wrote: > On Sat, 2013-12-14 at 13:30 +0100, Marek Vasut wrote: > > On Saturday, December 14, 2013 at 06:41:58 AM, Sergei Ianovich wrote: > > > I always used tftp in my test, so the first dm9000 on LP-8x4x was > > > always properly initialized. However, if the boot doesn't include > > > network related commands, linux will not find a valid MAC and will > > > complain. > > > > Please pass the ethernet address via DT, will that not work for you ? > > This will require a custom dtb object for every device recompiled every > time kernel is upgraded. The current way is to set once per device in > U-Boot environment variable, which doesn't need to change if kernel > changes. > > I strongly believe the current way is easier. I disagree :) IF you set 'ethaddr' variable in U-Boot THEN U-Boot will patch this 'ethaddr' value into your DT. The /aliases/ethernet0 node will be augmented with a new property 'local-mac-address', which will contain the MAC address from 'ethaddr' . The kernel will use this as the MAC address for that particular ethernet interface afterwards. NOTE: It is very important to have the alias set, it has to point to your ethernet device. A good example in Linux's arch/arm/boot/dts/imx28.dtsi, which even has two ethernet interfaces. Notice each of them has an alias. NOTE: If you have two interfaces, then 'eth1addr' is patches into /aliases/ethernet1 etc. The only problem here is the non-DT kernel. Do you need to support that? Is there no other way to pass MAC address of an ethernet interface to Linux but programming it into the ethernet interface itself ? Best regards, Marek Vasut
On Sat, 2013-12-14 at 18:12 +0100, Marek Vasut wrote: > On Saturday, December 14, 2013 at 04:39:00 PM, Sergei Ianovich wrote: > > I strongly believe the current way is easier. > > I disagree :) > > IF you set 'ethaddr' variable in U-Boot THEN > U-Boot will patch this 'ethaddr' value into your DT. The /aliases/ethernet0 > node will be augmented with a new property 'local-mac-address', which will > contain the MAC address from 'ethaddr' . The kernel will use this as the MAC > address for that particular ethernet interface afterwards. > > NOTE: It is very important to have the alias set, it has to point to your > ethernet device. A good example in Linux's arch/arm/boot/dts/imx28.dtsi, which > even has two ethernet interfaces. Notice each of them has an alias. > NOTE: If you have two interfaces, then 'eth1addr' is patches into > /aliases/ethernet1 etc. Thanks for explaining. It works. This is COOL! > The only problem here is the non-DT kernel. Do you need to support that? Is > there no other way to pass MAC address of an ethernet interface to Linux but > programming it into the ethernet interface itself ? Hardware vendor uses U-Boot environment values to init MAC-addresses. I think it is even worse than what I was doing in the patch. I don't think we need to support non-DT kernel. So we drop the patch.
On Saturday, December 14, 2013 at 09:53:48 PM, Sergei Ianovich wrote: > On Sat, 2013-12-14 at 18:12 +0100, Marek Vasut wrote: > > On Saturday, December 14, 2013 at 04:39:00 PM, Sergei Ianovich wrote: > > > I strongly believe the current way is easier. > > > > I disagree :) > > > > IF you set 'ethaddr' variable in U-Boot THEN > > > > U-Boot will patch this 'ethaddr' value into your DT. The > > /aliases/ethernet0 node will be augmented with a new property > > 'local-mac-address', which will contain the MAC address from 'ethaddr' > > . The kernel will use this as the MAC address for that particular > > ethernet interface afterwards. > > > > NOTE: It is very important to have the alias set, it has to point to your > > ethernet device. A good example in Linux's arch/arm/boot/dts/imx28.dtsi, > > which even has two ethernet interfaces. Notice each of them has an > > alias. NOTE: If you have two interfaces, then 'eth1addr' is patches into > > /aliases/ethernet1 etc. > > Thanks for explaining. It works. This is COOL! Glad it helps :) > > The only problem here is the non-DT kernel. Do you need to support that? > > Is there no other way to pass MAC address of an ethernet interface to > > Linux but programming it into the ethernet interface itself ? > > Hardware vendor uses U-Boot environment values to init MAC-addresses. I > think it is even worse than what I was doing in the patch. > > I don't think we need to support non-DT kernel. So we drop the > patch. OK, that's nice :) Best regards, Marek Vasut
On Sunday, December 15, 2013 at 12:53:18 AM, Sergei Ianovich wrote: > On Sat, 2013-12-14 at 21:57 +0100, Marek Vasut wrote: > > > I don't think we need to support non-DT kernel. So we drop the > > > patch. > > > > OK, that's nice :) > > It may even make sense to remove the code that initializes eth1, right? Which code? Best regards, Marek Vasut
On Sun, 2013-12-15 at 06:17 +0100, Marek Vasut wrote: > On Sunday, December 15, 2013 at 12:53:18 AM, Sergei Ianovich wrote: > > It may even make sense to remove the code that initializes eth1, right? > > Which code? I mean: lp8x4x_eth1_mac_init() in board/icpdas/lp8x4x/lp8x4x.c and its invocation in board_eth_init()
On Sunday, December 15, 2013 at 10:44:06 AM, Sergei Ianovich wrote: > On Sun, 2013-12-15 at 06:17 +0100, Marek Vasut wrote: > > On Sunday, December 15, 2013 at 12:53:18 AM, Sergei Ianovich wrote: > > > It may even make sense to remove the code that initializes eth1, right? > > > > Which code? > > I mean: > > lp8x4x_eth1_mac_init() in board/icpdas/lp8x4x/lp8x4x.c > > and its invocation in board_eth_init() Heh, yes. The DT will solve it through the ethernet1 alias, just set the eth1addr and test it to make sure please. Best regards, Marek Vasut
On Sun, 2013-12-15 at 16:05 +0100, Marek Vasut wrote: > Heh, yes. The DT will solve it through the ethernet1 alias, just set the > eth1addr and test it to make sure please. It works. I've updated the dts in kernel for next version of patch series. I will update this patch for v2.
diff --git a/board/icpdas/lp8x4x/lp8x4x.c b/board/icpdas/lp8x4x/lp8x4x.c index 1b68ef3..8396caa 100644 --- a/board/icpdas/lp8x4x/lp8x4x.c +++ b/board/icpdas/lp8x4x/lp8x4x.c @@ -112,10 +112,18 @@ void usb_board_stop(void) #ifdef CONFIG_DRIVER_DM9000 void lp8x4x_eth1_mac_init(void) { + u8 ethaddr[8]; u8 eth1addr[8]; int i; u8 reg; + eth_getenv_enetaddr_by_index("eth", 0, ethaddr); + if (is_valid_ether_addr(ethaddr)) { + for (i = 0, reg = 0x10; i < 6; i++, reg++) { + writeb(reg, (u8 *)(DM9000_IO)); + writeb(ethaddr[i], (u8 *)(DM9000_DATA)); + } + } eth_getenv_enetaddr_by_index("eth", 1, eth1addr); if (!is_valid_ether_addr(eth1addr)) return;
I always used tftp in my test, so the first dm9000 on LP-8x4x was always properly initialized. However, if the boot doesn't include network related commands, linux will not find a valid MAC and will complain. No longer. Signed-off-by: Sergei Ianovich <ynvich@gmail.com> --- board/icpdas/lp8x4x/lp8x4x.c | 8 ++++++++ 1 file changed, 8 insertions(+)