Hi Jonas, On Tue, Feb 18, 2014 at 11:16 AM, Jonas Gorski <j...@openwrt.org> wrote: > On Tue, Feb 18, 2014 at 10:52 AM, Helmut Schaa > <helmut.sc...@googlemail.com> wrote: >> On Tue, Feb 18, 2014 at 10:46 AM, Helmut Schaa >> <helmut.sc...@googlemail.com> wrote: >>> On Mon, Feb 17, 2014 at 5:37 PM, Jonas Gorski <j...@openwrt.org> wrote: >>>> On Thu, Feb 13, 2014 at 5:00 PM, Helmut Schaa >>>> <helmut.sc...@googlemail.com> wrote: >>>>> When setting the associated interface down and up again a new >>>>> switch device will be registered due to b53_phy_config_init >>>>> doing the necessary allocations and registrations. >>>>> >>>>> Instead, register the switch device already in b53_phy_probe. >>>>> >>>>> Signed-off-by: Helmut Schaa <helmut.sc...@googlemail.com> >>>>> --- >>>> >>>> It would be good to have some changelog here to know what changed >>>> between v1, v2 and v3. >>> >>> Right, sorry bout that ... >>> >>>> As far as I can tell you are now dropping the >>>> ethX alias completely? >>> >>> Correct, since the netdev is only known when attaching to the phy >>> which caused the problem I wanted to fix in the first place ... >>> >>> And I agree with your previous comment that it is not safe to >>> replace the phy name on attaching the netdev to the phy (for various >>> reasons) ... >>> >>> Now, the name is consistent, but the dev_name of the actual switch >>> is used. Which means you will get a switch device called "bcm53018" >>> or "bcm53125" (would be bad if you have multiple switches). >> >> Actually, this is what the mmap and spi versions do as far as I can see. >> Helmut > > You are thinking too complicated here, all switches are also available > under a generic switchX name, so usually you only need to do ethX => > switch0 replacement (unless there are two switches, but I never saw > such a device yet on bcm47xx or bcm63xx).
Indeed, didn't notice the switchX alias yet. I haven't found any time to update the patch yet. Might take some more time, until then I carry this patch in my tree ... > The bigger problem is the conversion from the old kmod-switch > standard, where you had to explicitly untag the cpu if you wanted it > untagged with 'u', and the cpu was always implicitly tagged. This seems unrelated or did I miss anything? Helmut _______________________________________________ openwrt-devel mailing list openwrt-devel@lists.openwrt.org https://lists.openwrt.org/cgi-bin/mailman/listinfo/openwrt-devel
