Bringing up new hardware, and GPIO definitions

[Philip Prindeville]

Seems logger isn't available in preinit, so I have to echo "$@" >/dev/kmsg instead.

I made this further change:

diff --git a/target/linux/x86/base-files/etc/board.d/02_network b/target/linux/x86/base-files/etc/board.d/02_network
index e00e8c04dd..188cff7e1c 100644
--- a/target/linux/x86/base-files/etc/board.d/02_network
+++ b/target/linux/x86/base-files/etc/board.d/02_network
@@ -75,6 +75,18 @@ traverse-technologies-geos)
        macaddr="$(cat /sys/class/net/eth0/address)" 2>/dev/null
        [ -n "$macaddr" ] && ucidef_set_interface_macaddr "wan" "$macaddr"
        ;;
+silicom-80500-0214-g02-sl01a)
+       ucidef_set_network_device_path "eth0" "pci0000:00/0000:00:17.0/0000:02:00.1"
+       ucidef_set_network_device_path "eth1" "pci0000:00/0000:00:17.0/0000:02:00.0"
+       ucidef_set_network_device_path "eth2" "pci0000:00/0000:00:16.0/0000:03:00.0"
+       ucidef_set_network_device_path "eth3" "pci0000:00/0000:00:16.0/0000:03:00.1"
+       ucidef_set_network_device_path "eth4" "pci0000:00/0000:00:0c.0/0000:04:00.0"
+       ucidef_set_network_device_path "eth5" "pci0000:00/0000:00:0e.0/0000:05:00.0"
+       ucidef_set_network_device_path "eth6" "pci0000:00/0000:00:0f.0/0000:06:00.0"
+       ucidef_set_network_device_path "eth7" "pci0000:00/0000:00:10.0/0000:07:00.0"
+
+       ucidef_set_interfaces_lan_wan "eth4 eth5 eth6 eth7" "eth0 eth1 eth2 eth3"
+       ;;
 esac
 board_config_flush


Digging into this, the problem seems to be here in preinit_config_board()

        # Find the current highest eth*
        max_eth=$(grep -o '^ *eth[0-9]*:' /proc/net/dev | tr -dc '[0-9]\n' | sort -n | tail -1)
        # Find and move netdevs using eth*s we are configuring
        json_get_keys keys "network_device"
        for netdev in $keys; do
                json_select "network_device"
                        json_select "$netdev"
                                json_get_vars path path
                                if [ -n "$path" -a -h "/sys/class/net/$netdev" ]; then
                                        ip link set "$netdev" down
                                        ip link set "$netdev" name eth$((++max_eth))
                                fi
                        json_select ..
                json_select ..
        done


Before this block runs, all that exists in /sys/class/net/* is "lo".  None of the other devices have yet been created, because they're of type "ixgbe" or "igc"...

The difference being that they get loaded not with /etc/modules-boot.d/* but with /etc/modules.d/* ... which happens *after* the preinit script runs.

Looking at netdevices.mk it seems a little arbitrary which modules have the "required for boot" flag and which ones don't:

philipp at ubuntu22:~/lede$ grep -e KernelPackage -e AUTOLOAD:= package/kernel/linux/modules/netdevices.mk 
...
define KernelPackage/e1000
  AUTOLOAD:=$(call AutoLoad,35,e1000)
...
define KernelPackage/igb
  AUTOLOAD:=$(call AutoLoad,35,igb,1)
...
define KernelPackage/ixgbe
  AUTOLOAD:=$(call AutoLoad,35,ixgbe)
...
define KernelPackage/i40e
  AUTOLOAD:=$(call AutoProbe,i40e)
...
define KernelPackage/igc
  AUTOLOAD:=$(call AutoProbe,igc)
...

So, "igb" is "required for boot" (it is???) but "e1000", "ixgbe", "i40e", and "igc" aren't.

Wha.....aaat?

So name assignments can only be done for a select few devices?  Why can't renumbering also be handled for devices that get loaded later?

Can boot() in /etc/init.d/boot be modified to finish the job?  It's not as simple as duplicating the code, because you don't want to do it to the devices that have already been handled...





> On Aug 21, 2023, at 1:49 PM, Philip Prindeville <philipp_subx at redfish-solutions.com> wrote:
> 
> On a somewhat related note, I got into my /etc/board.d/02_network file and made this change:
> 
> #
> # Copyright © 2017 OpenWrt.org
> #
> 
> . /lib/functions/system.sh
> . /lib/functions/uci-defaults.sh
> 
> board_config_update
> 
> +logger -p daemon.info <http://daemon.info/> "board_name: $(board_name)"
> 
> case "$(board_name)" in
> 
> But best as I can tell that line isn't being hit... or rather, the whole file isn't.
> 
> What would prevent that from happening?  I'm not very familiar with the preinit logic.
> 
> Thanks
> 
> 
> 
>> On Aug 21, 2023, at 11:34 AM, Philip Prindeville <philipp_subx at redfish-solutions.com> wrote:
>> 
>> I have a new x86 prototype box I'm working with, and it doesn't have a DTSI file (obviously) or a platform driver... and given my experiences with getting APU platform drivers upstream last time, I'm hesitant to do that all over again.
>> 
>> Anyway, is there an easy way to do GPIO assignments for LEDs and switches if the kernel isn't doing this for me?
>> 
>> Thanks
>> 
>> 
>> _______________________________________________
>> openwrt-devel mailing list
>> openwrt-devel at lists.openwrt.org
>> https://lists.openwrt.org/mailman/listinfo/openwrt-devel
> 
> 
> _______________________________________________
> openwrt-devel mailing list
> openwrt-devel at lists.openwrt.org
> https://lists.openwrt.org/mailman/listinfo/openwrt-devel