[PATCH rtems-libbsd] rtemsbsd: Use a separate header for test devices

[Chris Johns]

On 7/7/21 11:26 pm, Kinsey Moore wrote:
> On 7/6/2021 21:20, Chris Johns wrote:
>> On 7/7/21 12:03 pm, Kinsey Moore wrote:
>>> On 7/6/2021 20:57, Chris Johns wrote:
>>>> On 7/7/21 11:05 am, Kinsey Moore wrote:
>>>>> The need for the difference on ZynqMP is that there are 4 different CGEM
>>>>> interfaces of which dev boards primarily make use of CGEM3.
>>>> RTEMS_BSD_DRIVER_XILINX_ZYNQMP_CGEM0(ZYNQMP_IRQ_ETHERNET_0);
>>>> RTEMS_BSD_DRIVER_XILINX_ZYNQMP_CGEM1(ZYNQMP_IRQ_ETHERNET_1);
>>>> RTEMS_BSD_DRIVER_XILINX_ZYNQMP_CGEM2(ZYNQMP_IRQ_ETHERNET_2);
>>>> RTEMS_BSD_DRIVER_XILINX_ZYNQMP_CGEM3(ZYNQMP_IRQ_ETHERNET_3);
>>>>
>>>> ?
>>> Yes, this does technically work
>> Hmm, I suggest this is what we should support as a default.
>>
>>> if you can read the shell output past the log
>>> spam. The other interfaces trying and failing to come up throw a gargantuan
>>> amount of messages to the console.
>> Why do these interfaces fail to initialise? What are the errors?
>>
>> The removed probe check is based around FDT and so I suspect is the reason
>> Sebastian suggested FDT support. Needing FDT support would break existing Zynq
>> users.
> 
> The devices themselves are detected just fine and are likely operational. The
> MII busses are probably unterminated or pulled high/low which yields a ukphy
> detection on every available slot and constant PHY read/write timeouts. A small
> sample from enabling both CGEM2 and CGEM3 on this custom board:
> 
> ukphy13:  none, 10baseT, 10baseT-FDX, 100baseTX, 100baseTX-FDX, 100baseT4,
> 1000baseSX, 1000baseSX-FDX, 1000baseT, 1000baseT-master, 1000baseT-FDX,
> 1000baseT-FDX-master, auto
> cgem3: phy read timeout: 0
> ukphy14: <Generic IEEE 802.3u media interface> PHY 14 on miibus0
> cgem3: phy write timeout: 0
> cgem3: phy read timeout: 0
> cgem3: phy read timeout: 0
> cgem3: phy read timeout: 0
> cgem3: phy read timeout: 0
> cgem3: phy read timeout: 0
> cgem3: phy read timeout: 0
> 
> ...
> 
> cgem3: phy read timeout: 0

That is noisy and painful.

> This type of output reached ukphy31 before I was able to run the shutdown
> command with 50+ PHY read/write errors and more continuing afterward. I imagine
> it's worse if I enable all 4 interfaces.

The output is from the cgem driver which is good because it is localised. I
agree we need something to control what is enabled and I prefer we avoid special
build options, I will explain why I think this later. We can:

1. Add FDT support. This is something I would like to avoid as it adds an extra
layer of dependency and it complicates backwards compatibility for existing
users. I however wonder if this is just me and it is something that will be
needed more and more and cannot be avoided.

2. Add a weak call that is RTEMS and BSP specific to return a probe state. The
default state could be enabled and the tests provide something from the network
config. Not a great solution but it is simple and cheap to implement. The driver
already has a weak call to set the reference clock.

3. Try and detect a PHY in the probe? I am not sure if that is easy or hard. If
read works maybe that is something that may be suitable.


I would prefer we avoid special build options for BSPs. It is easy for us as
developers to make these changes and build a specific RTEMS plus we can be a
little narrow in our focus when delivering something for a client in the
defaults we select. Specific builds of a BSP to configure options becomes hard
for users where they have a few Zynq or ZynqMP designs all running RTEMS and
each with a different mix of network interfaces. This creates a complex set of
build options for common application code plus more configuration items to control.

Chris