I have seemingly done as you have suggested but the result remains the same. I do not even get any error messages to suggest I have modified anything incorrectly which again leads me to believe that maybe its an issue with how I have set up the folder but I don't see how as it follows the same conventions to the other plat from source files in libethdrivers. My second guess is simply that despite adding the parts for bootstrapping, it is making no attempt to do so. Again, apologies if my issues are trivial as this is my first time working with anything like this.

Best, Nkem

On Mon, 25 Jan 2021 at 03:27, Lee, Damon (Data61, Eveleigh) < Damon.Lee@data61.csiro.au> wrote:

Hi Nkem,

...

I have been trying to use the ethernet driver for the UNSW Asdvanced Operating Systems course but I still get these messages:

find_compatible_driver_module@io.c:434 No suitable driver was found for

path

...

/soc/ethernet@c9410000, ignoring server_init@ethdriver.c:382 Unable to find an ethernet device

I thought that I had to copy the driver to the utils lib folder and

provide a

...

preprocessor macro as someone else had suggested but I may be missing a

few

...

other things.

It seems that you're pretty close to getting everything complete. You're only missing one more step, and this is to add some bootstrapping code to get the driver to initialise and register itself with the interface registration service so that it can be found by the Ethdriver CAmkES component.

To do this, you can copy the example from the i.MX6 Ethernet driver and change it up a bit so that it works for the Odroid C2. More specifically, you'd copy this function [1], modify the contents of the function to use the correct initialisation function for the C2, and also modify the callbacks [2] and [3] so that it gets the correct registers and IRQs from the kernel's version of the Odroid C2's DTS. Finally, you also need copy this [4] and modify the following string array and the last argument in the preprocessor symbol to point to the function you copied earlier.

I hope this helps, Damon

[1]

https://github.com/seL4/util_libs/blob/master/libethdrivers/src/plat/imx6/im... [2]

https://github.com/seL4/util_libs/blob/master/libethdrivers/src/plat/imx6/im... [3]

https://github.com/seL4/util_libs/blob/master/libethdrivers/src/plat/imx6/im... [4]

https://github.com/seL4/util_libs/blob/master/libethdrivers/src/plat/imx6/im... _______________________________________________ Devel mailing list -- devel@sel4.systems To unsubscribe send an email to devel-leave@sel4.systems

Hi Nkem,

Just to clarify, a modification I made was calling the ethif_init function

ethernet.c file[1] and passing some parameters (base_addr,mac,

dma_ops,ethif_recv_callback). Currently, dma_ops and ethif_recv_callback

contain null values and I'm not too sure what I need to initialise them as.

The issue is more clear now, the issue lies with the fact that the Ethernet

driver structure layout is different from the expected layout in libethdrivers.

For the `dma_ops` case, this shouldn't be NULL, and depending on what CAmkES

application (e.g. picoserver) you are using, this will be handled automatically

for you. Otherwise, you can call `camkes_io_ops` and then you can access the

underlying `dma_ops` structure inside the `ps_io_ops_t` pointer.

As for the `ethif_recv_callback` this shouldn't be there, the libethdrivers

drivers instead require these functions to be filled in [1]. What I'm trying to

say here is that the model of the AOS Ethernet driver is different from those

expected in libethdrivers, and you would need to modify the AOS Ethernet driver

quite heavily so that it matches the model in libethdrivers.

Also, please could you clarify how I am meant to modify the callbacks so

that it gets the correct registers and IRQs from the DTS.

The callbacks are called like a lambda in a 'map' function like in other

languages. It's given the current IRQ or register information each time it's

called and the user can choose to take that information and use it to allocate

hardware resources. So in the case of the C2, looking at the kernel DTS for it

(kernel/tools/dts/odroidc2.dts), the Ethernet node (/soc/ethernet@c9410000)

contains one interrupt entry and two register entries. The interrupt entry is

simple, you could simple call `ps_irq_register` on the IRQ information in the

first invocation. As for the registers, you would need to call `ps_mem_map`

twice, once for each time the callback is called.

If aren't in a rush to solve these issues, the team is considering porting over

the AOS Ethernet driver for the C2 over to libethdrivers in the next few days.

Though this would take a few days to do, review, and have it pushed out to

upstream.

Hope this helps,

Damon

[1] https://github.com/seL4/util_libs/blob/master/libethdrivers/src/plat/imx6/im...

Sorry I accidentally sent the email without finishing it. The camkes file was set up as such:

import <std_connector.camkes>; import <global-connectors.camkes>; import <Ethdriver/Ethdriver.camkes>;

component Client{ uses Ethdriver eth_drv; control; }

assembly { composition { component Client client_1; component Client client_2;

component EthdriverARMPlat ethdriver; connection seL4Ethdriver eth_driver_conn1(from client_1.eth_drv, to ethdriver.client); connection seL4Ethdriver eth_driver_conn2(from client_2.eth_drv, to ethdriver.client); }

configuration { ethdriver.simple = true; ethdriver.cnode_size_bits = 12; ethdriver.simple_untyped20_pool = 2; ethdriver.heap_size = 0x10000; ethdriver.dma_pool = 0x200000; } }

and when run on the board I get an output of:

Hello World! Assertion failed: !"sys_socket not implemented" (/home/nkem/Test/projects/camkes-tool/libsel4camkes/src/sys_socket.c: camkes_sys_socket: 42) dwmac.687000 Waiting for PHY auto negotiation to complete.......... done Speed: 1000, full duplex Assertion failed: buf (/home/nkem/Test/projects/global-components/components/Ethdriver/src/ethdriver.c: server_init: 388)

Is this due to how i have set up the connections or something else, I'm not 100 percent sure on how to structure a camkes file.

Thanks, Nkem

On Wed, 24 Feb 2021 at 08:50, Nkem Ogosi ogosinkem@gmail.com wrote:

Thanks Damon,

I see that the driver has since been added to util_libs and I was just wondering about some errors I am getting when trying to use it. I made a quick hello world app and set up the camkes file as such:

Hello World! Assertion failed: !"sys_socket not implemented" (/home/nkem/Test/projects/camkes-tool/libsel4camkes/src/sys_socket.c: camkes_sys_socket: 42) dwmac.687000 Waiting for PHY auto negotiation to complete.......... done Speed: 1000, full duplex Assertion failed: buf (/home/nkem/Test/projects/global-components/components/Ethdriver/src/ethdriver.c: server_init: 388)

On Thu, 28 Jan 2021 at 01:19, Lee, Damon (Data61, Eveleigh) < Damon.Lee@data61.csiro.au> wrote:

...

Hi Nkem,

...

Just to clarify, a modification I made was calling the ethif_init function

...

ethernet.c file[1] and passing some parameters (base_addr,mac,

...

dma_ops,ethif_recv_callback). Currently, dma_ops and ethif_recv_callback

...

contain null values and I'm not too sure what I need to initialise them as.

The issue is more clear now, the issue lies with the fact that the Ethernet

driver structure layout is different from the expected layout in libethdrivers.

For the `dma_ops` case, this shouldn't be NULL, and depending on what CAmkES

application (e.g. picoserver) you are using, this will be handled automatically

for you. Otherwise, you can call `camkes_io_ops` and then you can access the

underlying `dma_ops` structure inside the `ps_io_ops_t` pointer.

As for the `ethif_recv_callback` this shouldn't be there, the libethdrivers

drivers instead require these functions to be filled in [1]. What I'm trying to

say here is that the model of the AOS Ethernet driver is different from those

expected in libethdrivers, and you would need to modify the AOS Ethernet driver

quite heavily so that it matches the model in libethdrivers.

...

Also, please could you clarify how I am meant to modify the callbacks so

...

that it gets the correct registers and IRQs from the DTS.

The callbacks are called like a lambda in a 'map' function like in other

languages. It's given the current IRQ or register information each time it's

called and the user can choose to take that information and use it to allocate

hardware resources. So in the case of the C2, looking at the kernel DTS for it

(kernel/tools/dts/odroidc2.dts), the Ethernet node (/soc/ethernet@c9410000)

contains one interrupt entry and two register entries. The interrupt entry is

simple, you could simple call `ps_irq_register` on the IRQ information in the

first invocation. As for the registers, you would need to call `ps_mem_map`

twice, once for each time the callback is called.

If aren't in a rush to solve these issues, the team is considering porting over

the AOS Ethernet driver for the C2 over to libethdrivers in the next few days.

Though this would take a few days to do, review, and have it pushed out to

upstream.

Hope this helps,

Damon

[1] https://github.com/seL4/util_libs/blob/master/libethdrivers/src/plat/imx6/im...