[PATCH 04/13] arm: mach-k3: j721s2_init: Probe AM65 CPSW NUSS for R5/A72 SPL

[Chintan Vankar]

On 23/01/25 15:10, Roger Quadros wrote:
> +Vignesh & Siddharth
> 
> On 22/01/2025 10:53, Chintan Vankar wrote:
>>
>>
>> On 07/01/25 20:03, Roger Quadros wrote:
>>>
>>>
>>> On 07/01/2025 11:38, Chintan Vankar wrote:
>>>> To support Ethernet boot on AM68-SK, probe AM65 CPSW NUSS driver in
>>>> board_init_f().
>>>>
>>>> Signed-off-by: Chintan Vankar <c-vankar at ti.com>
>>>> ---
>>>>    arch/arm/mach-k3/j721s2/j721s2_init.c | 10 ++++++++++
>>>>    1 file changed, 10 insertions(+)
>>>>
>>>> diff --git a/arch/arm/mach-k3/j721s2/j721s2_init.c b/arch/arm/mach-k3/j721s2/j721s2_init.c
>>>> index 6ce3eb87efb..7208bee5785 100644
>>>> --- a/arch/arm/mach-k3/j721s2/j721s2_init.c
>>>> +++ b/arch/arm/mach-k3/j721s2/j721s2_init.c
>>>> @@ -329,6 +329,16 @@ void board_init_f(ulong dummy)
>>>>          setup_qos();
>>>>    +    if (IS_ENABLED(CONFIG_SPL_ETH) && IS_ENABLED(CONFIG_TI_AM65_CPSW_NUSS) &&
>>>> +        spl_boot_device() == BOOT_DEVICE_ETHERNET) {
>>>> +        struct udevice *cpswdev;
>>>> +
>>>> +        ret = uclass_get_device_by_driver(UCLASS_MISC, DM_DRIVER_GET(am65_cpsw_nuss),
>>>> +                          &cpswdev);
>>>> +        if (ret)
>>>> +            printf("Failed to probe am65_cpsw_nuss driver..\n");
>>>> +    }
>>>> +
>>>
>>> This looks like a hack. Please find out why the Ethernet driver is not being
>>> probed when SPL tries to load image over net.
>>>
>>
>> Hello Roger,
>>
>> The driver is defined as UCLASS_MISC which should be probed explicitly,
> 
> The driver was originally built without UCLASS_MISC, but
> 
> commit 38922b1f4acc ("net: ti: am65-cpsw: Add support for multi port independent MAC mode")
> 
> Added UCLASS_MISC and states there that
> " Since top level driver is now UCLASS_MISC, board files would need to
>    instantiate this driver explicitly."
> 
> Not an elegant solution. So we need to fix something in the am65-cpuss driver.
> 
> Looking at drivers/net/mvpp2.c we can see a possible solution.
> 1) don't define .probe for the parent driver (UCLASS_MISC). Instead define .bind
> that will scan the device tree for ports and bind the port device and driver.
> e.g. see mvpp2_base_bind()
> 2) in port driver .probe (UCLASS_ETH), if parent has not been probed
> then call the parent probe (am65_cpsw_probe_nuss) . Also set a flag so parent probe
> only gets called once.
> 3) update all board files no not explicitly probe the am65-cpsw UCLASS_MISC driver.
> 
> Siddharth / Vignesh do you see any issues with this solution?
> 

Hello Roger, I have tried to understand your approach, I will work on
that. But for now, is it possible to merge the seris with the existing
approach ?

>> I have discussed the same with Nishanth in following thread:
>> https://lore.kernel.org/all/ee1d16fd-b99b-4b07-97bb-a896e179157a@ti.com/
>>
>>> We have the following defined at the am65-cpsw-nuss driver.
>>>
>>> U_BOOT_DRIVER(am65_cpsw_nuss) = {
>>>           .name   = "am65_cpsw_nuss",
>>>           .id     = UCLASS_MISC,
>>>           .of_match = am65_cpsw_nuss_ids,
>>>           .probe  = am65_cpsw_probe_nuss,
>>>           .priv_auto = sizeof(struct am65_cpsw_common),
>>> };
>>>
>>> U_BOOT_DRIVER(am65_cpsw_nuss_port) = {
>>>           .name   = "am65_cpsw_nuss_port",
>>>           .id     = UCLASS_ETH,
>>>           .probe  = am65_cpsw_port_probe,
>>>           .ops    = &am65_cpsw_ops,
>>>           .priv_auto      = sizeof(struct am65_cpsw_priv),
>>>           .plat_auto      = sizeof(struct eth_pdata),
>>>           .flags = DM_FLAG_ALLOC_PRIV_DMA | DM_FLAG_OS_PREPARE,
>>> };
>>>
>>> It looks like am65_cpsw_probe_nuss() is not being invoked for you.
>>>
>>> Can you please check if am65_cpsw_port_probe() was invoked?
>>> If yes but am65_cpsw_probe_nuss() was not then we need to fix the
>>> DM hierarchy for AM65_CPSW?
>>>
>> None of the probe function is getting invoked here, since we need
> 
> That is strange. am65_cpsw_probe_nuss() should be called at least for
> port 0 since it is defined as UCLASS_ETH.
> 

am65_cpsw_probe_nuss() is defined as UCLASS_MISC and not UCLASS_ETH, it
will get invoked if declared as UCLASS_ETH, as per your suggestion above
it will need driver changes to invoke the probe function without being
called explicitly and work properly.

>> Ethernet functionality here we need to probe function in board_init_f().
>> We have discussed the same at here:
>> https://lore.kernel.org/r/d68c8045-116a-49c0-9e74-d6a366e6f008@kernel.org/#t
>>
>>
>>>>        if (IS_ENABLED(CONFIG_CPU_V7R) && IS_ENABLED(CONFIG_K3_AVS0)) {
>>>>            ret = uclass_get_device_by_driver(UCLASS_MISC, DM_DRIVER_GET(k3_avs),
>>>>                              &dev);
>>>
>