On Thu, Aug 26, 2021 at 4:29 AM Alvin Šipraga [off-list ref] wrote:

Hi Saravana,

 From looking at the code, the Marvell DSA driver mv88e6xxx may also
suffer from the same problem if fw_devlink=on. Maybe somebody (Andrew?)
could test so that you know whether include a simlar patch to that
driver in your series.

Other drivers may be effected too - as Andrew said in the other thread,
this is not an uncommon pattern for DSA drivers.

On 8/26/21 9:45 AM, Saravana Kannan wrote:

quoted

This is just a quick fix to make this driver work with fw_devlink=on.
The proper fix might need a significant amount of rework of the driver
of the framework to use a component device model.

Signed-off-by: Saravana Kannan <redacted>

With the caveat that it's a test with my RFC rtl8365mb subdriver...

Tested-by: Alvin Šipraga <alsi@bang-olufsen.dk>

Thanks for testing. And just to be sure we are all on the same page:
Without this patch the PHYs get handled by the Generic PHY driver.
With this patch, the PHYs are handled by their specific driver.
Correct?

-Saravana

Kind regards,
Alvin

quoted

---
  drivers/net/dsa/realtek-smi-core.c | 7 +++++++
  1 file changed, 7 insertions(+)

diff --git a/drivers/net/dsa/realtek-smi-core.c b/drivers/net/dsa/realtek-smi-core.c
index 8e49d4f85d48..f79c174f4954 100644
--- a/drivers/net/dsa/realtek-smi-core.c
+++ b/drivers/net/dsa/realtek-smi-core.c

@@ -394,6 +394,13 @@ static int realtek_smi_probe(struct platform_device *pdev)
      var = of_device_get_match_data(dev);
      np = dev->of_node;

+     /* This driver assumes the child PHYs would be probed successfully
+      * before this functions returns. That's not a valid assumption, but
+      * let fw_devlink know so that this driver continues to function with
+      * fw_devlink=on.
+      */
+     np->fwnode.flags |= FWNODE_FLAG_BROKEN_PARENT;
+
      smi = devm_kzalloc(dev, sizeof(*smi) + var->chip_data_sz, GFP_KERNEL);
      if (!smi)
              return -ENOMEM;