On 2024-12-27 01:21, Andrew Lunn wrote:
On Thu, Dec 26, 2024 at 10:29:45AM +0800, Yijie Yang wrote:
On 2024-12-25 19:37, Krzysztof Kozlowski wrote:
On 25/12/2024 11:04, Yijie Yang wrote:
static int qcom_ethqos_probe(struct platform_device *pdev)
{
- struct device_node *np = pdev->dev.of_node;
+ struct device_node *np = pdev->dev.of_node, *root;
const struct ethqos_emac_driver_data *data;
struct plat_stmmacenet_data *plat_dat;
struct stmmac_resources stmmac_res;
@@ -810,6 +805,15 @@ static int qcom_ethqos_probe(struct platform_device *pdev)
ret = of_get_phy_mode(np, ðqos->phy_mode);
if (ret)
return dev_err_probe(dev, ret, "Failed to get phy mode\n");
+
+ root = of_find_node_by_path("/");
+ if (root && of_device_is_compatible(root, "qcom,sa8540p-ride"))
Nope, your drivers are not supposed to poke root compatibles. Drop and
fix your driver to behave correctly for all existing devices.
Since this change introduces a new flag in the DTS, we must maintain ABI
compatibility with the kernel. The new flag is specific to the board, so I
need to ensure root nodes are matched to allow older boards to continue
functioning as before. I'm happy to adopt that approach if there are any
more elegant solutions.
Why is it specific to this board? Does the board have a PHY which is
broken and requires this property? What we are missing are the details
needed to help you get to the correct way to solve the problem you are
facing.
Let me clarify why this bit is necessary and why it's board-specific.
The RX programming swap bit can introduce a time delay of half a clock
cycle. This bit, along with the clock delay adjustment functionality, is
implemented by a module called 'IO Macro.' This is a Qualcomm-specific
hardware design located between the MAC and PHY in the SoC, serving the
RGMII interface. The bit works in conjunction with delay adjustment to
meet the sampling requirements. The sampling of RX data is also handled
by this module.
During the board design stage, the RGMII requirements may not have been
strictly followed, leading to uncertainty in the relationship between
the clock and data waveforms when they reach the IO Macro. This means
the time delay introduced by the PC board may not be zero. Therefore,
it's necessary for software developers to tune both the RX programming
swap bit and the delay to ensure correct sampling.
Andrew
--
Best Regards,
Yijie
