Re: [PATCH v2] i2c: imx: support DMA defer probing

-----Original Message-----
From: Marc Kleine-Budde <mkl@pengutronix.de>
Sent: Tuesday, November 26, 2024 6:24 PM
To: Carlos Song <redacted>
Cc: Frank Li <frank.li@nxp.com>; o.rempel@pengutronix.de;
kernel@pengutronix.de; andi.shyti@kernel.org; shawnguo@kernel.org;
s.hauer@pengutronix.de; festevam@gmail.com; imx@lists.linux.dev;
linux-i2c@vger.kernel.org; linux-arm-kernel@lists.infradead.org;
linux-kernel@vger.kernel.org
Subject: [EXT] Re: [PATCH v2] i2c: imx: support DMA defer probing

On 26.11.2024 10:15:27, Carlos Song wrote:

quoted

quoted

quoted

 static void i2c_imx_dma_callback(void *arg) @@ -1803,6 +1804,13

@@ static int i2c_imx_probe(struct platform_device *pdev)
 	if (ret == -EPROBE_DEFER)
 		goto clk_notifier_unregister;

+	/* Init DMA config if supported */
+	ret = i2c_imx_dma_request(i2c_imx, phy_addr);
+	if (ret == -EPROBE_DEFER) {
+		dev_err(&pdev->dev, "DMA not ready, go defer probe!\n");
+		goto clk_notifier_unregister;
+	}

Don't spam the logs if the driver defers probing, it's not a error.
And it looks strange to ignore all other errors here. Either add a
comment here, something like "continue without DMA", or let the
function return
0 in case the driver should continue and propagate the error if the
caller should take care of it.

Hi,
Thank you for your suggestion! I agree with you.
I will change to this logic:
	ret = i2c_imx_dma_request(i2c_imx, phy_addr);
	if (ret) {
		if (ret == -EPROBE_DEFER)
			goto clk_notifier_unregister;
		dev_info(&pdev->dev, "use pio mode\n");
	}

Ret = 0  ----->  enable DMA successfully  -------> no print
Ret!=0  ----->  defer probe ---------> no print and try again
Ret!=0  ----->  fail to enable DMA ------> remind now is using pio
mode

Do you think the logic is acceptable?

Yes, the other option is to move the logic into i2c_imx_dma_request() and let it
return 0 in case of DMA or fallback to PIO, or an error in case of probe defer or a
fatal error.

This way the probe function will look like this:

     ret = i2c_imx_dma_request(i2c_imx, phy_addr);
     if (ret)
                return dev_err_probe(&pdev->dev, ret, "Failed to setup
DMA\n");

Sorry, I have some different ideas...
1. DMA mode should be optional for i2c-imx, because i2c-imx can accept DMA mode not enabled, because it still can work in CPU mode.
  If we use return dev_err_probe(), we have to return error at i2c_imx_dma_request() for "some fatal error", it will cause i2c_adapter can not be registered, then kill i2c adapter register.
  If we always return 0 at i2c_imx_dma_request(), dev_err_probe will not work forever. So from my point, if DMA is not working well, just output a log to remind now i2c is always
  working at CPU mode, we have no DMA, this is enough.
2. when really defer probe, return dev_err_probe will return defer probe directly, but we still need to goto clk_notifier_unregister branch to free irq, clk_notifier_unregister and disable runtime pm.
  So we still need more judgement at probe function to handle this.

So I prefer this logic:

ret = i2c_imx_dma_request(i2c_imx, phy_addr);
if (ret) {
	if (ret == -EPROBE_DEFER)
		goto clk_notifier_unregister;
	dev_info(&pdev->dev, "use pio mode\n");
}

regards,
Marc

--
Pengutronix e.K.                 | Marc Kleine-Budde          |
Embedded Linux                   | https://www.pengutronix.de |
Vertretung Nürnberg              | Phone: +49-5121-206917-129 |
Amtsgericht Hildesheim, HRA 2686 | Fax:   +49-5121-206917-9   |