On Tue, Jun 16, 2020 at 10:22:47AM +0530, Sandeep Maheswaram (Temp) wrote:

On 6/16/2020 1:12 AM, Matthias Kaehlcke wrote:

quoted

On Thu, Jun 04, 2020 at 04:16:31AM -0700, Stephen Boyd wrote:

quoted

Quoting Sandeep Maheswaram (Temp) (2020-06-04 02:43:09)

quoted

On 6/3/2020 11:06 PM, Stephen Boyd wrote:

quoted

Quoting Sandeep Maheswaram (2020-03-31 22:15:43)

quoted

diff --git a/drivers/usb/dwc3/dwc3-qcom.c b/drivers/usb/dwc3/dwc3-qcom.c
index 1dfd024..d33ae86 100644
--- a/drivers/usb/dwc3/dwc3-qcom.c
+++ b/drivers/usb/dwc3/dwc3-qcom.c

@@ -285,6 +307,101 @@ static int dwc3_qcom_resume(struct dwc3_qcom *qcom)
          return 0;
   }
+
+/**
+ * dwc3_qcom_interconnect_init() - Get interconnect path handles
+ * @qcom:                      Pointer to the concerned usb core.
+ *
+ */
+static int dwc3_qcom_interconnect_init(struct dwc3_qcom *qcom)
+{
+       struct device *dev = qcom->dev;
+       int ret;
+
+       if (!device_is_bound(&qcom->dwc3->dev))
+               return -EPROBE_DEFER;

How is this supposed to work? I see that this was added in an earlier
revision of this patch series but there isn't any mention of why
device_is_bound() is used here. It would be great if there was a comment
detailing why this is necessary. It sounds like maximum_speed is
important?

Furthermore, dwc3_qcom_interconnect_init() is called by
dwc3_qcom_probe() which is the function that registers the device for
qcom->dwc3->dev. If that device doesn't probe between the time it is
registered by dwc3_qcom_probe() and this function is called then we'll
fail dwc3_qcom_probe() with -EPROBE_DEFER. And that will remove the
qcom->dwc3->dev device from the platform bus because we call
of_platform_depopulate() on the error path of dwc3_qcom_probe().

So isn't this whole thing racy and can potentially lead us to a driver
probe loop where the wrapper (dwc3_qcom) and the core (dwc3) are probing
and we're trying to time it just right so that driver for dwc3 binds
before we setup interconnects? I don't know if dwc3 can communicate to
the wrapper but that would be more of a direct way to do this. Or maybe
the wrapper should try to read the DT property for maximum speed and
fallback to a worst case high bandwidth value if it can't figure it out
itself without help from dwc3 core.

This was added in V4 to address comments from Matthias in V3

https://patchwork.kernel.org/patch/11148587/

Yes, that why I said:

"I see that this was added in an earlier
  revision of this patch series but there isn't any mention of why
  device_is_bound() is used here. It would be great if there was a comment
  detailing why this is necessary. It sounds like maximum_speed is
  important?"

Can you please respond to the rest of my email?

I agree with Stephen that using device_is_bound() isn't a good option
in this case, when I suggested it I wasn't looking at the big picture
of how probing the core driver is triggered, sorry about that.

Reading the speed from the DT with usb_get_maximum_speed() as Stephen
suggests would be an option, the inconvenient is that we then
essentially require the property to be defined, while the core driver
gets a suitable value from hardware registers. Not sure if the wrapper
driver could read from the same registers.

One option could be to poll device_is_bound() for 100 ms (or so), with
sleeps between polls. It's not elegant but would probably work if we
don't find a better solution.

if (np)
        ret = dwc3_qcom_of_register_core(pdev);
    else
        ret = dwc3_qcom_acpi_register_core(pdev);

    if (ret) {
        dev_err(dev, "failed to register DWC3 Core, err=%d\n", ret);
        goto depopulate;
    }

    ret = dwc3_qcom_interconnect_init(qcom);
    if (ret)
        goto depopulate;

    qcom->mode = usb_get_dr_mode(&qcom->dwc3->dev);

Before calling dwc3_qcom_interconnect_init we are checking

    if (ret) {
        dev_err(dev, "failed to register DWC3 Core, err=%d\n", ret);
        goto depopulate;
    }

Doesn't  this condition confirm the core driver is probed?

Not really:

// called under the hood by of_platform_populate()
static int really_probe(struct device *dev, struct device_driver *drv)
{
	...

	if (dev->bus->probe) {
		ret = dev->bus->probe(dev);
		if (ret)
			goto probe_failed;
	} else if (drv->probe) {
		ret = drv->probe(dev);
	        if (ret)
	       		goto probe_failed;
        }

	...

probe_failed:
	...

	/*
         * Ignore errors returned by ->probe so that the next driver can try
         * its luck.
         */
        ret = 0;

	...

	return ret;
}

As a result of_platform_populate() in dwc3_qcom_of_register_core()
returns 0 even when probing the device failed:

[    0.244339] dwc3-qcom a6f8800.usb: DBG: populate
[    0.244772] dwc3 a600000.dwc3: DBG: dwc3_probe
[    0.245237] dwc3 a600000.dwc3: DBG: dwc3_probe err: -517
[    0.245264] dwc3-qcom a6f8800.usb: DBG: populate (done)
[    0.245317] dwc3-qcom a6f8800.usb: DBG: dwc3_qcom_interconnect_init() failed: -517

Probe fails because the interconnect stuff isn't ready yet, otherwise
it could access invalid data.

A later _populate() is successful and the probing of the core is done
synchronously, i.e. after _populate() the core driver is fully
initialized:

[    3.898106] dwc3-qcom a6f8800.usb: DBG: populate
[    3.908356] dwc3 a600000.dwc3: DBG: dwc3_probe
[    4.205104] dwc3 a600000.dwc3: DBG: dwc3_probe (done)
[    4.210305] dwc3-qcom a6f8800.usb: DBG: populate (done)

The synchronous probing in _populate() suggests that using device_is_bound()
would actually be a valid option, either the core device was successfully
probed or not, there should be no race.

I sent a patch that adds this check to dwc3_qcom_of_register_core(), which
is less confusing and makes clear that the core device is valid unless
this function returns an error:

  https://lore.kernel.org/patchwork/patch/1257279/

It might make sense to add your "driver core:Export the symbol
device_is_bound" patch, mine and this one to a single series.

Thanks

Matthias