Re: [PATCH v2 06/11] mfd: qcom-smd-rpm: Driver for the Qualcomm RPM over SMD

[Bjorn Andersson <bjorn.andersson@sonymobile.com>]

On Thu 23 Jul 06:22 PDT 2015, Lee Jones wrote:

> On Mon, 13 Jul 2015, Bjorn Andersson wrote:
> 
> > On Tue 07 Jul 05:37 PDT 2015, Lee Jones wrote:
> > 
> > > On Fri, 26 Jun 2015, bjorn@kryo.se wrote:
> > > 
> > > > From: Bjorn Andersson <bjorn.andersson@sonymobile.com>
> > [..]
> > 
> > > > diff --git a/drivers/mfd/Kconfig b/drivers/mfd/Kconfig
> > 
> > [..]
> > 
> > > > +config MFD_QCOM_SMD_RPM
> > > > +	tristate "Qualcomm Resource Power Manager (RPM) over SMD"
> > > > +	depends on QCOM_SMD && OF
> > > > +	help
> > > > +	  If you say yes to this option, support will be included for the
> > > > +	  Resource Power Manager system found in the Qualcomm 8974 based
> > > > +	  devices.
> > > > +
> > > > +	  This is required to access many regulators, clocks and bus
> > > > +	  frequencies controlled by the RPM on these devices.
> > > > +
> > > > +	  Say M here if you want to include support for the Qualcomm RPM as a
> > > > +	  module. This will build a module called "qcom-smd-rpm".
> > > 
> > > I'm not exactly sure what makes this an MFD device.
> > > 
> > 
> > It represents a piece of hardware (a micro-controller) that exposes
> > control of a multitude of regulators and clocks in the Qualcomm
> > platforms.
> > 
> > It's basically just a successor of the qcom_rpm driver - same
> > functionality but a new communication method is used.
> 
> My point still stands.  Please investigate moving this (and the
> qcom_rpm driver if it's the same) into either drivers/soc or
> drivers/platform.  The support in these two directories _seem_ to be
> pretty similar.
> 

We had this exact discussion last year and I argued that a piece of
hardware that exposes regulators and clocks - like most PMICs - is a
mfd and you agreed and picked the driver.

I will have a word with Andy about moving this and the qcom_rpm driver
out of mfd.

> > > > diff --git a/drivers/mfd/qcom-smd-rpm.c b/drivers/mfd/qcom-smd-rpm.c
> > 
> > [..]
> > 
> > > > +
> > > > +#define RPM_ERR_INVALID_RESOURCE "resource does not exist"
> > > 
> > > I don't like this at all.
> > > 
> > 
> > Which part of it?
> > 
> > It should probably be a static const char *, inlined in the function
> > below. Would that be to your liking?
> 
> It would be better, but I never really see the point in initialising
> variables with these types of messages.  I'd get rid of the
> superfluous chuff and just do:
> 
>   memcmp(msg->message, "resource does not exist", 23);
> 

The point was simply to not have to write:

  if (msg->length == 23 && memcmp(msg->message, ..., 23);

Simply because I don't like the first part of the expression. I'll
rewrite it...

> > > > +static int qcom_smd_rpm_callback(struct qcom_smd_device *qsdev,
> > > > +				 const void *data,
> > > > +				 size_t count)
> > > > +{
> > > > +	const struct qcom_rpm_header *hdr = data;
> > > > +	const struct qcom_rpm_message *msg;
> > > > +	const size_t inv_res_len = sizeof(RPM_ERR_INVALID_RESOURCE) - 1;
> > > > +	struct qcom_smd_rpm *rpm = dev_get_drvdata(&qsdev->dev);
> > > > +	const u8 *buf = data + sizeof(struct qcom_rpm_header);
> > > > +	const u8 *end = buf + hdr->length;
> > > > +	int status = 0;
> > > > +
> > > > +	if (hdr->service_type != RPM_SERVICE_TYPE_REQUEST ||
> > > > +	    hdr->length < sizeof(struct qcom_rpm_message)) {
> > > > +		dev_err(&qsdev->dev, "invalid request\n");
> > > > +		return 0;
> > > > +	}
> > > > +
> > > > +	while (buf < end) {
> > > > +		msg = (struct qcom_rpm_message *)buf;
> > > > +		switch (msg->msg_type) {
> > > > +		case RPM_MSG_TYPE_MSG_ID:
> > > > +			break;
> > > > +		case RPM_MSG_TYPE_ERR:
> > > > +			if (msg->length == inv_res_len &&
> > > > +			    !memcmp(msg->message,
> > > > +				    RPM_ERR_INVALID_RESOURCE,
> > > > +				    inv_res_len))
> > > 
> > > strncpy(msg->message, "resource does not exist", 23);
> > > 
> > 
> > No, I want to compare the content of msg->message with the string
> 
> Yes, I just noticed that.
> 
> > "resource does not exist" - as that's the only way to know what type of
> > error we got.
> > 
> > This is unfortunately how the protocol looks :/
> 
> What about either my memcmp suggestion above or this then:
> 
>   strncmp(msg->message, "resource does not exist", 23);
> 

That would require the string to be 0-terminated.

[..]

> > > > +static struct qcom_smd_driver qcom_smd_rpm_driver = {
> > > > +	.probe = qcom_smd_rpm_probe,
> > > > +	.remove = qcom_smd_rpm_remove,
> > > > +	.callback = qcom_smd_rpm_callback,
> > > > +	.driver  = {
> > > > +		.name  = "qcom_smd_rpm",
> > > > +		.owner = THIS_MODULE,
> > > 
> > > Remove this line.
> 
> Still not 100% sure why you need your own 'special' driver struct.  If
> it's for the .callback, there are other ways to do this without having
> to invent your own bus.
> 

Because the life cycle of these components are much like, say, USB -
they can come and go. As such e.g. a platform_driver is not a good fit.

Regards,
Bjorn