Re: [RFC PATCH 5/5] mtd: hyperbus: Add driver for TI's Hyperbus memory controller

[Vignesh R]

On 25/02/19 9:59 PM, Sergei Shtylyov wrote:
> Hello!
> 
> On 02/19/2019 09:36 AM, Vignesh R (by way of Boris Brezillon <bbrezillon@xxxxxxxxxx>) wrote:
> 
>> Add driver for Hyperbus memory controller on TI's AM654 SoC. Programming
>> IP is pretty simple and provides direct memory mapped access to
>> connected Flash devices.
> 
>    Are you sure you posted the _complete_ driver?
> 


Yes, it is... You can find controller doc here[1]. Default values in the
MCR/MTR registers are good enough to simple Hyperflash access.
More perf optimization and timing optimizations will come incrementally

[1] http://www.ti.com/lit/ug/spruid7b/spruid7b.pdf 12.3.3 Hyperbus Interface

>> Add basic support for the IP without DMA. Second ChipSelect is not
>> supported for now.
>>
>> Signed-off-by: Vignesh R <vigneshr@xxxxxx>
>> ---
>>  drivers/mtd/hyperbus/hbmc_am654.c | 105 ++++++++++++++++++++++++++++++
>>  1 file changed, 105 insertions(+)
>>  create mode 100644 drivers/mtd/hyperbus/hbmc_am654.c
>>
>> diff --git a/drivers/mtd/hyperbus/hbmc_am654.c b/drivers/mtd/hyperbus/hbmc_am654.c
>> new file mode 100644
>> index 000000000000..1f0d2dc52f9f
>> --- /dev/null
>> +++ b/drivers/mtd/hyperbus/hbmc_am654.c
>> @@ -0,0 +1,105 @@
> [...]
>> +static int am654_hbmc_probe(struct platform_device *pdev)
>> +{
>> +	struct device *dev = &pdev->dev;
>> +	struct am654_hbmc_priv *priv;
>> +	struct resource *res;
>> +	int err;
>> +
>> +	priv = devm_kzalloc(&pdev->dev, sizeof(*priv), GFP_KERNEL);
>> +	if (!priv)
>> +		return -ENOMEM;
>> +
>> +	platform_set_drvdata(pdev, priv);
>> +
>> +	res = platform_get_resource(pdev, IORESOURCE_MEM, 0);
>> +	if (IS_ERR(res)) {
>> +		dev_err(&pdev->dev, "failed to get memory resource\n");
>> +		return -ENOENT;
>> +	}
>> +
>> +	priv->regbase = devm_ioremap_resource(dev, res);
>> +	if (IS_ERR(priv->regbase)) {
>> +		dev_err(dev, "Cannot remap controller address.\n");
>> +		return PTR_ERR(priv->regbase);
>> +	}
>> +
>> +	pm_runtime_enable(&pdev->dev);
>> +	err = pm_runtime_get_sync(&pdev->dev);
> 
>    That's all nice but where's the code that accesses the actual registers?

Interface and functional clk needs to be enabled even to access MMIO
space to read/write data to flash (done by the map framework). So driver
currently just enables everything during probe and disables on remove

> 
>> +	if (err < 0) {
>> +		pm_runtime_put_noidle(&pdev->dev);
> 
>    Calling "put" with previously failed "get" sees strange...
> 

Basically pm_runtime_get_sync() increments usage_count even in case of
failure and pm_runtime_put_noidle() puts it back. You can find many
examples of above piece of code in kernel.

>> +		return err;
>> +	}
>> +
>> +	priv->hbdev.needs_calib = true;
>> +	priv->hbdev.dev = &pdev->dev;
>> +	priv->hbdev.np = of_get_next_child(dev->of_node, NULL);
>> +	err = hb_register_device(&priv->hbdev);
>> +	if (err) {
>> +		dev_err(&pdev->dev, "failed to register controller\n");
>> +		goto err_destroy;
>> +	}
>> +
>> +	return 0;
>> +
>> +err_destroy:
>> +	hb_unregister_device(&priv->hbdev);
> 
>    Calling "unregister" with "register" previously failed also looks strange...
> 

Agree, this is unneeded as hb_register_device() takes care of all
cleanups in err path.

>> +	pm_runtime_put_sync(&pdev->dev);
> 
>    Why the sync() version?
> 

Why not? Since the device is going away, I think its safer to ensure
device has definitely been put to idle state. I see its a common
practice in driver code.

>> +	return err;
>> +}
> [...]
> 
> MBR, Sergei
> 

-- 
Regards
Vignesh