Re: [External] Re: [PATCH v6 3/8] usb: mausb_host: HCD initialization

[Vladimir Stankovic <vladimir.stankovic@displaylink.com>]

On 15.5.20. 15:07, Greg KH wrote:
> On Fri, May 15, 2020 at 02:34:57PM +0200, Vladimir Stankovic wrote:
>> --- /dev/null
>> +++ b/drivers/usb/host/mausb/hcd.c
>> @@ -0,0 +1,120 @@
>> +// SPDX-License-Identifier: GPL-2.0
>> +/*
>> + * Copyright (c) 2019 - 2020 DisplayLink (UK) Ltd.
>> + */
>> +#include "hcd.h"
>> +
>> +#include "utils.h"
>> +
>> +static int mausb_bus_match(struct device *dev, struct device_driver *drv);
>> +
>> +static const struct file_operations mausb_fops;
>> +
>> +static unsigned int major;
>> +static unsigned int minor = 1;
>> +static dev_t devt;
>> +static struct device *device;
>> +
>> +struct mausb_hcd	*mhcd;
>> +static struct class	*mausb_class;
>> +static struct bus_type	mausb_bus_type = {
>> +	.name	= DEVICE_NAME,
>> +	.match	= mausb_bus_match,
>> +};
> 
> A static bus type???  For a single driver?
> 
>> +
>> +static struct device_driver mausb_driver = {
>> +	.name	= DEVICE_NAME,
>> +	.bus	= &mausb_bus_type,
>> +	.owner	= THIS_MODULE,
>> +};
> 
> Wait, what???  A static "raw" struct device_driver?  Why???
This was our initial driver setup that was "inherited" from some
in-tree drivers. We are currently revising driver setup. In general,
device driver will not work properly w/o bus being setup and the
only way to avoid explicit bus and simplify driver setup is to use
platform driver; however, we are not aware of any explicit dependency
on the platform, so not sure whether it's acceptable to switch to
platform device driver setup.
> 
>> +
>> +static void mausb_remove(void)
>> +{
>> +	struct usb_hcd *hcd, *shared_hcd;
>> +
>> +	hcd	   = mhcd->hcd_hs_ctx->hcd;
>> +	shared_hcd = mhcd->hcd_ss_ctx->hcd;
>> +
>> +	if (shared_hcd) {
>> +		usb_remove_hcd(shared_hcd);
>> +		usb_put_hcd(shared_hcd);
>> +		mhcd->hcd_ss_ctx = NULL;
>> +	}
>> +
>> +	usb_remove_hcd(hcd);
>> +	usb_put_hcd(hcd);
>> +	mhcd->hcd_hs_ctx = NULL;
>> +}
>> +
>> +static int mausb_bus_match(struct device *dev, struct device_driver *drv)
>> +{
>> +	if (strncmp(dev->bus->name, drv->name, strlen(drv->name)))
>> +		return 0;
>> +	else
>> +		return 1;
>> +}
>> +
>> +int mausb_init_hcd(void)
>> +{
>> +	int retval = register_chrdev(0, DEVICE_NAME, &mausb_fops);
>> +
>> +	if (retval < 0)
>> +		return retval;
>> +
>> +	major = (unsigned int)retval;
>> +	retval = bus_register(&mausb_bus_type);
>> +	if (retval)
>> +		goto bus_register_error;
>> +
>> +	mausb_class = class_create(THIS_MODULE, CLASS_NAME);
>> +	if (IS_ERR(mausb_class)) {
>> +		retval = (int)PTR_ERR(mausb_class);
>> +		goto class_error;
>> +	}
>> +
>> +	retval = driver_register(&mausb_driver);
>> +	if (retval)
>> +		goto driver_register_error;
>> +
>> +	mhcd = kzalloc(sizeof(*mhcd), GFP_ATOMIC);
>> +	if (!mhcd) {
>> +		retval = -ENOMEM;
>> +		goto mausb_hcd_alloc_failed;
>> +	}
>> +
>> +	devt = MKDEV(major, minor);
>> +	device = device_create(mausb_class, NULL, devt, mhcd, DEVICE_NAME);
>> +	if (IS_ERR(device)) {
>> +		retval = (int)IS_ERR(device);
>> +		goto device_create_error;
>> +	}
>> +
>> +	device->driver = &mausb_driver;
> 
> Why?  What is this device going to do?  What do you need it for?
Yeah, this one is not needed at all. Will be removed.
> 
>> +
>> +	return retval;
>> +device_create_error:
>> +	kfree(mhcd);
>> +	mhcd = NULL;
>> +mausb_hcd_alloc_failed:
>> +	driver_unregister(&mausb_driver);
>> +driver_register_error:
>> +	class_destroy(mausb_class);
>> +class_error:
>> +	bus_unregister(&mausb_bus_type);
>> +bus_register_error:
>> +	unregister_chrdev(major, DEVICE_NAME);
> 
> I thought you were using the misc device, what are you doing with a
> class and chardev?  Why is that still here?
> 
We're currently revising driver setup logic - this part should be much
cleaner in new patch version.
> 
> 
> 
>> +
>> +	return retval;
>> +}
>> +
>> +void mausb_deinit_hcd(void)
>> +{
>> +	if (mhcd) {
>> +		mausb_remove();
>> +		device_destroy(mausb_class, devt);
>> +		driver_unregister(&mausb_driver);
>> +		class_destroy(mausb_class);
>> +		bus_unregister(&mausb_bus_type);
>> +		unregister_chrdev(major, DEVICE_NAME);
>> +	}
>> +}
>> diff --git a/drivers/usb/host/mausb/hcd.h b/drivers/usb/host/mausb/hcd.h
>> new file mode 100644
>> index 000000000000..e2374af67663
>> --- /dev/null
>> +++ b/drivers/usb/host/mausb/hcd.h
>> @@ -0,0 +1,65 @@
>> +/* SPDX-License-Identifier: GPL-2.0 */
>> +/*
>> + * Copyright (c) 2019 - 2020 DisplayLink (UK) Ltd.
>> + */
>> +#ifndef __MAUSB_HCD_H__
>> +#define __MAUSB_HCD_H__
>> +
>> +#include <linux/slab.h>
>> +#include <linux/usb.h>
>> +#include <linux/usb/hcd.h>
>> +
>> +#define DEVICE_NAME "mausb_host_hcd"
>> +#define CLASS_NAME "mausb"
>> +
>> +#define NUMBER_OF_PORTS		15
> 
> Why this max?
Related to USB_SS_MAXPORTS - a worst case scenario.
> 
>> +
>> +#define MAX_USB_DEVICE_DEPTH	6
> 
> Where does this max come from?
Unknown origin and unused - will be removed.
> 
>> +
>> +#define RESPONSE_TIMEOUT	5000
> 
> Units?
ms - will be renamed to reflect the same.
> 
>> +
>> +#define MAUSB_PORT_20_STATUS_LOW_SPEED       0x0200
>> +#define MAUSB_PORT_20_STATUS_HIGH_SPEED      0x0400
> 
> Device ids?  Something else?
> 
> If something else, use BIT()?
> 
>> +
>> +enum mausb_device_type {
>> +	USBDEVICE = 0,
>> +	USB20HUB  = 1,
>> +	USB30HUB  = 2
> 
> Trailing , please
> 
>> +};
>> +
>> +enum mausb_device_speed {
>> +	LOW_SPEED	 = 0,
>> +	FULL_SPEED	 = 1,
>> +	HIGH_SPEED	 = 2,
>> +	SUPER_SPEED	 = 3,
>> +	SUPER_SPEED_PLUS = 4
> 
> , please.
> 
>> +};
> 
> Where do these values come from?
> 
>> +
>> +struct mausb_hcd {
>> +	spinlock_t	lock;	/* Protect HCD during URB processing */
>> +	struct device	*pdev;
>> +	u16		connected_ports;
> 
> Why u16?
Seems odd, since max number of ports is much less. Should be corrected.
> 
>> +
>> +	struct rb_root	mausb_urbs;
>> +	struct hub_ctx	*hcd_ss_ctx;
>> +	struct hub_ctx	*hcd_hs_ctx;
>> +	struct notifier_block power_state_listener;
>> +};
>> +
>> +struct mausb_dev {
>> +	u32		port_status;
>> +	struct rb_root	usb_devices;
>> +	u8		dev_speed;
>> +	void		*ma_dev;
>> +};
>> +
>> +struct hub_ctx {
>> +	struct mausb_hcd *mhcd;
>> +	struct usb_hcd	 *hcd;
>> +	struct mausb_dev ma_devs[NUMBER_OF_PORTS];
>> +};
>> +
>> +int mausb_init_hcd(void);
>> +void mausb_deinit_hcd(void);
>> +
>> +#endif /* __MAUSB_HCD_H__ */
>> diff --git a/drivers/usb/host/mausb/mausb_core.c b/drivers/usb/host/mausb/mausb_core.c
>> index 44f76a1b74de..485f241d2b4c 100644
>> --- a/drivers/usb/host/mausb/mausb_core.c
>> +++ b/drivers/usb/host/mausb/mausb_core.c
>> @@ -4,6 +4,7 @@
>>   */
>>  #include <linux/module.h>
>>  
>> +#include "hcd.h"
>>  #include "utils.h"
>>  
>>  MODULE_LICENSE("GPL");
>> @@ -11,12 +12,27 @@ MODULE_AUTHOR("DisplayLink (UK) Ltd.");
>>  
>>  static int mausb_host_init(void)
>>  {
>> -	return mausb_host_dev_register();
>> +	int status = mausb_host_dev_register();
>> +
>> +	if (status < 0)
>> +		goto exit;
>> +
>> +	status = mausb_init_hcd();
>> +	if (status < 0)
>> +		goto cleanup_dev;
>> +
>> +	return 0;
>> +
>> +cleanup_dev:
>> +	mausb_host_dev_deregister();
>> +exit:
>> +	return status;
>>  }
>>  
>>  static void mausb_host_exit(void)
>>  {
>>  	dev_info(mausb_host_dev.this_device, "Module unloading started...");
> 
> This is debugging statements, please remove.  If a driver works
> properly, it does not print anything out.
> 
> Especially as you never give the user the chance to see if module
> unloading ever finished :)
Agree :)
> 
> thanks,
> 
> greg k-h
> 


-- 
Regards,
Vladimir.