Re: [PATCH v3 1/2] usb: dwc2/gadget: add mutex to serialize init/deinit calls

[Marek Szyprowski <m.szyprowski@samsung.com>]

Hello,

On 2014-11-13 21:55, Paul Zimmerman wrote:
>> From: Marek Szyprowski [mailto:m.szyprowski@samsung.com]
>> Sent: Thursday, November 13, 2014 7:18 AM
>>
>> On 2014-10-31 19:46, Paul Zimmerman wrote:
>>>> From: Marek Szyprowski [mailto:m.szyprowski@samsung.com]
>>>> Sent: Friday, October 31, 2014 3:13 AM
>>>>
>>>> This patch adds mutex, which protects initialization and
>>>> deinitialization procedures against suspend/resume methods.
>>>>
>>>> Signed-off-by: Marek Szyprowski <m.szyprowski@samsung.com>
>>>> ---
>>>>    drivers/usb/dwc2/core.h   |  1 +
>>>>    drivers/usb/dwc2/gadget.c | 20 ++++++++++++++++++++
>>>>    2 files changed, 21 insertions(+)
>>>>
>>>> diff --git a/drivers/usb/dwc2/core.h b/drivers/usb/dwc2/core.h
>>>> index 9f77b4d1c5ff..58732a9a0019 100644
>>>> --- a/drivers/usb/dwc2/core.h
>>>> +++ b/drivers/usb/dwc2/core.h
>>>> @@ -187,6 +187,7 @@ struct s3c_hsotg {
>>>>    	struct s3c_hsotg_plat    *plat;
>>>>
>>>>    	spinlock_t              lock;
>>>> +	struct mutex		init_mutex;
>>>>
>>>>    	void __iomem            *regs;
>>>>    	int                     irq;
>>>> diff --git a/drivers/usb/dwc2/gadget.c b/drivers/usb/dwc2/gadget.c
>>>> index d8dda39c9e16..a2e4272a904e 100644
>>>> --- a/drivers/usb/dwc2/gadget.c
>>>> +++ b/drivers/usb/dwc2/gadget.c
>>>> @@ -21,6 +21,7 @@
>>>>    #include <linux/platform_device.h>
>>>>    #include <linux/dma-mapping.h>
>>>>    #include <linux/debugfs.h>
>>>> +#include <linux/mutex.h>
>>>>    #include <linux/seq_file.h>
>>>>    #include <linux/delay.h>
>>>>    #include <linux/io.h>
>>>> @@ -2908,6 +2909,7 @@ static int s3c_hsotg_udc_start(struct usb_gadget *gadget,
>>>>    		return -EINVAL;
>>>>    	}
>>>>
>>>> +	mutex_lock(&hsotg->init_mutex);
>>>>    	WARN_ON(hsotg->driver);
>>>>
>>>>    	driver->driver.bus = NULL;
>>>> @@ -2933,9 +2935,12 @@ static int s3c_hsotg_udc_start(struct usb_gadget *gadget,
>>>>
>>>>    	dev_info(hsotg->dev, "bound driver %s\n", driver->driver.name);
>>>>
>>>> +	mutex_unlock(&hsotg->init_mutex);
>>>> +
>>>>    	return 0;
>>>>
>>>>    err:
>>>> +	mutex_unlock(&hsotg->init_mutex);
>>>>    	hsotg->driver = NULL;
>>>>    	return ret;
>>>>    }
>>>> @@ -2957,6 +2962,8 @@ static int s3c_hsotg_udc_stop(struct usb_gadget *gadget,
>>>>    	if (!hsotg)
>>>>    		return -ENODEV;
>>>>
>>>> +	mutex_lock(&hsotg->init_mutex);
>>>> +
>>>>    	/* all endpoints should be shutdown */
>>>>    	for (ep = 1; ep < hsotg->num_of_eps; ep++)
>>>>    		s3c_hsotg_ep_disable(&hsotg->eps[ep].ep);
>>>> @@ -2974,6 +2981,8 @@ static int s3c_hsotg_udc_stop(struct usb_gadget *gadget,
>>>>
>>>>    	clk_disable(hsotg->clk);
>>>>
>>>> +	mutex_unlock(&hsotg->init_mutex);
>>>> +
>>>>    	return 0;
>>>>    }
>>>>
>>>> @@ -3002,6 +3011,7 @@ static int s3c_hsotg_pullup(struct usb_gadget *gadget, int is_on)
>>>>
>>>>    	dev_dbg(hsotg->dev, "%s: is_on: %d\n", __func__, is_on);
>>>>
>>>> +	mutex_lock(&hsotg->init_mutex);
>>>>    	spin_lock_irqsave(&hsotg->lock, flags);
>>>>    	if (is_on) {
>>>>    		clk_enable(hsotg->clk);
>>>> @@ -3013,6 +3023,7 @@ static int s3c_hsotg_pullup(struct usb_gadget *gadget, int is_on)
>>>>
>>>>    	hsotg->gadget.speed = USB_SPEED_UNKNOWN;
>>>>    	spin_unlock_irqrestore(&hsotg->lock, flags);
>>>> +	mutex_unlock(&hsotg->init_mutex);
>>>>
>>>>    	return 0;
>>>>    }
>>>> @@ -3507,6 +3518,7 @@ static int s3c_hsotg_probe(struct platform_device *pdev)
>>>>    	}
>>>>
>>>>    	spin_lock_init(&hsotg->lock);
>>>> +	mutex_init(&hsotg->init_mutex);
>>>>
>>>>    	hsotg->irq = ret;
>>>>
>>>> @@ -3652,6 +3664,8 @@ static int s3c_hsotg_suspend(struct platform_device *pdev, pm_message_t
>> state)
>>>>    	unsigned long flags;
>>>>    	int ret = 0;
>>>>
>>>> +	mutex_lock(&hsotg->init_mutex);
>>>> +
>>>>    	if (hsotg->driver)
>>>>    		dev_info(hsotg->dev, "suspending usb gadget %s\n",
>>>>    			 hsotg->driver->driver.name);
>>>> @@ -3674,6 +3688,8 @@ static int s3c_hsotg_suspend(struct platform_device *pdev, pm_message_t
>> state)
>>>>    		clk_disable(hsotg->clk);
>>>>    	}
>>>>
>>>> +	mutex_unlock(&hsotg->init_mutex);
>>>> +
>>>>    	return ret;
>>>>    }
>>>>
>>>> @@ -3683,7 +3699,9 @@ static int s3c_hsotg_resume(struct platform_device *pdev)
>>>>    	unsigned long flags;
>>>>    	int ret = 0;
>>>>
>>>> +	mutex_lock(&hsotg->init_mutex);
>>>>    	if (hsotg->driver) {
>>>> +
>>>>    		dev_info(hsotg->dev, "resuming usb gadget %s\n",
>>>>    			 hsotg->driver->driver.name);
>>>>
>>>> @@ -3699,6 +3717,8 @@ static int s3c_hsotg_resume(struct platform_device *pdev)
>>>>    	s3c_hsotg_core_connect(hsotg);
>>>>    	spin_unlock_irqrestore(&hsotg->lock, flags);
>>>>
>>>> +	mutex_unlock(&hsotg->init_mutex);
>>>> +
>>>>    	return ret;
>>>>    }
>>>>
>>> Hmm. I can't find any other UDC driver that uses a mutex in its
>>> suspend/resume functions. Can you explain why this is needed only
>>> for dwc2?
>> I've posted this version because I thought you were not convinced that
>> the patch
>> "usb: dwc2/gadget: rework suspend/resume code to correctly restore
>> gadget state"
>> can add code for initialization and deinitialization in suspend/resume
>> paths.
> My problem with that patch was that you were checking the ->enabled
> flag outside of the spinlock. To address that, you only need to move
> the check inside of the spinlock. I don't see why a mutex is needed.

It is not that simple. I can add spin_lock() before checking enabled, 
but then
I would need to spin_unlock() to call regulator_bulk_enable() and 
phy_enable(),
because both cannot be called from atomic context. This means that the 
spinlock
in such case will not protect anything and is simply useless.

Best regards
-- 
Marek Szyprowski, PhD
Samsung R&D Institute Poland