Re: [PATCH] usb: udc: add gadget state kobject uevent

[Felipe Balbi]

Hi,

On Thu, Jul 18, 2013 at 04:33:43PM +0800, Rong Wang wrote:
> On Wed, Jul 17, 2013 at 9:27 PM, Felipe Balbi <balbi@xxxxxx> wrote:
> > On Wed, Jul 17, 2013 at 09:04:54PM +0800, Rong Wang wrote:
> >> Hi Felipe,
> >>
> >> On Wed, Jul 17, 2013 at 3:57 PM, Felipe Balbi <balbi@xxxxxx> wrote:
> >> > Hi,
> >> >
> >> > On Mon, Jul 15, 2013 at 11:31:17PM -0700, Greg KH wrote:
> >> >> > The question is since we default GADGET, so the g_mass_storage.ko is
> >> >> > installed early but connecting to a host PC
> >> >> > is randomly, But the udev has no idea when a host PC connects our device.
> >> >> >
> >> >> > So we consider it's reasonable to let the udev know the GADGET device state.
> >> >> > Is there any alternative to our question?
> >> >>
> >> >> I thought we already export events for gadget device states, have you
> >> >> looked for them?  I can't dig through the code at the moment, but this
> >> >> seems like a pretty common issue...
> >> >>
> >> >> Felipe, any ideas?
> >> >
> >> > we already expose that in sysfs. IIRC udev can act on sysfs changes,
> >> > no ?
> >>
> >> I do not know if udev can polling sysfs file content change. I'll study this.
> >>
> >> But the change is triggered by calling usb_gadget_set_state, and I find
> >> composite framework do not call this. Then we should do this common work
> >> in every udc driver?
> >
> > yes. Only the UDC driver knows when the controller is moving among those
> > states.
> 
> OK. I got that.
> 
> But I think it may be more reasonable for the udc driver to maintain a
> work queue
> to handle the state change since this operation mostly happen in ISR ?

And that's the patch I want to test. Adding a workqueue to every single
UDC will be too much, so I tried to hide it in udc-core.c. I agree with
you we need to pass the sysfs_notification to a separate workqueue
though. If you can test the patch below, that would be great.

commit d32521bd775d48b600e67f23d363d70f71597ffd
Author: Felipe Balbi <balbi@xxxxxx>
Date:   Wed Jul 17 11:09:49 2013 +0300

    usb: gadget: udc-core: move sysfs_notify() to a workqueue
    
    usb_gadget_set_state() will call sysfs_notify()
    which might sleep. Some users might want to call
    usb_gadget_set_state() from the very IRQ handler
    which actually changes the gadget state.
    
    Instead of having every UDC driver add their own
    workqueue for such a simple notification, we're
    adding it generically to our struct usb_gadget,
    so the details are hidden from all UDC drivers.
    
    Signed-off-by: Felipe Balbi <balbi@xxxxxx>

diff --git a/drivers/usb/gadget/udc-core.c b/drivers/usb/gadget/udc-core.c
index ffd8fa5..b0d91b1 100644
--- a/drivers/usb/gadget/udc-core.c
+++ b/drivers/usb/gadget/udc-core.c
@@ -23,6 +23,7 @@
 #include <linux/list.h>
 #include <linux/err.h>
 #include <linux/dma-mapping.h>
+#include <linux/workqueue.h>
 
 #include <linux/usb/ch9.h>
 #include <linux/usb/gadget.h>
@@ -101,11 +102,18 @@ EXPORT_SYMBOL_GPL(usb_gadget_unmap_request);
 
 /* ------------------------------------------------------------------------- */
 
+static void usb_gadget_state_work(struct work_struct *work)
+{
+	struct usb_gadget	*gadget = work_to_gadget(work);
+
+	sysfs_notify(&gadget->dev.kobj, NULL, "status");
+}
+
 void usb_gadget_set_state(struct usb_gadget *gadget,
 		enum usb_device_state state)
 {
 	gadget->state = state;
-	sysfs_notify(&gadget->dev.kobj, NULL, "status");
+	schedule_work(&gadget->work);
 }
 EXPORT_SYMBOL_GPL(usb_gadget_set_state);
 
@@ -192,6 +200,7 @@ int usb_add_gadget_udc_release(struct device *parent, struct usb_gadget *gadget,
 		goto err1;
 
 	dev_set_name(&gadget->dev, "gadget");
+	INIT_WORK(&gadget->work, usb_gadget_state_work);
 	gadget->dev.parent = parent;
 
 	dma_set_coherent_mask(&gadget->dev, parent->coherent_dma_mask);
diff --git a/include/linux/usb/gadget.h b/include/linux/usb/gadget.h
index f1b0dca..942ef5e 100644
--- a/include/linux/usb/gadget.h
+++ b/include/linux/usb/gadget.h
@@ -22,6 +22,7 @@
 #include <linux/slab.h>
 #include <linux/scatterlist.h>
 #include <linux/types.h>
+#include <linux/workqueue.h>
 #include <linux/usb/ch9.h>
 
 struct usb_ep;
@@ -475,6 +476,7 @@ struct usb_gadget_ops {
 
 /**
  * struct usb_gadget - represents a usb slave device
+ * @work: (internal use) Workqueue to be used for sysfs_notify()
  * @ops: Function pointers used to access hardware-specific operations.
  * @ep0: Endpoint zero, used when reading or writing responses to
  *	driver setup() requests
@@ -520,6 +522,7 @@ struct usb_gadget_ops {
  * device is acting as a B-Peripheral (so is_a_peripheral is false).
  */
 struct usb_gadget {
+	struct work_struct		work;
 	/* readonly to gadget driver */
 	const struct usb_gadget_ops	*ops;
 	struct usb_ep			*ep0;
@@ -538,6 +541,7 @@ struct usb_gadget {
 	unsigned			out_epnum;
 	unsigned			in_epnum;
 };
+#define work_to_gadget(w)	(container_of((w), struct usb_gadget, work))
 
 static inline void set_gadget_data(struct usb_gadget *gadget, void *data)
 	{ dev_set_drvdata(&gadget->dev, data); }

-- 
balbi

Attachment: signature.asc
Description: Digital signature