Re: [PATCH 02/12] usb: usbtmc: Support Read Status Byte with SRQ per file handle

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

 




Zitat von Greg KH <gregkh@xxxxxxxxxxxxxxxxxxx>:

On Thu, May 17, 2018 at 07:03:26PM +0200, Guido Kiener wrote:
- Add 'struct usbtmc_file_data' for each file handle to cache last
  srq_byte (=Status Byte with SRQ) received by usbtmc_interrupt(..)

- usbtmc488_ioctl_read_stb returns cached srq_byte when available for
  each file handle to avoid race conditions of concurrent applications.

- SRQ now sets EPOLLPRI instead of EPOLLIN

- Caches other values TermChar, TermCharEnabled, auto_abort in
  'struct usbtmc_file_data' will not be changed by sysfs device
  attributes during an open file session.
  Future ioctl functions can change these values.

- use consistent error return value ETIMEOUT instead of ETIME

Tested-by: Dave Penkler <dpenkler@xxxxxxxxx>
Reviewed-by: Steve Bayless <steve_bayless@xxxxxxxxxxxx>
Signed-off-by: Guido Kiener <guido.kiener@xxxxxxxxxxxxxxxxx>
---
 drivers/usb/class/usbtmc.c | 176 ++++++++++++++++++++++++++++---------
 1 file changed, 136 insertions(+), 40 deletions(-)

diff --git a/drivers/usb/class/usbtmc.c b/drivers/usb/class/usbtmc.c
index 529295a17579..5754354429d8 100644
--- a/drivers/usb/class/usbtmc.c
+++ b/drivers/usb/class/usbtmc.c
@@ -67,6 +67,7 @@ struct usbtmc_device_data {
 	const struct usb_device_id *id;
 	struct usb_device *usb_dev;
 	struct usb_interface *intf;
+	struct list_head file_list;

 	unsigned int bulk_in;
 	unsigned int bulk_out;
@@ -87,12 +88,12 @@ struct usbtmc_device_data {
 	int            iin_interval;
 	struct urb    *iin_urb;
 	u16            iin_wMaxPacketSize;
-	atomic_t       srq_asserted;

 	/* coalesced usb488_caps from usbtmc_dev_capabilities */
 	__u8 usb488_caps;

 	/* attributes from the USB TMC spec for this device */
+	/* They are used as default values for file_data */
 	u8 TermChar;
 	bool TermCharEnabled;
 	bool auto_abort;
@@ -104,9 +105,26 @@ struct usbtmc_device_data {
 	struct mutex io_mutex;	/* only one i/o function running at a time */
 	wait_queue_head_t waitq;
 	struct fasync_struct *fasync;
+	spinlock_t dev_lock; /* lock for file_list */
 };
 #define to_usbtmc_data(d) container_of(d, struct usbtmc_device_data, kref)

+/*
+ * This structure holds private data for each USBTMC file handle.
+ */
+struct usbtmc_file_data {
+	struct usbtmc_device_data *data;
+	struct list_head file_elem;
+
+	u8             srq_byte;
+	atomic_t       srq_asserted;
+
+	/* These values are initialized with default values from device_data */
+	u8             TermChar;
+	bool           TermCharEnabled;

I don't remember, does TermChar and TermCharEnabled come from a
specification somewhere?  Is that why they are in InterCaps format?

TermChar and TermCharEnabled was introducted 10 years ago by your patches.
We can rename these fields in term_char and term_char_enabled as well.


And why duplicate these fields as they are already in the
device-specific data structure?

We do not need it in device-specific data structure.
We need it in file-specific data structure.
We keep it in device-specific data structure, since we do not want to break
previous applications that wants to change it with via sysfs.


+	bool           auto_abort;
+};
+
 /* Forward declarations */
 static struct usb_driver usbtmc_driver;

@@ -114,6 +132,7 @@ static void usbtmc_delete(struct kref *kref)
 {
 	struct usbtmc_device_data *data = to_usbtmc_data(kref);

+	pr_debug("%s - called\n", __func__);
 	usb_put_dev(data->usb_dev);
 	kfree(data);
 }
@@ -122,7 +141,7 @@ static int usbtmc_open(struct inode *inode, struct file *filp)
 {
 	struct usb_interface *intf;
 	struct usbtmc_device_data *data;
-	int retval = 0;
+	struct usbtmc_file_data *file_data;

 	intf = usb_find_interface(&usbtmc_driver, iminor(inode));
 	if (!intf) {
@@ -130,21 +149,54 @@ static int usbtmc_open(struct inode *inode, struct file *filp)
 		return -ENODEV;
 	}

+	file_data = kzalloc(sizeof(*file_data), GFP_KERNEL);
+	if (!file_data)
+		return -ENOMEM;
+
+	pr_debug("%s - called\n", __func__);

Please do not add "tracing" functions like this.  The kernel has a
wonderful built-in function tracing functionality, don't try to write
your own.  These lines should just be removed.

Ok, I agree. Sorry, I'm a newbie.



+
 	data = usb_get_intfdata(intf);
 	/* Protect reference to data from file structure until release */
 	kref_get(&data->kref);

+	mutex_lock(&data->io_mutex);
+	file_data->data = data;
+
+	/* copy default values from device settings */
+	file_data->TermChar = data->TermChar;
+	file_data->TermCharEnabled = data->TermCharEnabled;
+	file_data->auto_abort = data->auto_abort;
+
+	INIT_LIST_HEAD(&file_data->file_elem);
+	spin_lock_irq(&data->dev_lock);
+	list_add_tail(&file_data->file_elem, &data->file_list);
+	spin_unlock_irq(&data->dev_lock);
+	mutex_unlock(&data->io_mutex);
+
 	/* Store pointer in file structure's private data field */
-	filp->private_data = data;
+	filp->private_data = file_data;

-	return retval;
+	return 0;
 }

 static int usbtmc_release(struct inode *inode, struct file *file)
 {
-	struct usbtmc_device_data *data = file->private_data;
+	struct usbtmc_file_data *file_data = file->private_data;

-	kref_put(&data->kref, usbtmc_delete);
+	pr_debug("%s - called\n", __func__);

Again, these all need to be dropped.

I agree.


+
+	/* prevent IO _AND_ usbtmc_interrupt */
+	mutex_lock(&file_data->data->io_mutex);
+	spin_lock_irq(&file_data->data->dev_lock);
+
+	list_del(&file_data->file_elem);
+
+	spin_unlock_irq(&file_data->data->dev_lock);
+	mutex_unlock(&file_data->data->io_mutex);

You protect the list, but what about removing the data itself?

+
+	kref_put(&file_data->data->kref, usbtmc_delete);

What protects this from being called at the same time a kref_get is
being called?

Yeah, it previously probably already had this race, sorry I never
noticed that.

I will need more time to think about the race here.




+	file_data->data = NULL;
+	kfree(file_data);
 	return 0;
 }

@@ -369,10 +421,12 @@ static int usbtmc_ioctl_abort_bulk_out(struct usbtmc_device_data *data)
 	return rv;
 }

-static int usbtmc488_ioctl_read_stb(struct usbtmc_device_data *data,
+static int usbtmc488_ioctl_read_stb(struct usbtmc_file_data *file_data,
 				void __user *arg)
 {
+	struct usbtmc_device_data *data = file_data->data;
 	struct device *dev = &data->intf->dev;
+	int srq_asserted = 0;
 	u8 *buffer;
 	u8 tag;
 	__u8 stb;
@@ -381,15 +435,27 @@ static int usbtmc488_ioctl_read_stb(struct usbtmc_device_data *data,
 	dev_dbg(dev, "Enter ioctl_read_stb iin_ep_present: %d\n",
 		data->iin_ep_present);

+	spin_lock_irq(&data->dev_lock);
+	srq_asserted = atomic_xchg(&file_data->srq_asserted, srq_asserted);

That really frightens me. Why are you messing with atomic values here?
What is it supposed to be "protecting" or "doing"?

file_data->srq_asserted is of type atomic_t. In this patch we could also use
the simple type int. However in patch 07/12 we need an
atomic_read(&file_data->srq_asserted) != 0 in usbtmc488_ioctl_wait_srq.

I assumed that we should use the atomic_* functions when using atomic_t.


thanks,

greg k-h



--
To unsubscribe from this list: send the line "unsubscribe linux-usb" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html



[Index of Archives]     [Linux Media]     [Linux Input]     [Linux Audio Users]     [Yosemite News]     [Linux Kernel]     [Linux SCSI]     [Old Linux USB Devel Archive]

  Powered by Linux