Re: [PATCHv3 1/1] [RFC] uartclk from serial_core exposed to sysfs

[Marek Vasut <marek.vasut@gmail.com>]

Dear Greg KH,

> On Fri, Aug 17, 2012 at 06:30:36PM +0200, Tomas Hlavacek wrote:
> > Hello Greg!
> > 
> > On Fri, Aug 17, 2012 at 5:06 PM, Greg KH <gregkh@linuxfoundation.org> wrote:
> > >> @@ -2355,6 +2373,14 @@ int uart_add_one_port(struct uart_driver *drv,
> > >> struct uart_port *uport)
> > >> 
> > >>       }
> > >>       
> > >>       /*
> > >> 
> > >> +      * Expose uartclk in sysfs. Use driverdata of the tty device for
> > >> +      * referencing the UART port.
> > >> +      */
> > >> +     dev_set_drvdata(tty_dev, port);
> > >> +     if (device_create_file(tty_dev, &dev_attr_uartclk) < 0)
> > >> +             dev_err(tty_dev, "Failed to add uartclk attr\n");
> > > 
> > > I think you just raced with userspace in creating the file after the
> > > device was announced to userspace.  Are you sure it's ok?
> > > 
> > > If not (hint, I don't think so), please make it a default attribute of
> > > the device, which will then cause the file to be created before it is
> > > announced to userspace.  It will also be less code as you don't have to
> > > clean it up by hand :)
> > 
> > Do you mean I should modify the tty_register_device() function not to
> > use device_create() but it should rather do the device initialization
> > on it's own.
> 
> No, not at all.
> 
> > And I should add add the attribute (via struct attribute_group) to
> > struct device in between device_initialize() and device_add() calls.
> > Did I get it right?
> 
> No, make this a driver attribute, that way when the device is
> registered, it adds the attribute automagically to the device that is
> bound to it.

(hint, DEVICE_ATTR), right ?

> Does that make sense?
> 
> greg k-h

Best regards,
Marek Vasut