proper struct device selection for dev_printk()

From: Greg KH
Date: Thu May 03 2012 - 12:09:38 EST


Hi Kay,

I've been working on removing the old err() and dbg() functions in usb.h
that have been there since the 2.2 kernel and replace them with calls to
dev_err() and dev_dbg(), as that's what we want to have, especially with
your dev_printk() reworks.

In some recent changes in the input drivers, Dmitry noted that I was
picking the "wrong" struct device to pass to these functions. I was
using the "farthest down the tree" struct device that I could get to, in
the USB input driver's case, the struct device for the input device, a
"class" device.

But that seems to produce an output that is less than helpful. Dmitry
used this as an example output to show this for a serio device:
dev_warn(&input_dev->dev, "warning using input device\n");
dev_warn(&serio->dev, "warning using parent serio device\n");

Produces:
[ 1.903608] input input6: warning using input device
[ 1.903612] psmouse serio1: warning using parent serio device

Here it seems that the "one up from the lowest struct device" works
best.

So I tried this out with a usb to serial device, and got the following
results. With the code:
dev_err(&port->dev, "dev_err port->dev output\n");
dev_err(&serial->dev->dev, "dev_err serial->dev->dev output\n");
dev_err(&serial->interface->dev, "dev_err serial->interface->dev output\n");
dev_err(port->port.tty->dev, "dev_err port->port.tty->dev output\n");

I get:
[ 68.519639] pl2303 ttyUSB0: dev_err port->dev output
[ 68.519645] usb 2-1.2: dev_err serial->dev->dev output
[ 68.519649] pl2303 2-1.2:1.0: dev_err serial->interface->dev output
[ 68.519653] tty ttyUSB0: dev_err port->port.tty->dev output

All of these "describe" the device being operated on in one fashion or
the other, as they are struct devices that are easily accessable from
the driver.

My question is, what is the "best" thing to be doing here?

I still think the "lowest" struct device would be best (in this case,
the last line above from the port->port.tty->dev pointer), but what do
you think is best for userspace to have here?

And, in my conversions, I've realized that I need wrapper functions for
this for each subsystem, I'm tired of typing long -> -> -> pointer
chains for every debug message, that's madness and fragile to get right,
but that's something that I can work on later.

thanks,

greg k-h
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/