[PATCH] NVMe: Add a character device for each nvme device

[Greg KH]

On Fri, Jul 27, 2012 at 03:08:21PM -0400, Matthew Wilcox wrote:
> On Fri, Jul 27, 2012 at 11:25:46AM -0700, Greg KH wrote:
> > On Fri, Jul 27, 2012 at 02:12:12PM -0400, Matthew Wilcox wrote:
> > > I don't see a problem here, but I'm no expert at sysfs / character devices.
> > > Alan, Greg, anyone else see any problems with how this character device is
> > > created / destroyed?
> > 
> > Yes, see below:
> 
> Thanks!
> 
> > > > +	device_create(nvme_char_cl, NULL, MKDEV(nvme_char_major, dev->instance),
> > > > +						NULL, "nvme%d", dev->instance);
> > 
> > You just created a device at the "root" of sysfs, which is wrong,
> > especially when you do have a parent device here.  Please use it.
> 
> OK, that makes sense; this device should be the child of the pci_dev that
> it belongs to.
> 
> > Also, why are you creating your own class?  Can't this just be a misc
> > device?  And if you want to create your own class, please don't, use a
> > bus, as that is what is really happening here, right?  We are trying to
> > move away from using 'struct class' wherever possible (one of these days
> > we'll just remove it...)
> 
> What we're trying to achieve here is to create one character device
> per NVMe controller that gets plugged in.  Each NVMe controller is-a
> PCI function.  The reason we're trying to do this is so that we can send
> commands to the NVMe controller, even when there is no storage present
> (eg a drive is shipped from the factory with no configured storage).
> 
> So we have no particular desire to create a new struct class, or struct
> bus.  If we can create a misc device per PCIe function that's bound to our
> driver, that's great!  Can you recommend a driver that does this already?

I don't think there is one, but it shouldn't be that hard to just create
a 'struct misdevice' for each one of the devices you want to create,
would it?

But, as you really are a "specific type", a bus_type might be overkill,
so the original use of device_create() should be fine.  Just be sure to
fix the parent pointer issue, and you should be fine, right?

greg k-h