/dev/disk0 vs /dev/mmc0

[David Jander]

On Mon, 7 Oct 2013 22:19:36 +0200
Sascha Hauer <s.hauer at pengutronix.de> wrote:

> On Mon, Oct 07, 2013 at 11:57:35AM +0200, David Jander wrote:
> > On Mon, 7 Oct 2013 08:41:11 +0200
> > Sascha Hauer <s.hauer at pengutronix.de> wrote:
> > 
> > > On Mon, Oct 07, 2013 at 08:32:03AM +0200, David Jander wrote:
> > > > 
> > > > Dear Sascha,
> > > > 
> > > > On Sun, 6 Oct 2013 12:39:50 +0200
> > > > Sascha Hauer <s.hauer at pengutronix.de> wrote:
> > > > > 
> > > > > It doesn't interfere with the kernel. The kernel currently ignores
> > > > > this aliases. There are patches floating to let the kernel honor this
> > > > > aliases, but then they should simply have the same effect as they
> > > > > have in barebox.
> > > > 
> > > > That doesn't _feel_ right. Changing standard device names in Linux via
> > > > aliases in the DT might be a fancy idea, but it'd have a different
> > > > reason/use-case than in the case of barebox. IMHO using the same DT for
> > > > both seems to be the Right Thing (tm) to do, but then the semantics
> > > > must be the same also. If I need aliases in the DT only to be able to
> > > > tell devices apart from each other in barebox, while in Linux the
> > > > effect would only be a rather inconvenient renaming of devices with no
> > > > other added value, I think we need a different way to differentiate
> > > > devices in barebox. Why not just use a simple driver-provided prefix
> > > > (mci, mmc, usb, sata, etc...) for now?
> > > 
> > > That's not enough. We also need a fixed numbering. Otherwise a
> > > nonremovable eMMC and a removable SD card change their device names
> > > depending on the detect order.
> > 
> > True. Can host->dev.id be used? Should be fixed AFAICS...
> 
> host->dev.id depends on the probe order.

:-(

> > > Using aliases to provide a numbering is done in the Kernel aswell at
> > > least for gpios, uarts and i2c busses, so expanding this scheme to
> > > mmc/sd slots doesn't feel too wrong to me.
> > 
> > Yes, but we are doing it in barebox now only because there is no other way
> > to tell devices apart from each other. While the MMC device being called
> > "mmcblk0" or whatever in Linux is perfectly fine (no need for alias),
> 
> Linux has exactly the same problem. There are enough systems on which
> the eMMC changes its name depending on a SD card being plugged in or
> not.

Yes, but you have udev/mdev hotplugging and the sysfs interface to figure out
what you need to know. Barebox has nothing like that AFAICS.

> > in
> > barebox the device is now named "disk0", and there is no way of knowing
> > what "disk0" actually is. Using DT-aliases for that purpose seems wrong to
> > me. Why was this changed anyway? Introduction of some common "disk" layer
> > (like scsi-disk on Linux)? Or just for the sake of confusing it with other
> > "disk"-like devices?
> 
> It has always been diskx on barebox, this behaviour hasn't changed.

You are right. Sorry for the confusion. There were aliases in barebox's
version of imx53.dtsi which went missing somehow in my tree. I have been
merging stuff from mainline Linux....
That leads me to the question: What is the intended relationship between
barebox and Linux kernel DTS files? Which one is supposed to be the "master" or
"upstream" version? Aren't they supposed to be kept in sync somehow? At least
the basic SoC-specific .dtsi files?

Best regards,

-- 
David Jander
Protonic Holland.