808684 - PandaBoard: Out-of-the-box support for Linaro partitions

Status: RESOLVED FIXED

(Firefox OS Graveyard :: General, defect)

Description

[Thomas Zimmermann [:tzimmermann] [:tdz]]

A Linaro-based partition layout is used for automated testing on the PandaBoard. Mounting these partitions during start-up needs a change to the init scripts. The necessary patch is available at [1], but ideally the differences in the partitions should be transparent.

We need a mechanism to determine whether we're on a AOSP or Linaro setup, and mount the partitions accordingly. Android init scripts don't seem to support conditionals, so I implemented a mount script to translate the device paths automatically. This didn't work, because the shell is not available in the root partition. Instead, we could implement this idea as binary program.

[1] https://github.com/tdz/android-device-panda/commits/bug-778248-linaro

Comment 1

[Thomas Zimmermann [:tzimmermann] [:tdz]]

Michael, Dave,

what you you think about adding a little wrapper around mount for translating the device path on Linaro. The conversion is as follows:

  - if the device file does exist then mount it;
  - if the device file does not exist, but a translation to a Linaro device file exists, then
    mount the Linaro file; or otherwise
  - forward the everything to mount and let mount decide.

Comment 2

[Michael Wu [:mwu]]

A simpler alternative may be to add a parameter to the kernel commandline that the init scripts can look for. https://github.com/android/platform_system_core/blob/master/init/init.c#L579 has a list of kernel options which init imports into properties. Maybe there's one you can abuse to switch between partition configurations?

Comment 3

[Dave Hylands [:dhylands]]

If its just going to be hard-coded mount A or mount B, then you could just try to mount them both. One will fail. One should succeed.

Will init keep executing stuff if something fails? If not, then creating a wrapper which always returns success if probably easier.

Comment 4

[Michael Wu [:mwu]]

(In reply to Dave Hylands [:dhylands] from comment #3)
> If its just going to be hard-coded mount A or mount B, then you could just
> try to mount them both. One will fail. One should succeed.
> 

That's the best option and should work - init doesn't care whether a command succeeds or not.

Comment 5

[Thomas Zimmermann [:tzimmermann] [:tdz]]

(In reply to Michael Wu [:mwu] from comment #4)
> (In reply to Dave Hylands [:dhylands] from comment #3)
> > If its just going to be hard-coded mount A or mount B, then you could just
> > try to mount them both. One will fail. One should succeed.
> > 
> 
> That's the best option and should work - init doesn't care whether a command
> succeeds or not.

TBH this seems a bit ugly to me. I'd prefer to not have invalid or unnecessary error messages in the log files.

I think I'll spend a day investigating the command-line approach. We can probably use the simple approach as a fall back.

Comment 6

[Michael Wu [:mwu]]

(In reply to Thomas Zimmermann from comment #5)
> (In reply to Michael Wu [:mwu] from comment #4)
> > (In reply to Dave Hylands [:dhylands] from comment #3)
> > > If its just going to be hard-coded mount A or mount B, then you could just
> > > try to mount them both. One will fail. One should succeed.
> > > 
> > 
> > That's the best option and should work - init doesn't care whether a command
> > succeeds or not.
> 
> TBH this seems a bit ugly to me. I'd prefer to not have invalid or
> unnecessary error messages in the log files.
> 

This is what Android actually does by default - The default android init.rc always tries to mount some yaffs2 partitions from an mtd device, and there's usually a device specific init file which has proper mount commands when using ext4.

Comment 7

[Dave Hylands [:dhylands]]

From a cursory look at the init code, it appears that any option of the form androidboot.* will create a property r.boot.* (replace * with arbitrary string)

https://github.com/android/platform_system_core/blob/master/init/init.c#L605

Comment 8

[Thomas Zimmermann [:tzimmermann] [:tdz]]

(In reply to Dave Hylands [:dhylands] from comment #7)
> From a cursory look at the init code, it appears that any option of the form
> androidboot.* will create a property r.boot.* (replace * with arbitrary
> string)
> 
> https://github.com/android/platform_system_core/blob/master/init/init.c#L605

Yes, I though about something link 'r.boot.environment=linaro'. If it's not set, the init script might be able to fall back to the old method.

Comment 9

[Thomas Zimmermann [:tzimmermann] [:tdz]]

Michael,

Please have a look at the pull request

  https://github.com/mozilla-b2g/android-device-panda/pull/7

and merge it if appropriate. Thanks!