Missing volume groups during boot with SCSI and MD

[Simon Richter]

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

Hi,

During the BSP I've taken a look at this, and my current theory is
that the following sequence of events happens:

1. udev event for the new device is generated (this applies to both
SCSI and MD device nodes)
2. the program responsible for creating the new device exits
3. "udevadm settle" is invoked, loaded from cache and executed, finds
that udevd has no pending events and exits.
4. the event is delivered to udevd.

The race condition is between the startup of "udevadm settle" and the
event arriving in udevd.

A possible solution would be to extend "udevadm settle" to include an
explicit synchronisation mechanism, where udevadm would generate a
test event and wait for udevd to process it before then waiting for
the udevd queue to be empty.

The SCSI cases should also be solved by this, provided scanning is
synchronous and the SCSI host adapter modules are loaded via udevd.

If that makes sense, should we reassign these bugs to udev?

   Simon
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1

iJwEAQECAAYFAlRyC7wACgkQ0sfeulffv7vHQAP9Ff1O7KQDwquesmlDjwNyEUgg
qF5HP2uoFkQu3XVFbW3CD/OFWVHHuvBw4I68+0xTlhuVULPqIM8/n0O6z93A0P2j
h2HbvaxfN5JTDRq+N918qWGb+W0t5uBwxuZ294ET+ZyO3k5GBTAHlI1iImpYQ9SM
DUwqlFKs5aRWnhsjVX4=
=Trol
-----END PGP SIGNATURE-----