All,
my apologies in advance for the wide distribution - it
was recommended that I contact these aliases but if
there is a more appropriate one, please let me know...
I have received the following EFI disk-related
questions from the EMC PowerPath team who
would like to provide more complete support
for EFI disks on Sun platforms...
I would appreciate help in answering these questions...
thanks...
-sharlene
~~~~~~~~~~~~~~~~~~~
1.
Reserving the minor numbers that would have been used for the "h"
slice
for the "wd"
nodes instead creates a "hole". Is this the long term design and
therefore not likely to change?
2. The efi_alloc_and_read(3EXT) function provides a pointer to a
dk_gpt_t structure that
contains an array of efi_nparts dk_part structures.
(efi_partition.h) Will efi_nparts be 15 or 16?
3. Will dk_part[7] correspond to sd@0,0:h or to sd@0,0:wd or is it
somehow reserved?
4. Does the fact that efi_nparts is a member of the dk_gpt_t structure
suggest that it may
not always be { 15 | 16 }? (If so, we should probably query it and
create the number
of nodes indicated).
5. If efi_nparts can be bigger than 16, what happens to the naming of
the partition
nodes ("q" through "t")? If smaller?
6. When changing label types (SMI -> EFI, EFI -> SMI), I assume
the DKIOCSVTOC and
DKIOCSETEFI ioctls are being trapped by the driver to cause the
proper device nodes
to be created, what is the preferred method for cleaning up the
device nodes? e.g. is a
reboot always required? Changing from an EFI to an SMI disk label
leaves the wd node
hanging around, a reboot deletes it. Is this a bug? How are the
/dev/dsk and /dev/rdsk
links manipulated at label change time, is format involved in some
way other than calling
the two aforementioned ioctls or is it accomplished entirely within
the driver? If so, using
what mechanism?
7. For disks that are not immediately accessible (i.e. iSCSI, not ready
disk or fabric) what
is done to create the proper device nodes and links? Which of the
following is Sparc/x86
doing and how?
a. The attach is failed, and no device nodes are created. The attach
is reattempted at
some point in the future triggered by some mechanism (different
for sparc AND x86?).
If so, what event is used to force an attach?
b. The attach is failed and a default set of nodes is created and
adjusted when the device is
accessable?
c. The attach is successful and everything is cleaned up later when the
device is available.
8. Is it possible to query the path_to_inst data via a system call to
determine the minors created
for a disk during early boot? This question is
related to question 7; If the Solaris disk driver
makes a default
assumption about the label type before the device is available and
creates
nodes accordingly, how can this decision be determined? If
the device configuration is delayed
until the device is available the
question is not germane.
9. On Opteron, does a plug and play event trigger the attach, ensuring
that the proper nodes
are created as the label type will be available?
10. On Opteron, why do the cXtXdXp[1-4] links/nodes not address the
corresponding fdisk
partitions? Multiple fdisk partitions can be created but are not
addressable unless each
in turn is set as the active partition and then addressed via p0.
11. An EFI labeled disk must constitute the whole disk, so there can
only be one fdisk partition,
any existing fdisk partitions are overwritten. Is this by design?
It seems to violate the
definition of fdisk partitions?
_______________________________________________
zfs-discuss mailing list
zfs-discuss@opensolaris.org
http://mail.opensolaris.org/mailman/listinfo/zfs-discuss
