gmirror & sparc64 oddness

[Chris Elsworth]

On Mon, Apr 11, 2005 at 03:57:37PM +0200, Pawel Jakub Dawidek wrote:
> 
> No, the problem is that geom_sunlabel class operates only on rank#1
> providers (disks, etc.).
> Mirrors are not rank#1 of course, because there are providers under them.
> This was done to prevent recursive labels creations.
> 
> geom_bsd and geom_mbr are already fixed and are using MD5 to verify
> if metadata were already used for slices/labels creation.
> 
> I don't have HW to test this, but please try this patch and tell me
> if it works for you:

Hello Pawel,

This seems to work fine - no problems so far and it's happily booted
off a gmirror device. The only thing I've noticed so far is that
writes during a synchronization of components don't seem to be
appearing in gstat for the mirror:

dT: 1.010  flag_I 1000000us  sizeof 288  i -1
 L(q)  ops/s    r/s   kBps   ms/r    w/s   kBps   ms/w   %busy Name
    0      0      0      0    0.0      0      0    0.0    0.0| acd0
    1     77      0      0    0.0     77   9757    6.6   50.9| da0
    0     77     77   9884    1.8      0      0    0.0   13.8| da1
    0     77     77   9884    1.8      0      0    0.0   14.1| mirror/gm0

This is a sync from da1 to da0, but notice how the writes values for
the mirror are zero. I can't recall offhand whether this happens on
i386 as well, but I just thought I'd point it out as an oddity ; or
this is recognised behaviour?

Thanks very much for the patch, hope it makes it into the tree :)

Cheers,
-- 
Chris