Red Hat Bugzilla – Bug 38111
RAID funkiness (md: bug in file md.c...)
Last modified: 2005-10-31 17:00:50 EST
Things get "weird" when trying to boot the wolverine-alpha2 kernel on this
A relevant detail may be that the 2.2 kernel sometimes has to spin /dev/sdg
up during boot. Otherwise, everything seems to function normally under
I will attach the 2.4 boot log and the raidtab file. Please let me know
what other information might be useful.
Created attachment 16689 [details]
kernel-smp-2.4.3-2.14.4 boot log on ostrich-deluxe.labs.redhat.com
Created attachment 16690 [details]
Going to add Bryce to the Cc: list just because he claimed it had been fixed,
and it still hasn't. :)
<Grumble> I can't fix what I can't replicate
As and when I can, I'll retackle this problem
Linux/Alpha 2.2.x: sizeof(mdp_super_t) == 4104, offsetof(mdp_super_t,
gstate_reserved) == 168
Linux/Alpha 2.4.x: sizeof(mdp_super_t) == 4096, offsetof(mdp_super_t,
gstate_reserved) == 164.
It looks like someone seriously broke on-disk backwards compat during the
attempt to whittle mdp_super_t down to a dividable-by-1024 size.
Ingo added to Cc's because he probably knows what is going on.
Changing priority to 'low' because I hacked a util to do the conversion.
This ends up being an alignment bug.
The event counter was a 64 bit item and as such was 64bit aligned so there was a
32bit pad before the counter in the 2.2 version of the structure.
the 2.4 version of the structure has 2 32 bit items as the event counter. These
are now 32bit aligned and the padding has gone away.
I have a small patch to check for this condition and repare the raidtable but I
am not sure if it is truly ok to do this.
I can confirm your description because I found the same thing out when creating
the conversion utility. The only question is whether the kernel dudes will ever
do something about it :)
/me twiddles thumbs,...
Created attachment 60825 [details]
The utility I used to solve the problem, for historical interest only
If you're not gonna fix it, say so...