During the past year I've been developing software RAID server management
tools based on kernels 2.2.16 and lower. Yesterday, I upgraded my
development system to linux-2.2.17 and applied Mingo's raid-2.2.17-A0
patch.
After upgrading, I performed a series of regression tests and discovered
the creation of a RAID-1 array causes a continuous stream of "raid1: out of
memory, retrying..." errors. I rebooted the machine after a couple minutes
and examined raid1.c to see what I could find. I'm not a kernel developer
(hacker if you prefer), so I didn't find too much more than a while() loop
that repeatedly attempts to malloc memory if it fails.
Taking what little information I found, I went to the linux-raid archive and
found one thread that might be related. The first post (by Neil Brown on
July 18) was titled, "PATCH raid1.c - possible deadlock under high memory
pressure" (http://marc.theaimsgroup.com/?l=linux-raid&m=96388716303114&w=2)
mentioned a possible deadlock between kflushd and raid1.
Having nothing to lose and the assumption that this might be a related
problem, I applied the changes to raid1.c (GFP_KERNEL -> GFP_BUFFER) and
tried again. This time it *worked*. I even wrote a script to repeatedly
create RAID-1 devices while I went to lunch; no sign of errors when I came
back.
I apologize for the long story, so here is my question:
Am I on the right track or did I just hide the problem?
Has anyone else had similar problems?
BTW, my development platform has 64MB of memory, although not an enormous
amount, it should be enough to create a RAID-1 device right? That's why
I thought this problem was related to Neil's deadlock situation when I
found the article.
Any information would be appreciated.
-kral
============================================================================
Brian M. Kral [EMAIL PROTECTED]
ApplianceWare, Inc. www.applianceware.com
Fremont, CA
-
To unsubscribe from this list: send the line "unsubscribe linux-raid" in
the body of a message to [EMAIL PROTECTED]
