I have created multiple RAID5/6 on top of multipath devices. There need to many disks in each set (i.e. 9 disk). I tried with 5 in each set an it didn't show. When reading from all of them simultaneously the kernel crashes with this bug (trace in attached file):<br />
<br />
BUG: scheduling while atomic: md4_raid5/22460/0x10000200<br />
kernel BUG at crypto/async_tx/async_tx.c:180!<br />
<br />
If there is no physical multipath it still crashes, when going through multipath.<br />
<br />
Both I/O directly from the md block devices and filesystems mounted on top makes it crash.<br />
<br />
Using md RAID0 works fine.<br />
<br />
Running the same md sets directly on the disks (/dev/sd<x>), without multipath, works fine.<br />
<br />
If I turn of I/O AT in the BIOS everything works fine.<br />
<br />
I guess is that md uses I/O AT for checksum offloading and it clashes with multipath.
↧