geom mirror in error-prone situations

Ivan Voras ivoras at freebsd.org
Thu Mar 20 17:20:07 UTC 2008


Adam Pordzik wrote:
> Hello,
> 
> recentlty, I was faced a strange situation, where ata(4) reports
> irregularly interrupt storms followed by DMA_WRITE timeouts.
> 
> Mar 20 01:00:30 brain kernel: interrupt storm detected on "irq14:";
> throttling interrupt source
> Mar 20 01:00:30 brain last message repeated 4 times
> Mar 20 01:00:30 brain kernel: ata5: reiniting channel ..
> Mar 20 01:00:30 brain kernel: ata5: SATA connect time=0ms
> Mar 20 01:00:30 brain kernel: ata5: reset tp1 mask=01 ostat0=51 ostat1=00
> Mar 20 01:00:30 brain kernel: ata5: stat0=0x50 err=0x01 lsb=0x00 msb=0x00
> Mar 20 01:00:30 brain kernel: ata5: reset tp2 stat0=50 stat1=00
> devices=0x1<ATA_MASTER>
> Mar 20 01:00:30 brain kernel: ata5: reinit done ..
> Mar 20 01:00:30 brain kernel: ad10: TIMEOUT - WRITE_DMA48 retrying (0
> retries left) LBA=871709696
> 
> The disk was (and still is) pysically faultless.
> 
> How would gmirror handle such a disk? Would that disc be removed
> at the first error? And would it be inserted the next boot? If so,
> does it first have to be rebuild?

If the GEOM system (actually the driver) itself doesn't disconnect the
drive (i.e. "/dev/adX" entry disappears), geom_mirror will still use it.



More information about the freebsd-geom mailing list