Re: running short on DMA buffers during badblocks
On Sat, Sep 20 2003, Michael D. Crawford wrote:
> When I give the following command to do a DESTRUCTIVE bad block check:
>
> badblocks -o 11.bb -b 4096 -v -w /dev/sda11
>
> the following message appears repeatedly on the console:
>
> Warning - running *really* short on DMA buffers
Upgrade your kernel, this bug is fixed in 2.4.20 and newer.
> However, the test seems to proceed normally. Is this a problem? Each time
> the message appears, it repeats a half dozen times immediately, and then
> the half dozen repetitions come again ten or twenty seconds later. The
> message only seems to happen while writing to the drive, not while reading.
It's because of fragmentation in the scsi dma pool, and with writes you
typically have a lot more commands pending. As you note, the message
doesn't cause corruption: it's like the message says, a warning telling
you about slightly degraded io performance.
--
Jens Axboe
Reply to: