Re: CD burn buffer underruns on 2.6

[Jens Axboe <axboe@suse.de>]

On Sun, Dec 28 2003, craig duncan wrote:
> My first attempts at burning on 2.6 resulted in a couple of coasters
> (i'm running the "stock" 2.6 Debian kernel: kernel-image-2.6.0-test9-1-386).
> Burn speed was 8x (max speed for my PleXwrite 8/4/32A).  With the 2.4
> kernel (2.4.22-1-k7 on a 1.3 gig Athlon) i never had a problem burning
> at 8x (in fact, when burning completes, it always tells me:
> Buffer fill min 93%/max 100%.
> 
> At 4x, burns on 2.6 work fine, though.  At 8x, the burn always craps out
> shortly after starting.  This is completely consistent... 8x fails, 4x
> works (see cdrdao output below).  Nothing else is going on on my system.
> Running as root or regular user doesn't matter.
> 
> After a few of these failed burns ("cdrdao simulate...", actually) i saw
> this in /var/log/kern.log:
> 
> Dec 24 08:24:44 cdw kernel: cdrom_newpc_intr: 110 residual after xfer
> Dec 24 08:24:57 cdw kernel: cdrom_newpc_intr: 104 residual after xfer
> Dec 24 08:25:43 cdw kernel: cdrom_newpc_intr: 110 residual after xfer
> Dec 24 08:25:55 cdw kernel: cdrom_newpc_intr: 104 residual after xfer
> Dec 24 08:26:52 cdw kernel: cdrom_newpc_intr: 110 residual after xfer
> Dec 24 08:27:03 cdw kernel: cdrom_newpc_intr: 104 residual after xfer
> 
> As a side note, under 2.6, i see these "resid: 110", "resid: 104", etc.
> messages in the cdrdao output (below) which i never saw under 2.4.
> 
> Here's my cdrdao output:
> -----
> Starting write simulation resid: 110
> at speed 8...
> Pausing 10 seconds - hit CTRL-C to abort.
> Process can be aborted with QUIT signal (usually CTRL-\).
> resid: 104
> Writing CD-TEXT lead-in...
> Writing track 01 (mode AUDIO/AUDIO )...
> Writing track 02 (mode AUDIO/AUDIO )...
> ?: Success.  : scsi sendcmd: no error
> CDB:  2A 00 00 00 26 80 00 00 1A 00
> status: 0x2 (CHECK CONDITION)
> Sense Bytes: F0 00 03 00 00 27 18 0A 00 00 00 00 0C 09 00 00
> Sense Key: 0x3 Medium Error, Segment 0
> Sense Code: 0x0C Qual 0x09 (write error - loss of streaming) Fru 0x0
> Sense flags: Blk 10008 (valid) 
> resid: 61152
> cmd finished after 0.014s timeout 180s
> ERROR: Write data failed.
> ERROR: Writing failed - buffer under run?
> ERROR: Simulation failed.
> 
> I looked through the kernel archives and didn't see anyone else
> reporting this problem but it seems kind of serious to me.

Maybe cdrdao doesn't align its buffers correctly? Can you send me a
vmstat 1 output while attempting a 8x burn on 2.6?

-- 
Jens Axboe