2.4.20-pre7-ac3 IDE taskfile io woes

2.4.20-pre7-ac3 IDE taskfile io woes

[Paul Cassella]

I turned on taskfile io on my last kernel, and had some stability
problems.  This is ia32, all ide but with ide-scsi.

I experienced severe performance disruptions followed by what appeared to
be a system hang twice.  Before that, I was recieving an "hdd: lost
interrupt"  message every few hours, and the system felt sluggish.  (The
ide-scsi device is hdc, not hdd.)  I have not had these problems with a
kernel compiled without taskfile io.

Is this a known problem?  I can provide more information if it would be
helpful.


I also got the following, which looks like what Nick Evgeniev reported May
24 (2.4.19-pre8 ide bugs).  Andre said to him

> Obviously you have not enable taskfile IO, if you have then you have my
> attention.

As I had enabled taskfile IO, I figured I'd mention this:


11:42:29 bug: kernel timer added twice at c01b0107.

c01b00b8 T ide_set_handler
c01b0110 t atapi_reset_pollfunc

11:45:35 scsi0 channel 0 : resetting for second half of retries.
11:45:35 SCSI bus is being reset for host 0 channel 0.
11:45:35 hdd: ide_set_handler: handler not null; old=c01b0110, new=c01b1474

c01b0110 t atapi_reset_pollfunc
c01b1474 T task_in_intr

11:45:35 bug: kernel timer added twice at c01b0107.

11:47:19 kernel: scsi0 channel 0 : resetting for second half of retries.
11:47:19 kernel: SCSI bus is being reset for host 0 channel 0.
11:47:19 kernel: hdd: ide_set_handler: handler not null; old=c01b0110, new=c01b177c

c01b0110 t atapi_reset_pollfunc
c01b177c T task_out_intr


-- 
Paul Cassella

Re: 2.4.20-pre7-ac3 IDE taskfile io woes

[Alan Cox]

On Mon, 2002-09-30 at 06:10, Paul Cassella wrote:
> I experienced severe performance disruptions followed by what appeared to
> be a system hang twice.  Before that, I was recieving an "hdd: lost
> interrupt"  message every few hours, and the system felt sluggish.  (The
> ide-scsi device is hdc, not hdd.)  I have not had these problems with a
> kernel compiled without taskfile io.

taskfile I/O write is broken and loses IRQ's. Non taskfile I/O write
should be fine. Thats one of the known problems that needs debugging


> 11:42:29 bug: kernel timer added twice at c01b0107.

Thats always a bug.

> 
> c01b00b8 T ide_set_handler
> c01b0110 t atapi_reset_pollfunc
> 
> 11:45:35 scsi0 channel 0 : resetting for second half of retries.
> 11:45:35 SCSI bus is being reset for host 0 channel 0.
> 11:45:35 hdd: ide_set_handler: handler not null; old=c01b0110, new=c01b1474

We ended up trying to add a handler before we had finished getting rid
of the previous one.