Eric Youngdale <[EMAIL PROTECTED]> wrote:
> For what it is worth, I was testing the new queueing code in the 2.3
>kernels using a 1542 and an old scratched up CDROM that gives lots of sector
>errors.
Most of my testing is with a clean cdrom but code that reads lots of scattered
sectors all over the cd. I think the thrashes cause the cdrom to become
more and more behind in handling the request (cdroms seek badly) until
it starts timing out.
Sometimes if I preload the cache, it works fine.
>Eventually device got into such a weird state that the error handling code
>took the device offline. Once this took place, the machine was back to
>normal
[...]
> I think I should test with an additional device on the 1542 bus,
I have a number of devices on my 1542 bus. Several of them are cdroms,
and if I can manage to trigger behavior that slows them down enough,
I get bus resets. Actually, I get bus resets on anything that should
probably have been an I/O error. I've never seen an I/O error, however.
Things don't return after a bus reset to give an I/O error.
The other devices on the bus are unaffected before, during, and after
the reset. In fact, one of the devices is a cd changer, and the other
sub-devices (linux side) work fine even after one has errored out.
This pretty much convinces me that the problem is purely a linux software
problem, and not a hardware problem at all. I'm sure that
a) nothing is hanging the bus, as other devices on the bus work fine
b) the device that caused the reset is not hanging the bus, because
other subdevices of the same device work fine too
I think perhaps there are at least two bugs here--the one that caused
the (unnecessary?) reset, and the one that didn't properly clean up
the device after the reset.
(BTW: the changer dies randomly if I stick in certain commercial mixed
multisession cd's. It seems to get confused; it works if I read things
in a certain order, but just thrashes forever trying to find the TOC if I
ask for the wrong thing first. The bus reset DOES reset it. Sometimes.)
Steve
-
To unsubscribe from this list: send the line "unsubscribe linux-scsi" in
the body of a message to [EMAIL PROTECTED]
