Joerg Schilling wrote:
> <snip/>
>
> As I cannot see any progress with ATAPI within the last 3 years,
> I susect that the version does not matter.
>
> It is a vanilla SuSe 7.2 with 2.4.4-4GB kernel.
>
> I am unsure why you ask for the writer interface, I already mentioned it:
>
> PCATA
Ah, I thought that was the brand. So _all_ ATAPI cdwriters
lock up on Sony VAIO notebooks? It seems strange that
it locks up on /dev/sg1 in a scan. That implies there
was something on /dev/sg0 (perhaps a cd/dvd reader?).
The output of "cat /proc/scsi/sg/*" may help me
understand what is happening.
>>Looking at sg_open() the only "wait" done there is
>>associated with the exclusive lock and that wait is
>>interruptible. sg_open() does call the linux scsi
>>mid-level code [scsi_block_when_processing_errors()]
>>which could wait forever in non-interruptible state.
>>This would imply that the device was continually in
>>the "recovering from error" state.
>>
>
>>If you have access to the machine and cdwriter in
>>question and the machine is still useable after the
>>reported problem, then this command may help find
>>where open("/dev/sg1", ...) is hanging inside the
>>kernel:
>> ps -eo cmd,wchan
>>
>
> It hangs in "scsi_block_when_processing_errors"
So that rules out a _sg_ driver bug :-)
>>There may also be some information in the /var/log/messages
>>file (potentially showing ongoing device resets). If the
>>cdwriter is ATAPI and the kernel is in the 2.4 series
>>then turning off (or reducing the speed of) DMA may be
>>helpful.
>>
>
> There is absolutely nothing in the message files.
>
> I am not sure if you don't know this, but Linux does not support
> DMA for PCATA drives. Some people state that this is not true,
> but those people could never have tried to use Linux this way.
> Unfortunately, Alan Cox is ne of these persons ;-(
>
> If you turn on DMA at the device, Linux will not detect it.
As I have noted before, the IDE subsystem in the lk 2.4
series tries to maximize the DMA mode, even to cdwriters.
This causes problems with cdwriters that can't actually
handle that speed (and some IDE chipsets).
Also the lockup that you are reporting would be avoided
if the bus scan was performed with the O_NONBLOCK flag
on open() [as my documentation recommends].
Doug Gilbert
P.S. Email I send to you directly bounces.
--
To UNSUBSCRIBE, email to [EMAIL PROTECTED]
with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]
