Re: OOPS in idescsi_end_request

[Brian King <brking@charter.net>]

James Stevenson wrote:
> [LARGE SNIP]
> 
> 
>>Would you agree to test the patch (possibly next week).
> 
> 
> yeah sure.
> 


I would be happy to test it as well.

-Brian

> 
> 
>>cheers
>>
>>-andrey
>>
>>
>>>>If you can reliably reproduce the problem you could give it a try.
>>>>
>>>>Anybody sees yet another race condition here? :))
>>>>
>>>>-andrey
>>>>
>>>>
>>>>
>>>>>While burning a CD tonight I ended up taking an oops on my system. I
> 
> had
> 
>>>>>the lkcd patch applied to my 2.4.19 kernel, so I was able to look at
> 
> the
> 
>>>>> oops after my system rebooted. After digging into it a little and
>>>>>looking at the ide-scsi code I think I found the problem but am not
>>>>>sure. How can idescsi_reset simply return SCSI_RESET_SUCCESS to the
> 
> scsi
> 
>>>>>mid layer? I think what is happening is that a command times out,
>>>>>idescsi_abort is called, which returns SCSI_ABORT_SNOOZE. Later on
>>>>>idescsi_reset gets called, which returns SCSI_RESET_SUCCESS. At this
>>>>>point the scsi mid-layer owns the scsi_cmnd and returns the failure
> 
> back
> 
>>>>>up the chain. Later on, the command gets run through
>>>>>idescsi_end_request, which then tries to access the scsi_cmnd
> 
> structure
> 
>>>>>which is it no longer owns.
>>>>>
>>>>>Any help is appreciated. I have a complete lkcd dump of the failure if
>>>>>anyone would like more information...
>>>>>
> 
> 
> 
> 


-- 
Some days it's just not worth chewing through the restraints...