[ewg] Re: [ofa-general] ANNOUNCE ofed backports for 2.6.22 kernel bits

[Mike Christie]

Erez Zilber wrote:
>> Erez, and other iser maintainers, I had a problem with RHEL4 iscsi backports
>> (scsi_flush_work isn't exported) I decided that since it isn't
>> called on older kernels it's reasonably safe to just comment it out,
>> but would be interested to hear you opinion.
>> See it in this sub-directory:
>> kernel_patches/backport/2.6.9_U2/libiscsi_no_flush_to_2_6_9.patch
>>   
> 
> This leads me to something that I thought about in the past. Old kernels
> (i.e. the RH4 kernels) don't have the SCSI work queue. Therefore, I used
> schedule_work instead of scsi_queue_work. Now, I cannot replace
> scsi_flush_work with flush_workqueue because I'm using a workqueue which
> does not belong to me (and, therefore, I cannot flush it).
> 
> I'm thinking about adding a backport that will create a workqueue for
> each session in open-iscsi. With this, I can queue & flush. Mike - what
> do you think about that? I think that creating a workqueue in open-iscsi
> per session will be the closer thing to the SCSI workqueue that we have
> in new kernels.
> 

Yeah, that sounds fine. Just to be clear, you would want to create the
single threaded work queue (create_singlethread_workqueue) instead of
the normal thread per cpu work queue for each session.