Re: [v5 PATCH 2/4] bnx2fc: Firmware interface and ELS handling

[Mike Christie <michaelc@cs.wisc.edu>]

On 02/02/2011 09:41 PM, Bhanu Gollapudi wrote:
>>
>>> +				spin_unlock_bh(&fps->fp_work_lock);
>>> +
>>> +				/* Pending work request completion */
>>> +				if (fps->iothread&&   work)
>>> +					wake_up_process(fps->iothread);
>>
>>
>> This looks like blk io poll but with a thread instead of a softirq. Use
>> what is in the kernel already.
>
> I'll have to take a look at implementation details of blk io poll. This
> will be a change in IO path, and I'll have to do a lot of testing. So, I
> cannot promise I can do this change in v6 submittal. Probably will stage
> it later.

Did you just add the completion in a thread code? I do not remember it 
being there before.

What is the benefit?

It seems like if you just call bnx2fc_process_cq_compl from here then 
you go from

network driver isr-> wake network soft irq -> network soft irq -> calls 
driver then driver ends up calling bnx2fc_process_new_cqes?

Why do you want to push that to another thread? And then have that 
thread push it to another soft irq (the block irq)?

Are you not supposed to do a lot of work in the network soft irq or did 
I misread the code that this gets run from the network driver isr?