Re: libsas - sas_queue's policy of re-queuing all the tasks ...

[Chetan Loke <chetanloke@gmail.com>]

On Thu, Dec 24, 2009 at 7:26 AM, 王金浦 <jinpuwang@gmail.com> wrote:
>
>
> 2009/12/24 Chetan Loke <chetanloke@gmail.com>
>>
>> Hello,
>>
>> I've a question regarding the libsas core.
>>
>> So, sas_queue() will dispatch the coalesced tasks. It then invokes the
>> SAS_LLDD's execute_task. Let's assume mv_sas driver.In this case
>> mvs_task_exec(). Say, it received 'N' tasks the first time it was
>> called.However, the LLDD was only able to service M tasks(where M<N).
>> Now it will return rc(non-zero value). sas_queue will then requeue all
>> the tasks. But only (N-M) tasks should have been queued, correct? So
>> why are we re-trying all the tasks again?
>>
>
> if you see the code , at mv_init , in function mvs_post_sas_ha_init set
> sha->lldd_max_execute_num = 1;
> and libsas invoke sas_queue_up when sha->lldd_max_execute_num> 1 .
>

Ignore the max_execute_num for this discussion. What I'm saying is, if
you do 'coalesce' the tasks.
mv_sas core clearly has 256+ slots in their adapter. That means you
could coalesce them. Well, that's not the point.
What I would like to know is why does the libsas core wants to requeue
all the tasks?
What was the rationale behind this logic?

> Jack Wang
>

Chetan Loke