[Cbe-oss-dev] [PATCH 7/10] MARS: workload queue api fix

[Geoff Levand]

Yuji Mano wrote:
> Kazunori Asayama wrote:
>> Yuji Mano wrote:
>>> This modifies the internal workload queue API in the host-side library. When
>>> scheduling a new workload, the workload model may need to initialize parts of
>>> the workload context once it has been reserved in an available slot of the
>>> workload queue.
>>> 
>>> Previously a separate call that returns the pointer to the
>>> workload context was provided. However, that may cause simultaneous access
>>> problems since the workload queue will not be locked during the whole
>>> scheduling call.
>>> 
>>> This patch removes the separate API call, and instead adds a callback interface
>>> to allow for the workload model to modify the workload context within the
>>> mutex protected 'workload_queue_schedule' call.
>> 
>> Or alternative solution is to introduce another state such as
>> 'MARS_WORKLOAD_STATE_SCHEDULING' for intermediate state between
>> workload_queue_context() and workload_queue_schedule().
> 
> Yes that is another possible solution.
> 
> workload_queue_schedule_begin()
>  - mutex lock
>  - changes state to MARS_WORKLOAD_STATE_SCHEDULING
>  - mutex unlock
>  - returns pointer to the workload context for workload specific initialization
> 
> workload_queue_schedule_end()
>  - mutex lock
>  - sets necessary bits in workload queue
>  - changes state to MARS_WORKLOAD_STATE_READY
>  - mutex unlock
> 
> Is this preferable to the current callback solution and why?

Callbacks are generally troublesome both in debugging and in reliability,
so something that is not to complex and avoids a callback is usually
preferred.

-Geoff