Re: [PATCH v3 04/13] mailbox: mediatek: cmdq: clear task in channel before shutdown

[Dennis-YC Hsieh <dennis-yc.hsieh@mediatek.com>]

Hi CK,

Thanks for your comment.

On Fri, 2020-02-28 at 23:57 +0800, CK Hu wrote:
> Hi, Dennis:
> 
> On Fri, 2020-02-28 at 21:44 +0800, Dennis YC Hsieh wrote:
> > Do success callback in channel when shutdown. For those task not finish,
> > callback with error code thus client has chance to cleanup or reset.
> > 
> > Signed-off-by: Dennis YC Hsieh <dennis-yc.hsieh@mediatek.com>
> > ---
> >  drivers/mailbox/mtk-cmdq-mailbox.c | 38 ++++++++++++++++++++++++++++++
> >  1 file changed, 38 insertions(+)
> > 
> > diff --git a/drivers/mailbox/mtk-cmdq-mailbox.c b/drivers/mailbox/mtk-cmdq-mailbox.c
> > index 7246b7e21a2e..50dec015593f 100644
> > --- a/drivers/mailbox/mtk-cmdq-mailbox.c
> > +++ b/drivers/mailbox/mtk-cmdq-mailbox.c
> > @@ -387,6 +387,12 @@ static int cmdq_mbox_send_data(struct mbox_chan *chan, void *data)
> >  
> >  	if (list_empty(&thread->task_busy_list)) {
> >  		WARN_ON(clk_enable(cmdq->clock) < 0);
> > +		/*
> > +		 * The thread reset will clear thread related register to 0,
> > +		 * including pc, end, priority, irq, suspend and enable. Thus
> > +		 * set CMDQ_THR_ENABLED to CMDQ_THR_ENABLE_TASK will enable
> > +		 * thread and make it running.
> > +		 */
> >  		WARN_ON(cmdq_thread_reset(cmdq, thread) < 0);
> >  
> >  		writel(task->pa_base >> cmdq->shift_pa,
> > @@ -450,6 +456,38 @@ static int cmdq_mbox_startup(struct mbox_chan *chan)
> >  
> >  static void cmdq_mbox_shutdown(struct mbox_chan *chan)
> >  {
> > +	struct cmdq_thread *thread = (struct cmdq_thread *)chan->con_priv;
> > +	struct cmdq *cmdq = dev_get_drvdata(chan->mbox->dev);
> > +	struct cmdq_task *task, *tmp;
> > +	unsigned long flags;
> > +
> > +	spin_lock_irqsave(&thread->chan->lock, flags);
> > +	if (list_empty(&thread->task_busy_list))
> > +		goto done;
> > +
> > +	WARN_ON(cmdq_thread_suspend(cmdq, thread) < 0);
> > +
> > +	/* make sure executed tasks have success callback */
> > +	cmdq_thread_irq_handler(cmdq, thread);
> > +	if (list_empty(&thread->task_busy_list))
> > +		goto done;
> > +
> > +	list_for_each_entry_safe(task, tmp, &thread->task_busy_list,
> > +				 list_entry) {
> > +		cmdq_task_exec_done(task, -ECONNABORTED);
> 
> cmdq_task_exec_done(task, CMDQ_CB_ERROR) ? However, I've like to use the
> standard error as you write here.
> 

Ok, for consistent with current design, I will change to CMDQ_CB_ERROR.
And send another patch to change all cmdq error definition to standard
error next time.


Regards,
Dennis


> Regards,
> CK
> 
> > +		kfree(task);
> > +	}
> > +
> > +	cmdq_thread_disable(cmdq, thread);
> > +	clk_disable(cmdq->clock);
> > +done:
> > +	/*
> > +	 * The thread->task_busy_list empty means thread already disable. The
> > +	 * cmdq_mbox_send_data() always reset thread which clear disable and
> > +	 * suspend statue when first pkt send to channel, so there is no need
> > +	 * to do any operation here, only unlock and leave.
> > +	 */
> > +	spin_unlock_irqrestore(&thread->chan->lock, flags);
> >  }
> >  
> >  static const struct mbox_chan_ops cmdq_mbox_chan_ops = {
> 
>