Re: [PATCH v4 5/5] scsi: ufs: ufs-mediatek: Add MCQ support for MTK platform

["Powen Kao (高伯文)" <Powen.Kao@mediatek.com>]

On Mon, 2023-03-13 at 14:36 -0700, Bart Van Assche wrote:
> On 3/6/23 22:54, Po-Wen Kao wrote:
> > +static unsigned int mtk_mcq_irq[UFSHCD_MAX_Q_NR];
> 
> Shouldn't there be one instance of this array per controller such
> that 
> this driver can support multiple host controllers instead of only
> one?
> 

True, I will fix the flow to get irq after `struct ufs_hba` is
allocated so that these infomation can be stored as per host instance. 

> > -	err = ufshcd_make_hba_operational(hba);
> > +	if (!hba->mcq_enabled) {
> > +		err = ufshcd_make_hba_operational(hba);
> > +	} else {
> > +		ufs_mtk_config_mcq(hba, false);
> > +		ufshcd_mcq_make_queues_operational(hba);
> > +		ufshcd_mcq_config_mac(hba, hba->nutrs);
> > +		ufshcd_writel(hba, ufshcd_readl(hba, REG_UFS_MEM_CFG) |
> > 0x1,
> > +			      REG_UFS_MEM_CFG);
> > +	}
> 
> ufshcd_config_mcq() in the UFSHCD core already calls 
> ufshcd_mcq_config_mac(). Why is there another call to 
> ufshcd_mcq_config_mac() in the MediaTek driver?

MCQ configuration will be reset through HCE cycle on our host
controller, hence we need to reconfigure those registers. 

> 
> > +	/*
> > +	 * Disable MCQ_CQ_EVENT interrupt.
> > +	 * Use CQ Tail Entry Push Status instead.
> > +	 */
> > +	ufshcd_disable_intr(hba, MCQ_CQ_EVENT_STATUS);
> 
> UFS host controller drivers should not call ufshcd_disable_intr().
> 
>  From the UFSHCI 4.0 specification: "MCQ CQ Event Status (CQES):
> This 
> bit is transparent and becomes ‘1’ when all of the following
> conditions 
> are met:
> • Controller is operating in MCQ mode (Config.QT=1)
> • ESI is not enabled (Config.ESIE=0)
> • CQES set only for Events in Queues that do not have interrupt 
> aggregation enabled or the Events that do not belong to
> MCQIACRy.IACTH 
> counter operation criteria.
> • At least one bit in CQISy is set and associated bit in CQIEy is
> set. 
> y=0..31"
> 
> Is there perhaps a bug in the MediaTek controller that causes the MCQ
> CQ 
> Event Status to be set in ESI mode? If not, can the above 
> ufshcd_disable_intr() call be left out?

We did not implement ESI at hardware level but per queue hw interrupt.
Without disabling MCQ_CQ_EVENT_STATUS, there will be two interrupts,
CQES (traditional interrupt) and CQ Tail Entry Push Interrupt (per
queue hw interrupt), raised on a signle command arrival.

I wouldn't consider it as a bug, but different interrupt design.
> 
> Thanks,
> 
> Bart.
> 

Thanks for your review

Powen