Re: [PATCH v1 1/2] scsi: ufs: add required delay after gating reference clock

From: Stanley Chu <stanley.chu@mediatek.com>
To: Can Guo <cang@codeaurora.org>
Cc: <linux-scsi@vger.kernel.org>, <martin.petersen@oracle.com>,
	Asutosh Das <asutoshd@codeaurora.org>, <hongwus@codeaurora.org>,
	<avri.altman@wdc.com>, <alim.akhtar@samsung.com>,
	<jejb@linux.ibm.com>, <beanhuo@micron.com>,
	<matthias.bgg@gmail.com>, <bvanassche@acm.org>,
	<linux-mediatek@lists.infradead.org>,
	<linux-arm-kernel@lists.infradead.org>,
	<linux-kernel@vger.kernel.org>, <kuohong.wang@mediatek.com>,
	<peter.wang@mediatek.com>, <chun-hung.wu@mediatek.com>,
	<andy.teng@mediatek.com>
Subject: Re: [PATCH v1 1/2] scsi: ufs: add required delay after gating reference clock
Date: Thu, 20 Feb 2020 21:30:40 +0800	[thread overview]
Message-ID: <1582205440.26304.50.camel@mtksdccf07> (raw)
In-Reply-To: <bbb0b0637d9667d4691a9a28f9988dea@codeaurora.org>

Hi Can,

On Wed, 2020-02-19 at 18:33 +0800, Can Guo wrote:
> Hi Stanley,
> 
> On 2020-02-19 17:11, Stanley Chu wrote:
> > Hi Can,
> > 
> > On Wed, 2020-02-19 at 10:35 +0800, Can Guo wrote:
> > 
> >> Since we all need this delay here, how about put the delay in the
> >> entrence of ufshcd_setup_clocks(), before vops_setup_clocks()?
> >> If so, we can remove all the delays we added in our vops since the
> >> delay anyways delays everything inside ufshcd_setup_clocks().
> >> 
> > 
> > Always putting the delay in the entrance of ufshcd_setup_clocks() may
> > add unwanted delay for vendors, just like your current implementation,
> > or some other vendors who do not want to disable the reference clock.
> > 
> > I think current patch is more reasonable because the delay is applied 
> > to
> > clock only named as "ref_clk" specifically.
> > 
> > If you needs to keep "ref_clk" in DT, would you consider to remove the
> > delay in your ufs_qcom_dev_ref_clk_ctrl() and let the delay happens via
> > common ufshcd_setup_clocks() only? However you may still need delay if
> > call path comes from ufs_qcom_pwr_change_notify().
> > 
> > What do you think?
> > 
> 
> I agree current change is more reasonable from what it looks, but the 
> fact
> is that I canont remove the delay in ufs_qcom_dev_ref_clk_ctrl() even 
> with
> this change. On our platforms, ref_clk in DT serves multipule purposes,
> the ref_clk provided to UFS device is actually controlled in
> ufs_qcom_dev_ref_clk_ctrl(), which comes before where this change kicks 
> start,
> so if I remove the delay in ufs_qcom_dev_ref_clk_ctrl(), this change 
> cannot
> provide us the correct delay before gate the ref_clk provided to UFS 
> device.

> > Always putting the delay in the entrance of ufshcd_setup_clocks() may
> > add unwanted delay for vendors, just like your current implementation,
> > or some other vendors who do not want to disable the reference clock.
> 
> I meant if we put the delay in the entrance, I will be able to remove
> the delay in ufs_qcom_dev_ref_clk_ctrl(). Meanwhile, we can add proper
> checks before the delay to make sure it is initiated only if ref_clk 
> needs
> to be disabled, i.e:
> 
> if(!on && !skip_ref_clk && hba->dev_info.clk_gating_wait_us)
>      usleep_range();
> 
> Does this look better to you?

Firstly thanks so much for above details.

Again this statement may also add unwanted delay if some other vendors
does not have "ref_clk" in DT or they don't/can't disable the reference
clock provided to UFS device.

> 
> Anyways, we will see regressions with this change on our platforms, can 
> we
> have more discussions before get it merged? It should be OK if you go 
> with
> patch #2 alone first, right? Thanks.

Now the fact is that this change will impact your flow and it seems no
solid conclusion yet. Sure I could drop patch #1 and submit patch #2
only first : )

Thanks,
Stanley Chu