Re: [PATCH v1 4/4] scsi: ufs: Fix WriteBooster flush during runtime suspend

["Asutosh Das (asd)" <asutoshd@codeaurora.org>]

On 5/14/2020 7:49 AM, Stanley Chu wrote:
> Hi Asutosh,
> 
> On Thu, 2020-05-14 at 10:23 +0800, Stanley Chu wrote:
>> Hi Asutosh,
>>
>> On Wed, 2020-05-13 at 12:31 -0700, Asutosh Das (asd) wrote:
>>> On 5/12/2020 3:47 AM, Stanley Chu wrote:
>>>> Currently UFS host driver promises VCC supply if UFS device
>>>> needs to do WriteBooster flush during runtime suspend.
>>>>
>>>> However the UFS specification mentions,
>>>>
>>>> "While the flushing operation is in progress, the device is
>>>> in Active power mode."
>>>>
>>>> Therefore UFS host driver needs to promise more: Keep UFS
>>>> device as "Active power mode", otherwise UFS device shall not
>>>> do any flush if device enters Sleep or PowerDown power mode.
>>>>
>>>> Fix this by not changing device power mode if WriteBooster
>>>> flush is required in ufshcd_suspend().
>>>>
>>>> Signed-off-by: Stanley Chu <stanley.chu@mediatek.com>
>>>> ---
>>>>    drivers/scsi/ufs/ufs.h    |  1 -
>>>>    drivers/scsi/ufs/ufshcd.c | 39 +++++++++++++++++++--------------------
>>>>    2 files changed, 19 insertions(+), 21 deletions(-)
>>>>
>>>> diff --git a/drivers/scsi/ufs/ufs.h b/drivers/scsi/ufs/ufs.h
>>>> index b3135344ab3f..9e4bc2e97ada 100644
>>>> --- a/drivers/scsi/ufs/ufs.h
>>>> +++ b/drivers/scsi/ufs/ufs.h
>>>> @@ -577,7 +577,6 @@ struct ufs_dev_info {
>>>>    	u32 d_ext_ufs_feature_sup;
>>>>    	u8 b_wb_buffer_type;
>>>>    	u32 d_wb_alloc_units;
>>>> -	bool keep_vcc_on;
>>>>    	u8 b_presrv_uspc_en;
>>>>    };
>>>>    
>>>> diff --git a/drivers/scsi/ufs/ufshcd.c b/drivers/scsi/ufs/ufshcd.c
>>>> index 169a3379e468..2d0aff8ac260 100644
>>>> --- a/drivers/scsi/ufs/ufshcd.c
>>>> +++ b/drivers/scsi/ufs/ufshcd.c
>>>> @@ -8101,8 +8101,7 @@ static void ufshcd_vreg_set_lpm(struct ufs_hba *hba)
>>>>    	    !hba->dev_info.is_lu_power_on_wp) {
>>>>    		ufshcd_setup_vreg(hba, false);
>>>>    	} else if (!ufshcd_is_ufs_dev_active(hba)) {
>>>> -		if (!hba->dev_info.keep_vcc_on)
>>>> -			ufshcd_toggle_vreg(hba->dev, hba->vreg_info.vcc, false);
>>>> +		ufshcd_toggle_vreg(hba->dev, hba->vreg_info.vcc, false);
>>>>    		if (!ufshcd_is_link_active(hba)) {
>>>>    			ufshcd_config_vreg_lpm(hba, hba->vreg_info.vccq);
>>>>    			ufshcd_config_vreg_lpm(hba, hba->vreg_info.vccq2);
>>>> @@ -8172,6 +8171,7 @@ static int ufshcd_suspend(struct ufs_hba *hba, enum ufs_pm_op pm_op)
>>>>    	enum ufs_pm_level pm_lvl;
>>>>    	enum ufs_dev_pwr_mode req_dev_pwr_mode;
>>>>    	enum uic_link_state req_link_state;
>>>> +	bool keep_curr_dev_pwr_mode = false;
>>>>    
>>>>    	hba->pm_op_in_progress = 1;
>>>>    	if (!ufshcd_is_shutdown_pm(pm_op)) {
>>>> @@ -8226,28 +8226,27 @@ static int ufshcd_suspend(struct ufs_hba *hba, enum ufs_pm_op pm_op)
>>>>    			/* make sure that auto bkops is disabled */
>>>>    			ufshcd_disable_auto_bkops(hba);
>>>>    		}
>>>> +
>>> Unnecessary newline, perhaps?
>>
>> Yap, I will remove it in next version.
>>
>>>>    		/*
>>>> -		 * With wb enabled, if the bkops is enabled or if the
>>>> -		 * configured WB type is 70% full, keep vcc ON
>>>> -		 * for the device to flush the wb buffer
>>>> +		 * If device needs to do BKOP or WB buffer flush, keep device
>>>> +		 * power mode as "active power mode" and its VCC supply.
>>>>    		 */
>>>> -		if ((hba->auto_bkops_enabled && ufshcd_is_wb_allowed(hba)) ||
>>>> -		    ufshcd_wb_keep_vcc_on(hba))
>>>> -			hba->dev_info.keep_vcc_on = true;
>>>> -		else
>>>> -			hba->dev_info.keep_vcc_on = false;
>>>> -	} else {
>>>> -		hba->dev_info.keep_vcc_on = false;
>>>> +		keep_curr_dev_pwr_mode = hba->auto_bkops_enabled ||
>>>> +			ufshcd_wb_keep_vcc_on(hba);
>>> Should the device be in UFS_ACTIVE_PWR_MODE to perform auto-bkops?
>>>
>>> Also, is it needed to keep the device in UFS_ACTIVE_PWR_MODE , if flush
>>> on hibern8 is enabled and the link is being put to hibern8 mode during
>>> runtime-suspend? Perhaps that should also be factored in here?
>>
>> Both auto-bkops and WriteBooster flush during Hibern8 need device power
>> mode to be "Active Power Mode".
>>
>> For auto-bkops, the spec mentions,
>>
>> "If the background operations enable bit is set and the device is in
>> Active power mode or Idle power mode, then the device is allowed to
>> execute any internal operations."
>>
>> For WriteBooster flush during Hibern8, the spec mentions,
>>
>> "While the flushing operation is in progress, the device is in Active
>> power mode."
>>
>> Therefore here we can use an unified "keep_curr_dev_pwr_mode" to
>> indicate the same requirements of above both features.
>>
>> Besides, both operations may access flash array inside UFS device thus
>> VCC supply shall be also kept.
>>
>> Before this patch, the original code will keep device power mode (stay
>> in Active Power Mode) if hba->auto_bkops_enabled is set as true during
>> runtime-suspend with UFSHCD_CAP_AUTO_BKOPS_SUSPEND capability is
>> enabled. This patch will not change this decision, just add
>> "WriteBooster flush during Hibern8" feature as another condition to do
>> so.
>>
>> Thank you so much to remind me that "Link shall be put in Hibern8" is a
>> necessary condition for "WriteBooster flush during Hibern8". I will add
>> more checking for keep_curr_dev_pwr_mode to prevent unnecessary power
>> drain.
>>
>>>>    	}
>>>>    
>>>> -	if ((req_dev_pwr_mode != hba->curr_dev_pwr_mode) &&
>>>> -	    ((ufshcd_is_runtime_pm(pm_op) && !hba->auto_bkops_enabled) ||
>>>> -	    !ufshcd_is_runtime_pm(pm_op))) {
>>>> -		/* ensure that bkops is disabled */
>>>> -		ufshcd_disable_auto_bkops(hba);
>>>> -		ret = ufshcd_set_dev_pwr_mode(hba, req_dev_pwr_mode);
>>>> -		if (ret)
>>>> -			goto enable_gating;
>>>> +	if (req_dev_pwr_mode != hba->curr_dev_pwr_mode) {
>>>> +		if ((ufshcd_is_runtime_pm(pm_op) && !hba->auto_bkops_enabled) ||
>>>> +		    !ufshcd_is_runtime_pm(pm_op)) {
>>>> +			/* ensure that bkops is disabled */
>>>> +			ufshcd_disable_auto_bkops(hba);
>>>> +		}
>>>> +
>>>> +		if (!keep_curr_dev_pwr_mode) {
>>>> +			ret = ufshcd_set_dev_pwr_mode(hba, req_dev_pwr_mode);
>>>
>>> Now, when the WB buffer is completely flushed out, the device should be
>>> put back into UFS_SLEEP_PWR_MODE or UFS_POWERDOWN_PWR_MODE. Say, the
>>> device buffer has to be flushed and during runtime-suspend, the device
>>> is put to UFS_ACTIVE_PWR_MODE and Vcc is kept ON; the device doesn't
>>> resume nor does the system enters suspend for a very long time, and with
>>> AH8 and hibern8 disabled, there will be an unnecessary power drain for
>>> that much time.
> 
> Another thought is that if keep_curr_dev_pwr_mode will be set as true
> only if link is put in Hibern8 or Auto-Hibern8 is enabled. By this way,
> the power consumption shall be very small after flush or auto-bkop is
> finished.
> 
> Then the checking of flush status during runtime-suspend may be not
> necessary.
> 
>>>
>>> How about a periodic interval checking of flush status if
>>> keep_curr_dev_pwr_mode evaluates to be true?
>>
>> This is a good point!
>>
>> The same thing also happens for auto-bkops. How about add a timer to
>> leave runtime suspend if keep_curr_dev_pwr_mode is set as true? This is
>> simple and also favors power. The timeout value could be adjustable
>> according to the available WriteBooster buffer size.
>>
>> A periodic interval checking of flush status needs to re-activate link
>> to communicate with the device. This would be tricky and the
>> re-activation flow is just like runtime-resume.
>>
>> What would you think?
>>
>> Thanks.
>> Stanley Chu
>>
>>
>> _______________________________________________
>> Linux-mediatek mailing list
>> Linux-mediatek@lists.infradead.org
>> http://lists.infradead.org/mailman/listinfo/linux-mediatek
> 

Hi Stanley,
I think that'd work, but there's definitely a penalty of keeping Vcc ON.
And if we do want to keep it ON, then we'd have to measure how much 
excess power is being used - after the flush is done.

I think setting keep_curr_dev_pwr_mode to true iff h8 and ah8 are 
enabled is a good idea. In addition to that, adding a timer to check 
flush status if keep_curr_dev_pwr_mode is set to true would keep the 
power consumption to a minimum. So I suggest to have the delayed check 
of flush status as well.

Thanks,
-asd

-- 
The Qualcomm Innovation Center, Inc. is a member of the Code Aurora Forum,
Linux Foundation Collaborative Project