linux-kernel.vger.kernel.org archive mirror
 help / color / mirror / Atom feed
From: Can Guo <cang@codeaurora.org>
To: 황두현 <dh0421.hwang@samsung.com>
Cc: adrian.hunter@intel.com, alim.akhtar@samsung.com,
	asutoshd@codeaurora.org, avri.altman@wdc.com, beanhuo@micron.com,
	hongwus@codeaurora.org, jejb@linux.ibm.com,
	kernel-team@android.com, linux-kernel@vger.kernel.org,
	linux-scsi@vger.kernel.org, martin.petersen@oracle.com,
	nguyenb@codeaurora.org, rnayak@codeaurora.org,
	salyzyn@google.com, saravanak@google.com, satyat@google.com,
	stanley.chu@mediatek.com, grant.jung@samsung.com,
	jt77.jang@samsung.com, junwoo80.lee@samsung.com,
	sh043.lee@samsung.com, jangsub.yi@samsung.com,
	cw9316.lee@samsung.com, sh8267.baek@samsung.com,
	wkon.kim@samsung.com
Subject: Re: [PATCH v9 1/3] scsi: ufs: Protect some contexts from unexpected clock scaling
Date: Wed, 13 Jan 2021 18:49:25 +0800	[thread overview]
Message-ID: <eeb48b97d094d0240813f6c7d6953d54@codeaurora.org> (raw)
In-Reply-To: <00e401d6e994$e62e1660$b28a4320$@samsung.com>

On 2021-01-13 18:14, 황두현 wrote:
>> From: Can Guo <cang@codeaurora.org>
>> 
>> In contexts like suspend, shutdown and error handling, we need to 
>> suspend
>> devfreq to make sure these contexts won't be disturbed by clock 
>> scaling.
>> However, suspending devfreq is not enough since users can still 
>> trigger a
>> clock scaling by manipulating the devfreq sysfs nodes like 
>> min/max_freq and
>> governor even after devfreq is suspended. Moreover, mere suspending 
>> devfreq
>> cannot synchroinze a clock scaling which has already been invoked 
>> through
>> these sysfs nodes. Add one more flag in struct clk_scaling and wrap 
>> the
>> entire func ufshcd_devfreq_scale() with the clk_scaling_lock, so that 
>> we
>> can use this flag and clk_scaling_lock to control and synchronize 
>> clock
>> scaling invoked through devfreq sysfs nodes.
>> 
>> Signed-off-by: Can Guo <cang@codeaurora.org>
>> ---
>>  drivers/scsi/ufs/ufshcd.c | 86 
>> +++++++++++++++++++++++++++++------------------
>>  drivers/scsi/ufs/ufshcd.h |  6 +++-
>>  2 files changed, 59 insertions(+), 33 deletions(-)
>> 
>> diff --git a/drivers/scsi/ufs/ufshcd.c b/drivers/scsi/ufs/ufshcd.c
>> index 82ad317..64218e6 100644
>> --- a/drivers/scsi/ufs/ufshcd.c
>> +++ b/drivers/scsi/ufs/ufshcd.c
>> @@ -1181,19 +1181,33 @@ static int ufshcd_clock_scaling_prepare(struct 
>> ufs_hba *hba)
>>  	 */
>>  	ufshcd_scsi_block_requests(hba);
>>  	down_write(&hba->clk_scaling_lock);
>> -	if (ufshcd_wait_for_doorbell_clr(hba, DOORBELL_CLR_TOUT_US)) {
>> +
> 
> I think another value is better than -EAGAIN.
> Ex) -EPERM because the clk scaling is not allowed state.

-EPERM is usually used by IOCTL to tell users that
they don't have right permissions. -EAGAIN here tells
the caller that if it tries again, it may success, i.e
after hba is runtime resumed or error recovery is completed.

> 
>> +	if (!hba->clk_scaling.is_allowed)
>> +		ret = -EAGAIN;
> 
> I think ufshcd_hold() has to be called before access host registers,
> so calling ufshcd_hold() is better before calling
> ufshcd_clock_scaling_prepare().
> 
> If you want to do not anything when hba->clk_scaling.is_allowed is 
> false,
> it is recommended to check hba->clk_scaling.is_allowed before calling
> ufshcd_hold() in ufshcd_devfreq_scale().

Check ufshcd_wait_for_doorbell_clr(), it calls ufshcd_hold/release() 
inside.

> 
>> +	else if (ufshcd_wait_for_doorbell_clr(hba, DOORBELL_CLR_TOUT_US))
>>  		ret = -EBUSY;
>> +
>> +	if (ret) {
>>  		up_write(&hba->clk_scaling_lock);
>>  		ufshcd_scsi_unblock_requests(hba);
>> +		goto out;
>>  	}
>> 
>> +	/* let's not get into low power until clock scaling is completed */
>> +	ufshcd_hold(hba, false);
>> +
>> +out:
>>  	return ret;
>>  }
>> 
>> -static void ufshcd_clock_scaling_unprepare(struct ufs_hba *hba)
>> +static void ufshcd_clock_scaling_unprepare(struct ufs_hba *hba, bool 
>> writelock)
>>  {
>> -	up_write(&hba->clk_scaling_lock);
>> +	if (writelock)
>> +		up_write(&hba->clk_scaling_lock);
>> +	else
>> +		up_read(&hba->clk_scaling_lock);
>>  	ufshcd_scsi_unblock_requests(hba);
>> +	ufshcd_release(hba);
>>  }
>> 
>>  /**
>> @@ -1208,13 +1222,11 @@ static void 
>> ufshcd_clock_scaling_unprepare(struct ufs_hba *hba)
>>  static int ufshcd_devfreq_scale(struct ufs_hba *hba, bool scale_up)
>>  {
>>  	int ret = 0;
>> -
>> -	/* let's not get into low power until clock scaling is completed */
>> -	ufshcd_hold(hba, false);
>> +	bool is_writelock = true;
>> 
> 
> I think it would be nice to call ufshcd_hold() here.

No, if hba is runtime suspended (rpm_lvl = 3/4/5),
calling ufshcd_hold() here shall fail, because you
are trying to exit hibern8 even when UFS is not
powered or VCCQ2/VCCQ in LPM.

So ufshcd_hold MUST come after hba->clk_scaling.is_allowed
check in prepare(), and the code is clean and tight if
ufshcd_hold() is put in prepare().

> 
>>  	ret = ufshcd_clock_scaling_prepare(hba);
>>  	if (ret)
> 
> Below change is not needed if calling ufshcd_hold() is before calling
> ufshcd_clock_scaling_prepare().
> 

See my comments above.

>> -		goto out;
>> +		return ret;
>> 
>>  	/* scale down the gear before scaling down clocks */
>>  	if (!scale_up) {
>> @@ -1240,14 +1252,12 @@ static int ufshcd_devfreq_scale(struct ufs_hba 
>> *hba, bool scale_up)
>>  	}
>> 
>>  	/* Enable Write Booster if we have scaled up else disable it */
>> -	up_write(&hba->clk_scaling_lock);
>> +	downgrade_write(&hba->clk_scaling_lock);
>> +	is_writelock = false;
>>  	ufshcd_wb_ctrl(hba, scale_up);
>> -	down_write(&hba->clk_scaling_lock);
>> 
>>  out_unprepare:
>> -	ufshcd_clock_scaling_unprepare(hba);
>> -out:
>> -	ufshcd_release(hba);
>> +	ufshcd_clock_scaling_unprepare(hba, is_writelock);
>>  	return ret;
>>  }
>> 
>> @@ -1521,7 +1531,7 @@ static ssize_t 
>> ufshcd_clkscale_enable_show(struct device *dev,
>>  {
>>  	struct ufs_hba *hba = dev_get_drvdata(dev);
>> 
>> -	return snprintf(buf, PAGE_SIZE, "%d\n", 
>> hba->clk_scaling.is_allowed);
>> +	return snprintf(buf, PAGE_SIZE, "%d\n", 
>> hba->clk_scaling.is_enabled);
>>  }
>> 
>>  static ssize_t ufshcd_clkscale_enable_store(struct device *dev,
>> @@ -1535,7 +1545,7 @@ static ssize_t 
>> ufshcd_clkscale_enable_store(struct device *dev,
>>  		return -EINVAL;
>> 
>>  	value = !!value;
>> -	if (value == hba->clk_scaling.is_allowed)
>> +	if (value == hba->clk_scaling.is_enabled)
>>  		goto out;
>> 
>>  	pm_runtime_get_sync(hba->dev);
>> @@ -1544,7 +1554,7 @@ static ssize_t 
>> ufshcd_clkscale_enable_store(struct device *dev,
>>  	cancel_work_sync(&hba->clk_scaling.suspend_work);
>>  	cancel_work_sync(&hba->clk_scaling.resume_work);
>> 
>> -	hba->clk_scaling.is_allowed = value;
>> +	hba->clk_scaling.is_enabled = value;
>> 
>>  	if (value) {
>>  		ufshcd_resume_clkscaling(hba);
>> @@ -1882,8 +1892,6 @@ static void ufshcd_init_clk_scaling(struct 
>> ufs_hba *hba)
>>  	snprintf(wq_name, sizeof(wq_name), "ufs_clkscaling_%d",
>>  		 hba->host->host_no);
>>  	hba->clk_scaling.workq = create_singlethread_workqueue(wq_name);
>> -
>> -	ufshcd_clkscaling_init_sysfs(hba);
>>  }
>> 
>>  static void ufshcd_exit_clk_scaling(struct ufs_hba *hba)
>> @@ -1891,6 +1899,8 @@ static void ufshcd_exit_clk_scaling(struct 
>> ufs_hba *hba)
>>  	if (!ufshcd_is_clkscaling_supported(hba))
>>  		return;
>> 
>> +	if (hba->clk_scaling.enable_attr.attr.name)
>> +		device_remove_file(hba->dev, &hba->clk_scaling.enable_attr);
>>  	destroy_workqueue(hba->clk_scaling.workq);
>>  	ufshcd_devfreq_remove(hba);
>>  }
>> @@ -1955,7 +1965,7 @@ static void ufshcd_clk_scaling_start_busy(struct 
>> ufs_hba *hba)
>>  	if (!hba->clk_scaling.active_reqs++)
>>  		queue_resume_work = true;
>> 
>> -	if (!hba->clk_scaling.is_allowed || hba->pm_op_in_progress)
>> +	if (!hba->clk_scaling.is_enabled || hba->pm_op_in_progress)
>>  		return;
>> 
>>  	if (queue_resume_work)
>> @@ -5071,7 +5081,8 @@ static void __ufshcd_transfer_req_compl(struct 
>> ufs_hba *hba,
>>  				update_scaling = true;
>>  			}
>>  		}
>> -		if (ufshcd_is_clkscaling_supported(hba) && update_scaling)
>> +		if (ufshcd_is_clkscaling_supported(hba) && update_scaling &&
>> +		    hba->clk_scaling.active_reqs > 0)
>>  			hba->clk_scaling.active_reqs--;
>>  	}
>> 
>> @@ -5737,18 +5748,24 @@ static void ufshcd_err_handling_prepare(struct 
>> ufs_hba *hba)
>>  		ufshcd_vops_resume(hba, pm_op);
>>  	} else {
>>  		ufshcd_hold(hba, false);
>> -		if (hba->clk_scaling.is_allowed) {
>> +		if (hba->clk_scaling.is_enabled) {
>>  			cancel_work_sync(&hba->clk_scaling.suspend_work);
>>  			cancel_work_sync(&hba->clk_scaling.resume_work);
>>  			ufshcd_suspend_clkscaling(hba);
>>  		}
>> +		down_write(&hba->clk_scaling_lock);
>> +		hba->clk_scaling.is_allowed = false;
>> +		up_write(&hba->clk_scaling_lock);
>>  	}
>>  }
>> 
>>  static void ufshcd_err_handling_unprepare(struct ufs_hba *hba)
>>  {
>>  	ufshcd_release(hba);
>> -	if (hba->clk_scaling.is_allowed)
>> +	down_write(&hba->clk_scaling_lock);
>> +	hba->clk_scaling.is_allowed = true;
>> +	up_write(&hba->clk_scaling_lock);
>> +	if (hba->clk_scaling.is_enabled)
>>  		ufshcd_resume_clkscaling(hba);
>>  	pm_runtime_put(hba->dev);
>>  }
>> @@ -7728,12 +7745,14 @@ static int ufshcd_add_lus(struct ufs_hba *hba)
>>  			sizeof(struct ufs_pa_layer_attr));
>>  		hba->clk_scaling.saved_pwr_info.is_valid = true;
>>  		if (!hba->devfreq) {
>> +			hba->clk_scaling.is_allowed = true;
>>  			ret = ufshcd_devfreq_init(hba);
>>  			if (ret)
>>  				goto out;
>> -		}
>> 
>> -		hba->clk_scaling.is_allowed = true;
>> +			hba->clk_scaling.is_enabled = true;
>> +			ufshcd_clkscaling_init_sysfs(hba);
>> +		}
>>  	}
>> 
>>  	ufs_bsg_probe(hba);
>> @@ -8650,11 +8669,14 @@ static int ufshcd_suspend(struct ufs_hba *hba, 
>> enum ufs_pm_op pm_op)
>>  	ufshcd_hold(hba, false);
>>  	hba->clk_gating.is_suspended = true;
>> 
>> -	if (hba->clk_scaling.is_allowed) {
>> +	if (hba->clk_scaling.is_enabled) {
>>  		cancel_work_sync(&hba->clk_scaling.suspend_work);
>>  		cancel_work_sync(&hba->clk_scaling.resume_work);
>>  		ufshcd_suspend_clkscaling(hba);
>>  	}
>> +	down_write(&hba->clk_scaling_lock);
>> +	hba->clk_scaling.is_allowed = false;
>> +	up_write(&hba->clk_scaling_lock);
>> 
>>  	if (req_dev_pwr_mode == UFS_ACTIVE_PWR_MODE &&
>>  			req_link_state == UIC_LINK_ACTIVE_STATE) {
>> @@ -8751,8 +8773,6 @@ static int ufshcd_suspend(struct ufs_hba *hba, 
>> enum ufs_pm_op pm_op)
>>  	goto out;
>> 
>>  set_link_active:
>> -	if (hba->clk_scaling.is_allowed)
>> -		ufshcd_resume_clkscaling(hba);
>>  	ufshcd_vreg_set_hpm(hba);
>>  	/*
>>  	 * Device hardware reset is required to exit DeepSleep. Also, for
>> @@ -8776,7 +8796,10 @@ static int ufshcd_suspend(struct ufs_hba *hba, 
>> enum ufs_pm_op pm_op)
>>  	if (!ufshcd_set_dev_pwr_mode(hba, UFS_ACTIVE_PWR_MODE))
>>  		ufshcd_disable_auto_bkops(hba);
>>  enable_gating:
>> -	if (hba->clk_scaling.is_allowed)
>> +	down_write(&hba->clk_scaling_lock);
>> +	hba->clk_scaling.is_allowed = true;
>> +	up_write(&hba->clk_scaling_lock);
>> +	if (hba->clk_scaling.is_enabled)
>>  		ufshcd_resume_clkscaling(hba);
>>  	hba->clk_gating.is_suspended = false;
>>  	hba->dev_info.b_rpm_dev_flush_capable = false;
>> @@ -8879,7 +8902,10 @@ static int ufshcd_resume(struct ufs_hba *hba, 
>> enum ufs_pm_op pm_op)
>> 
>>  	hba->clk_gating.is_suspended = false;
>> 
>> -	if (hba->clk_scaling.is_allowed)
>> +	down_write(&hba->clk_scaling_lock);
>> +	hba->clk_scaling.is_allowed = true;
>> +	up_write(&hba->clk_scaling_lock);
>> +	if (hba->clk_scaling.is_enabled)
>>  		ufshcd_resume_clkscaling(hba);
>> 
>>  	/* Enable Auto-Hibernate if configured */
>> @@ -8903,8 +8929,6 @@ static int ufshcd_resume(struct ufs_hba *hba, 
>> enum ufs_pm_op pm_op)
>>  	ufshcd_vreg_set_lpm(hba);
>>  disable_irq_and_vops_clks:
>>  	ufshcd_disable_irq(hba);
>> -	if (hba->clk_scaling.is_allowed)
>> -		ufshcd_suspend_clkscaling(hba);
>>  	ufshcd_setup_clocks(hba, false);
>>  	if (ufshcd_is_clkgating_allowed(hba)) {
>>  		hba->clk_gating.state = CLKS_OFF;
>> @@ -9131,8 +9155,6 @@ void ufshcd_remove(struct ufs_hba *hba)
>> 
>>  	ufshcd_exit_clk_scaling(hba);
>>  	ufshcd_exit_clk_gating(hba);
>> -	if (ufshcd_is_clkscaling_supported(hba))
>> -		device_remove_file(hba->dev, &hba->clk_scaling.enable_attr);
>>  	ufshcd_hba_exit(hba);
>>  }
>>  EXPORT_SYMBOL_GPL(ufshcd_remove);
>> diff --git a/drivers/scsi/ufs/ufshcd.h b/drivers/scsi/ufs/ufshcd.h
>> index aa9ea35..2863af1 100644
>> --- a/drivers/scsi/ufs/ufshcd.h
>> +++ b/drivers/scsi/ufs/ufshcd.h
>> @@ -419,7 +419,10 @@ struct ufs_saved_pwr_info {
>>   * @suspend_work: worker to suspend devfreq
>>   * @resume_work: worker to resume devfreq
>>   * @min_gear: lowest HS gear to scale down to
>> - * @is_allowed: tracks if scaling is currently allowed or not
>> + * @is_enabled: tracks if scaling is currently enabled or not, 
>> controlled by
>> +		clkscale_enable sysfs node
>> + * @is_allowed: tracks if scaling is currently allowed or not, used 
>> to block
>> +		clock scaling which is not invoked from devfreq governor
>>   * @is_busy_started: tracks if busy period has started or not
>>   * @is_suspended: tracks if devfreq is suspended or not
>>   */
>> @@ -434,6 +437,7 @@ struct ufs_clk_scaling {
>>  	struct work_struct suspend_work;
>>  	struct work_struct resume_work;
>>  	u32 min_gear;
>> +	bool is_enabled;
>>  	bool is_allowed;
>>  	bool is_busy_started;
>>  	bool is_suspended;
>> --
>> Qualcomm Innovation Center, Inc. is a member of Code Aurora Forum, a 
>> Linux Foundation Collaborative Project.
> 
> --
> DooHyun Hwang
> Samsung Electronics

  reply	other threads:[~2021-01-13 10:50 UTC|newest]

Thread overview: 3+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
     [not found] <CGME20210113101438epcas1p21192b49fa1a079353ab36555478f32d8@epcas1p2.samsung.com>
2021-01-13 10:14 ` [PATCH v9 1/3] scsi: ufs: Protect some contexts from unexpected clock scaling 황두현
2021-01-13 10:49   ` Can Guo [this message]
     [not found] <1609909486-21053-1-git-send-email-cang@codeaurora.org>
2021-01-06  5:04 ` Can Guo

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=eeb48b97d094d0240813f6c7d6953d54@codeaurora.org \
    --to=cang@codeaurora.org \
    --cc=adrian.hunter@intel.com \
    --cc=alim.akhtar@samsung.com \
    --cc=asutoshd@codeaurora.org \
    --cc=avri.altman@wdc.com \
    --cc=beanhuo@micron.com \
    --cc=cw9316.lee@samsung.com \
    --cc=dh0421.hwang@samsung.com \
    --cc=grant.jung@samsung.com \
    --cc=hongwus@codeaurora.org \
    --cc=jangsub.yi@samsung.com \
    --cc=jejb@linux.ibm.com \
    --cc=jt77.jang@samsung.com \
    --cc=junwoo80.lee@samsung.com \
    --cc=kernel-team@android.com \
    --cc=linux-kernel@vger.kernel.org \
    --cc=linux-scsi@vger.kernel.org \
    --cc=martin.petersen@oracle.com \
    --cc=nguyenb@codeaurora.org \
    --cc=rnayak@codeaurora.org \
    --cc=salyzyn@google.com \
    --cc=saravanak@google.com \
    --cc=satyat@google.com \
    --cc=sh043.lee@samsung.com \
    --cc=sh8267.baek@samsung.com \
    --cc=stanley.chu@mediatek.com \
    --cc=wkon.kim@samsung.com \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).