diff mbox series

[v4,2/3] scsi: ufs: core: Introduce a new clock_gating lock

Message ID 20241118144117.88483-3-avri.altman@wdc.com
State Superseded
Headers show
Series Untie the host lock entanglement - part 2 | expand

Commit Message

Avri Altman Nov. 18, 2024, 2:41 p.m. UTC 
Introduce a new clock gating lock to serialize access to some of the
clock gating members instead of the host_lock.

While at it, simplify the code with the guard() macro and co for
automatic cleanup of the new lock. There are some explicit
spin_lock_irqsave/spin_unlock_irqrestore snaking instances I left behind
because I couldn't make heads or tails of it.

Additionally, move the trace_ufshcd_clk_gating() call from inside the
region protected by the lock as it doesn't needs protection.

Signed-off-by: Avri Altman <avri.altman@wdc.com>
---
 drivers/ufs/core/ufshcd.c | 109 ++++++++++++++++++++------------------
 include/ufs/ufshcd.h      |   8 ++-
 2 files changed, 62 insertions(+), 55 deletions(-)

Comments

Bart Van Assche Nov. 21, 2024, 8:44 p.m. UTC | #1 
On 11/18/24 6:41 AM, Avri Altman wrote:
> diff --git a/drivers/ufs/core/ufshcd.c b/drivers/ufs/core/ufshcd.c
> index be5fe2407382..638d9c0e2603 100644
> --- a/drivers/ufs/core/ufshcd.c
> +++ b/drivers/ufs/core/ufshcd.c
> @@ -1816,19 +1816,17 @@ static void ufshcd_exit_clk_scaling(struct ufs_hba *hba)
>   static void ufshcd_ungate_work(struct work_struct *work)
>   {
>   	int ret;
> -	unsigned long flags;
>   	struct ufs_hba *hba = container_of(work, struct ufs_hba,
>   			clk_gating.ungate_work);
>   
>   	cancel_delayed_work_sync(&hba->clk_gating.gate_work);
>   
> -	spin_lock_irqsave(hba->host->host_lock, flags);
> -	if (hba->clk_gating.state == CLKS_ON) {
> -		spin_unlock_irqrestore(hba->host->host_lock, flags);
> -		return;
> +	scoped_guard(spinlock_irqsave, &hba->clk_gating.lock)
> +	{
> +		if (hba->clk_gating.state == CLKS_ON)
> +			return;
>   	}

Here and elsewhere, please move "{" to the end of the "scoped_guard()"
line since that is the style used in all other Linux kernel code (I know
that clang-format gets this wrong).

>   /* host lock must be held before calling this variant */

Please remove this comment since your patch makes it incorrect and
replace it with a lockdep_assert_held() call.

> +	spin_lock_irqsave(hba->host->host_lock, flags);
> +	if (ufshcd_has_pending_tasks(hba) ||
> +	    hba->ufshcd_state != UFSHCD_STATE_OPERATIONAL) {
> +		spin_unlock_irqrestore(hba->host->host_lock, flags);
> +		return;
> +	}
> +	spin_unlock_irqrestore(hba->host->host_lock, flags);

Why explicit lock/unlock calls instead of using scoped_guard()?

> diff --git a/include/ufs/ufshcd.h b/include/ufs/ufshcd.h
> index d7aca9e61684..8f9997b0dbf9 100644
> --- a/include/ufs/ufshcd.h
> +++ b/include/ufs/ufshcd.h
> @@ -403,6 +403,8 @@ enum clk_gating_state {
>    * delay_ms
>    * @ungate_work: worker to turn on clocks that will be used in case of
>    * interrupt context
> + * @clk_gating_workq: workqueue for clock gating work.
> + * @lock: serialize access to some struct ufs_clk_gating members

Please document that @lock is the outer lock relative to the host lock.

Thanks,

Bart.
Avri Altman Nov. 21, 2024, 9:06 p.m. UTC | #2 
> On 11/18/24 6:41 AM, Avri Altman wrote:
> > diff --git a/drivers/ufs/core/ufshcd.c b/drivers/ufs/core/ufshcd.c
> > index be5fe2407382..638d9c0e2603 100644
> > --- a/drivers/ufs/core/ufshcd.c
> > +++ b/drivers/ufs/core/ufshcd.c
> > @@ -1816,19 +1816,17 @@ static void ufshcd_exit_clk_scaling(struct
> ufs_hba *hba)
> >   static void ufshcd_ungate_work(struct work_struct *work)
> >   {
> >       int ret;
> > -     unsigned long flags;
> >       struct ufs_hba *hba = container_of(work, struct ufs_hba,
> >                       clk_gating.ungate_work);
> >
> >       cancel_delayed_work_sync(&hba->clk_gating.gate_work);
> >
> > -     spin_lock_irqsave(hba->host->host_lock, flags);
> > -     if (hba->clk_gating.state == CLKS_ON) {
> > -             spin_unlock_irqrestore(hba->host->host_lock, flags);
> > -             return;
> > +     scoped_guard(spinlock_irqsave, &hba->clk_gating.lock)
> > +     {
> > +             if (hba->clk_gating.state == CLKS_ON)
> > +                     return;
> >       }
> 
> Here and elsewhere, please move "{" to the end of the "scoped_guard()"
> line since that is the style used in all other Linux kernel code (I know that
> clang-format gets this wrong).
Yeah - I was running clang-format.
Done.

Thanks,
Avri

> 
> >   /* host lock must be held before calling this variant */
> 
> Please remove this comment since your patch makes it incorrect and replace
> it with a lockdep_assert_held() call.
Done.

> 
> > +     spin_lock_irqsave(hba->host->host_lock, flags);
> > +     if (ufshcd_has_pending_tasks(hba) ||
> > +         hba->ufshcd_state != UFSHCD_STATE_OPERATIONAL) {
> > +             spin_unlock_irqrestore(hba->host->host_lock, flags);
> > +             return;
> > +     }
> > +     spin_unlock_irqrestore(hba->host->host_lock, flags);
> 
> Why explicit lock/unlock calls instead of using scoped_guard()?
Should I apply those to host_lock as well?
I find it a bit confusing because in this change using guard et al. is limited to the new locks only. 

> 
> > diff --git a/include/ufs/ufshcd.h b/include/ufs/ufshcd.h index
> > d7aca9e61684..8f9997b0dbf9 100644
> > --- a/include/ufs/ufshcd.h
> > +++ b/include/ufs/ufshcd.h
> > @@ -403,6 +403,8 @@ enum clk_gating_state {
> >    * delay_ms
> >    * @ungate_work: worker to turn on clocks that will be used in case of
> >    * interrupt context
> > + * @clk_gating_workq: workqueue for clock gating work.
> > + * @lock: serialize access to some struct ufs_clk_gating members
> 
> Please document that @lock is the outer lock relative to the host lock.
Not sure what you mean?
host_lock is nested in one place only, should this goes to the @lock documentation?

Thanks,
Avri
> 
> Thanks,
> 
> Bart.
Bart Van Assche Nov. 21, 2024, 9:14 p.m. UTC | #3 
On 11/21/24 1:06 PM, Avri Altman wrote:
>> On 11/18/24 6:41 AM, Avri Altman wrote:
>>> +     spin_lock_irqsave(hba->host->host_lock, flags);
>>> +     if (ufshcd_has_pending_tasks(hba) ||
>>> +         hba->ufshcd_state != UFSHCD_STATE_OPERATIONAL) {
>>> +             spin_unlock_irqrestore(hba->host->host_lock, flags);
>>> +             return;
>>> +     }
>>> +     spin_unlock_irqrestore(hba->host->host_lock, flags);
>>
>> Why explicit lock/unlock calls instead of using scoped_guard()?
> Should I apply those to host_lock as well?

Yes, please use scoped_guard() and guard() in new code. I expect that
using scoped_guard() here will lead to code that is easier to read.

>>> + * @clk_gating_workq: workqueue for clock gating work.
>>> + * @lock: serialize access to some struct ufs_clk_gating members
>>
>> Please document that @lock is the outer lock relative to the host lock.
> Not sure what you mean?
> host_lock is nested in one place only, should this goes to the @lock documentation?

Whenever locks are nested, the nesting order must be consistent
everywhere. Otherwise there is a risk of triggering an ABBA deadlock.
So I think it is a good practice to document in which order locks should
be nested.

Thanks,

Bart.
Avri Altman Nov. 21, 2024, 9:39 p.m. UTC | #4 
> On 11/21/24 1:06 PM, Avri Altman wrote:
> >> On 11/18/24 6:41 AM, Avri Altman wrote:
> >>> +     spin_lock_irqsave(hba->host->host_lock, flags);
> >>> +     if (ufshcd_has_pending_tasks(hba) ||
> >>> +         hba->ufshcd_state != UFSHCD_STATE_OPERATIONAL) {
> >>> +             spin_unlock_irqrestore(hba->host->host_lock, flags);
> >>> +             return;
> >>> +     }
> >>> +     spin_unlock_irqrestore(hba->host->host_lock, flags);
> >>
> >> Why explicit lock/unlock calls instead of using scoped_guard()?
> > Should I apply those to host_lock as well?
> 
> Yes, please use scoped_guard() and guard() in new code. I expect that using
> scoped_guard() here will lead to code that is easier to read.
Done.

> 
> >>> + * @clk_gating_workq: workqueue for clock gating work.
> >>> + * @lock: serialize access to some struct ufs_clk_gating members
> >>
> >> Please document that @lock is the outer lock relative to the host lock.
> > Not sure what you mean?
> > host_lock is nested in one place only, should this goes to the @lock
> documentation?
> 
> Whenever locks are nested, the nesting order must be consistent
> everywhere. Otherwise there is a risk of triggering an ABBA deadlock.
> So I think it is a good practice to document in which order locks should be
> nested.
Done.

Thanks,
Avri

> 
> Thanks,
> 
> Bart.
diff mbox series

Patch

diff --git a/drivers/ufs/core/ufshcd.c b/drivers/ufs/core/ufshcd.c
index be5fe2407382..638d9c0e2603 100644
--- a/drivers/ufs/core/ufshcd.c
+++ b/drivers/ufs/core/ufshcd.c
@@ -1816,19 +1816,17 @@  static void ufshcd_exit_clk_scaling(struct ufs_hba *hba)
 static void ufshcd_ungate_work(struct work_struct *work)
 {
 	int ret;
-	unsigned long flags;
 	struct ufs_hba *hba = container_of(work, struct ufs_hba,
 			clk_gating.ungate_work);
 
 	cancel_delayed_work_sync(&hba->clk_gating.gate_work);
 
-	spin_lock_irqsave(hba->host->host_lock, flags);
-	if (hba->clk_gating.state == CLKS_ON) {
-		spin_unlock_irqrestore(hba->host->host_lock, flags);
-		return;
+	scoped_guard(spinlock_irqsave, &hba->clk_gating.lock)
+	{
+		if (hba->clk_gating.state == CLKS_ON)
+			return;
 	}
 
-	spin_unlock_irqrestore(hba->host->host_lock, flags);
 	ufshcd_hba_vreg_set_hpm(hba);
 	ufshcd_setup_clocks(hba, true);
 
@@ -1863,7 +1861,7 @@  void ufshcd_hold(struct ufs_hba *hba)
 	if (!ufshcd_is_clkgating_allowed(hba) ||
 	    !hba->clk_gating.is_initialized)
 		return;
-	spin_lock_irqsave(hba->host->host_lock, flags);
+	spin_lock_irqsave(&hba->clk_gating.lock, flags);
 	hba->clk_gating.active_reqs++;
 
 start:
@@ -1879,11 +1877,11 @@  void ufshcd_hold(struct ufs_hba *hba)
 		 */
 		if (ufshcd_can_hibern8_during_gating(hba) &&
 		    ufshcd_is_link_hibern8(hba)) {
-			spin_unlock_irqrestore(hba->host->host_lock, flags);
+			spin_unlock_irqrestore(&hba->clk_gating.lock, flags);
 			flush_result = flush_work(&hba->clk_gating.ungate_work);
 			if (hba->clk_gating.is_suspended && !flush_result)
 				return;
-			spin_lock_irqsave(hba->host->host_lock, flags);
+			spin_lock_irqsave(&hba->clk_gating.lock, flags);
 			goto start;
 		}
 		break;
@@ -1912,17 +1910,17 @@  void ufshcd_hold(struct ufs_hba *hba)
 		 */
 		fallthrough;
 	case REQ_CLKS_ON:
-		spin_unlock_irqrestore(hba->host->host_lock, flags);
+		spin_unlock_irqrestore(&hba->clk_gating.lock, flags);
 		flush_work(&hba->clk_gating.ungate_work);
 		/* Make sure state is CLKS_ON before returning */
-		spin_lock_irqsave(hba->host->host_lock, flags);
+		spin_lock_irqsave(&hba->clk_gating.lock, flags);
 		goto start;
 	default:
 		dev_err(hba->dev, "%s: clk gating is in invalid state %d\n",
 				__func__, hba->clk_gating.state);
 		break;
 	}
-	spin_unlock_irqrestore(hba->host->host_lock, flags);
+	spin_unlock_irqrestore(&hba->clk_gating.lock, flags);
 }
 EXPORT_SYMBOL_GPL(ufshcd_hold);
 
@@ -1933,26 +1931,32 @@  static void ufshcd_gate_work(struct work_struct *work)
 	unsigned long flags;
 	int ret;
 
-	spin_lock_irqsave(hba->host->host_lock, flags);
-	/*
-	 * In case you are here to cancel this work the gating state
-	 * would be marked as REQ_CLKS_ON. In this case save time by
-	 * skipping the gating work and exit after changing the clock
-	 * state to CLKS_ON.
-	 */
-	if (hba->clk_gating.is_suspended ||
-		(hba->clk_gating.state != REQ_CLKS_OFF)) {
-		hba->clk_gating.state = CLKS_ON;
-		trace_ufshcd_clk_gating(dev_name(hba->dev),
-					hba->clk_gating.state);
-		goto rel_lock;
+	scoped_guard(spinlock_irqsave, &hba->clk_gating.lock)
+	{
+		/*
+		 * In case you are here to cancel this work the gating state
+		 * would be marked as REQ_CLKS_ON. In this case save time by
+		 * skipping the gating work and exit after changing the clock
+		 * state to CLKS_ON.
+		 */
+		if (hba->clk_gating.is_suspended ||
+		    hba->clk_gating.state != REQ_CLKS_OFF) {
+			hba->clk_gating.state = CLKS_ON;
+			trace_ufshcd_clk_gating(dev_name(hba->dev),
+						hba->clk_gating.state);
+			return;
+		}
+
+		if (hba->clk_gating.active_reqs)
+			return;
 	}
 
+	spin_lock_irqsave(hba->host->host_lock, flags);
 	if (ufshcd_is_ufs_dev_busy(hba) ||
-	    hba->ufshcd_state != UFSHCD_STATE_OPERATIONAL ||
-	    hba->clk_gating.active_reqs)
-		goto rel_lock;
-
+	    hba->ufshcd_state != UFSHCD_STATE_OPERATIONAL) {
+		spin_unlock_irqrestore(hba->host->host_lock, flags);
+		return;
+	}
 	spin_unlock_irqrestore(hba->host->host_lock, flags);
 
 	/* put the link into hibern8 mode before turning off clocks */
@@ -1964,7 +1968,7 @@  static void ufshcd_gate_work(struct work_struct *work)
 					__func__, ret);
 			trace_ufshcd_clk_gating(dev_name(hba->dev),
 						hba->clk_gating.state);
-			goto out;
+			return;
 		}
 		ufshcd_set_link_hibern8(hba);
 	}
@@ -1984,32 +1988,37 @@  static void ufshcd_gate_work(struct work_struct *work)
 	 * prevent from doing cancel work multiple times when there are
 	 * new requests arriving before the current cancel work is done.
 	 */
-	spin_lock_irqsave(hba->host->host_lock, flags);
+	guard(spinlock_irqsave)(&hba->clk_gating.lock);
 	if (hba->clk_gating.state == REQ_CLKS_OFF) {
 		hba->clk_gating.state = CLKS_OFF;
 		trace_ufshcd_clk_gating(dev_name(hba->dev),
 					hba->clk_gating.state);
 	}
-rel_lock:
-	spin_unlock_irqrestore(hba->host->host_lock, flags);
-out:
-	return;
 }
 
 /* host lock must be held before calling this variant */
 static void __ufshcd_release(struct ufs_hba *hba)
 {
+	unsigned long flags;
+
 	if (!ufshcd_is_clkgating_allowed(hba))
 		return;
 
 	hba->clk_gating.active_reqs--;
 
 	if (hba->clk_gating.active_reqs || hba->clk_gating.is_suspended ||
-	    hba->ufshcd_state != UFSHCD_STATE_OPERATIONAL ||
-	    ufshcd_has_pending_tasks(hba) || !hba->clk_gating.is_initialized ||
+	    !hba->clk_gating.is_initialized ||
 	    hba->clk_gating.state == CLKS_OFF)
 		return;
 
+	spin_lock_irqsave(hba->host->host_lock, flags);
+	if (ufshcd_has_pending_tasks(hba) ||
+	    hba->ufshcd_state != UFSHCD_STATE_OPERATIONAL) {
+		spin_unlock_irqrestore(hba->host->host_lock, flags);
+		return;
+	}
+	spin_unlock_irqrestore(hba->host->host_lock, flags);
+
 	hba->clk_gating.state = REQ_CLKS_OFF;
 	trace_ufshcd_clk_gating(dev_name(hba->dev), hba->clk_gating.state);
 	queue_delayed_work(hba->clk_gating.clk_gating_workq,
@@ -2019,11 +2028,8 @@  static void __ufshcd_release(struct ufs_hba *hba)
 
 void ufshcd_release(struct ufs_hba *hba)
 {
-	unsigned long flags;
-
-	spin_lock_irqsave(hba->host->host_lock, flags);
+	guard(spinlock_irqsave)(&hba->clk_gating.lock);
 	__ufshcd_release(hba);
-	spin_unlock_irqrestore(hba->host->host_lock, flags);
 }
 EXPORT_SYMBOL_GPL(ufshcd_release);
 
@@ -2038,11 +2044,9 @@  static ssize_t ufshcd_clkgate_delay_show(struct device *dev,
 void ufshcd_clkgate_delay_set(struct device *dev, unsigned long value)
 {
 	struct ufs_hba *hba = dev_get_drvdata(dev);
-	unsigned long flags;
 
-	spin_lock_irqsave(hba->host->host_lock, flags);
+	guard(spinlock_irqsave)(&hba->clk_gating.lock);
 	hba->clk_gating.delay_ms = value;
-	spin_unlock_irqrestore(hba->host->host_lock, flags);
 }
 EXPORT_SYMBOL_GPL(ufshcd_clkgate_delay_set);
 
@@ -2070,7 +2074,6 @@  static ssize_t ufshcd_clkgate_enable_store(struct device *dev,
 		struct device_attribute *attr, const char *buf, size_t count)
 {
 	struct ufs_hba *hba = dev_get_drvdata(dev);
-	unsigned long flags;
 	u32 value;
 
 	if (kstrtou32(buf, 0, &value))
@@ -2078,9 +2081,10 @@  static ssize_t ufshcd_clkgate_enable_store(struct device *dev,
 
 	value = !!value;
 
-	spin_lock_irqsave(hba->host->host_lock, flags);
+	guard(spinlock_irqsave)(&hba->clk_gating.lock);
+
 	if (value == hba->clk_gating.is_enabled)
-		goto out;
+		return count;
 
 	if (value)
 		__ufshcd_release(hba);
@@ -2088,8 +2092,7 @@  static ssize_t ufshcd_clkgate_enable_store(struct device *dev,
 		hba->clk_gating.active_reqs++;
 
 	hba->clk_gating.is_enabled = value;
-out:
-	spin_unlock_irqrestore(hba->host->host_lock, flags);
+
 	return count;
 }
 
@@ -2131,6 +2134,8 @@  static void ufshcd_init_clk_gating(struct ufs_hba *hba)
 	INIT_DELAYED_WORK(&hba->clk_gating.gate_work, ufshcd_gate_work);
 	INIT_WORK(&hba->clk_gating.ungate_work, ufshcd_ungate_work);
 
+	spin_lock_init(&hba->clk_gating.lock);
+
 	hba->clk_gating.clk_gating_workq = alloc_ordered_workqueue(
 		"ufs_clk_gating_%d", WQ_MEM_RECLAIM | WQ_HIGHPRI,
 		hba->host->host_no);
@@ -9120,7 +9125,6 @@  static int ufshcd_setup_clocks(struct ufs_hba *hba, bool on)
 	int ret = 0;
 	struct ufs_clk_info *clki;
 	struct list_head *head = &hba->clk_list_head;
-	unsigned long flags;
 	ktime_t start = ktime_get();
 	bool clk_state_changed = false;
 
@@ -9171,11 +9175,10 @@  static int ufshcd_setup_clocks(struct ufs_hba *hba, bool on)
 				clk_disable_unprepare(clki->clk);
 		}
 	} else if (!ret && on) {
-		spin_lock_irqsave(hba->host->host_lock, flags);
-		hba->clk_gating.state = CLKS_ON;
+		scoped_guard(spinlock_irqsave, &hba->clk_gating.lock)
+			hba->clk_gating.state = CLKS_ON;
 		trace_ufshcd_clk_gating(dev_name(hba->dev),
 					hba->clk_gating.state);
-		spin_unlock_irqrestore(hba->host->host_lock, flags);
 	}
 
 	if (clk_state_changed)
diff --git a/include/ufs/ufshcd.h b/include/ufs/ufshcd.h
index d7aca9e61684..8f9997b0dbf9 100644
--- a/include/ufs/ufshcd.h
+++ b/include/ufs/ufshcd.h
@@ -403,6 +403,8 @@  enum clk_gating_state {
  * delay_ms
  * @ungate_work: worker to turn on clocks that will be used in case of
  * interrupt context
+ * @clk_gating_workq: workqueue for clock gating work.
+ * @lock: serialize access to some struct ufs_clk_gating members
  * @state: the current clocks state
  * @delay_ms: gating delay in ms
  * @is_suspended: clk gating is suspended when set to 1 which can be used
@@ -413,11 +415,14 @@  enum clk_gating_state {
  * @is_initialized: Indicates whether clock gating is initialized or not
  * @active_reqs: number of requests that are pending and should be waited for
  * completion before gating clocks.
- * @clk_gating_workq: workqueue for clock gating work.
  */
 struct ufs_clk_gating {
 	struct delayed_work gate_work;
 	struct work_struct ungate_work;
+	struct workqueue_struct *clk_gating_workq;
+
+	spinlock_t lock;
+
 	enum clk_gating_state state;
 	unsigned long delay_ms;
 	bool is_suspended;
@@ -426,7 +431,6 @@  struct ufs_clk_gating {
 	bool is_enabled;
 	bool is_initialized;
 	int active_reqs;
-	struct workqueue_struct *clk_gating_workq;
 };
 
 /**