Skip to contentSkip to conversation

ARM: SAMSUNG: Bug fix spin_lock recursion in clk_enable() and clk_disable()

PATCH New Jaecheol Lee 5 participants 6 messages Oct 12, 2010 – Oct 21, 2010

Subsystems (3)
lore
Share & tools
lore
#0 lore raw

PATCH 1/1 ARM: SAMSUNG: Bug fix spin_lock recursion in clk_enable() and clk_disable()

From: Minho Ban <[email protected]>

The clk_enable() and clk_disable() can be used process and ISR either. So spin_lock_irqsave should be used instead.

---
 arch/arm/plat-samsung/clock.c |   12 ++++++++----
 1 files changed, 8 insertions(+), 4 deletions(-)
arch/arm/plat-samsung/clock.c +8 -4
index e8d20b0..2a991a5 100644
@@ -138,31 +138,35 @@ void clk_put(struct clk *clk)
 
 int clk_enable(struct clk *clk)
 {
+	unsigned long flags;
+
 	if (IS_ERR(clk) || clk == NULL)
 		return -EINVAL;
 
 	clk_enable(clk->parent);
 
-	spin_lock(&clocks_lock);
+	spin_lock_irqsave(&clocks_lock, flags);
 
 	if ((clk->usage++) == 0)
 		(clk->enable)(clk, 1);
 
-	spin_unlock(&clocks_lock);
+	spin_unlock_irqrestore(&clocks_lock, flags);
 	return 0;
 }
 
 void clk_disable(struct clk *clk)
 {
+	unsigned long flags;
+
 	if (IS_ERR(clk) || clk == NULL)
 		return;
 
-	spin_lock(&clocks_lock);
+	spin_lock_irqsave(&clocks_lock, flags);
 
 	if ((--clk->usage) == 0)
 		(clk->enable)(clk, 0);
 
-	spin_unlock(&clocks_lock);
+	spin_unlock_irqrestore(&clocks_lock, flags);
 	clk_disable(clk->parent);
 }
 
Signature · 1 line
1.6.2.5
#1 lore raw

PATCH 1/1 ARM: SAMSUNG: Bug fix spin_lock recursion in clk_enable() and clk_disable()

Quoted 43 lines · Jaecheol Lee

From: Minho Ban <[email protected]>

The clk_enable() and clk_disable() can be used process and ISR either. So spin_lock_irqsave should be used instead.

---
 arch/arm/plat-samsung/clock.c |   12 ++++++++----
 1 files changed, 8 insertions(+), 4 deletions(-)
arch/arm/plat-samsung/clock.c +6 -2
index e8d20b0..2a991a5 100644
@@ -138,31 +138,35 @@ void clk_put(struct clk *clk)
 
 int clk_enable(struct clk *clk)
 {
+	unsigned long flags;
+
 	if (IS_ERR(clk) || clk == NULL)
 		return -EINVAL;
 
 	clk_enable(clk->parent);
 
-	spin_lock(&clocks_lock);
+	spin_lock_irqsave(&clocks_lock, flags);
 
 	if ((clk->usage++) == 0)
 		(clk->enable)(clk, 1);
 
-	spin_unlock(&clocks_lock);
+	spin_unlock_irqrestore(&clocks_lock, flags);
 	return 0;
 }
 
 void clk_disable(struct clk *clk)
 {
+	unsigned long flags;
+
 	if (IS_ERR(clk) || clk == NULL)
 		return;

arch/arm/plat-samsung/clock.c void clk_put(struct clk *clk) · line 170

Quoted 10 lines
arch/arm/plat-samsung/clock.c +2 -2
 
-	spin_lock(&clocks_lock);
+	spin_lock_irqsave(&clocks_lock, flags);
 
 	if ((--clk->usage) == 0)
 		(clk->enable)(clk, 0);
 
-	spin_unlock(&clocks_lock);
+	spin_unlock_irqrestore(&clocks_lock, flags);
 	clk_disable(clk->parent);

I'm not sure, but I don't belive that the clk_ api has ever been callable from non-sleepable contexts such as interrupt handlers.

I would welcome RMK's response (or any other response) about this issue?

Personally, given that some clk_ calls may sleep esp. for pll enables, I would prefer to see this patch dropped in favour of fixing the users.

Signature · 1 line
Ben
#2 lore raw

PATCH 1/1 ARM: SAMSUNG: Bug fix spin_lock recursion in clk_enable() and clk_disable()

Quoted 59 lines · Ben Dooks
Quoted 43 lines · Jaecheol Lee

From: Minho Ban <[email protected]>

The clk_enable() and clk_disable() can be used process and ISR either. So spin_lock_irqsave should be used instead.

---
?arch/arm/plat-samsung/clock.c | ? 12 ++++++++---- ?1 files changed, 8 insertions(+), 4 deletions(-)

arch/arm/plat-samsung/clock.c +6 -2
index e8d20b0..2a991a5 100644
@@ -138,31 +138,35 @@ void clk_put(struct clk *clk)
 
 ?int clk_enable(struct clk *clk)
 ?{
+ ? ? unsigned long flags;
+
 ? ? ? if (IS_ERR(clk) || clk == NULL)
 ? ? ? ? ? ? ? return -EINVAL;
 
 ? ? ? clk_enable(clk->parent);
 
- ? ? spin_lock(&clocks_lock);
+ ? ? spin_lock_irqsave(&clocks_lock, flags);
 
 ? ? ? if ((clk->usage++) == 0)
 ? ? ? ? ? ? ? (clk->enable)(clk, 1);
 
- ? ? spin_unlock(&clocks_lock);
+ ? ? spin_unlock_irqrestore(&clocks_lock, flags);
 ? ? ? return 0;
 ?}
 
 ?void clk_disable(struct clk *clk)
 ?{
+ ? ? unsigned long flags;
+
 ? ? ? if (IS_ERR(clk) || clk == NULL)
 ? ? ? ? ? ? ? return;

arch/arm/plat-samsung/clock.c void clk_put(struct clk *clk) · line 170

Quoted 10 lines
arch/arm/plat-samsung/clock.c +2 -2
 
- ? ? spin_lock(&clocks_lock);
+ ? ? spin_lock_irqsave(&clocks_lock, flags);
 
 ? ? ? if ((--clk->usage) == 0)
 ? ? ? ? ? ? ? (clk->enable)(clk, 0);
 
- ? ? spin_unlock(&clocks_lock);
+ ? ? spin_unlock_irqrestore(&clocks_lock, flags);
 ? ? ? clk_disable(clk->parent);

I'm not sure, but I don't belive that the clk_ api has ever been callable from non-sleepable contexts such as interrupt handlers.

Quoted 6 lines

I would welcome RMK's response (or any other response) about this issue?

Personally, given that some clk_ calls may sleep esp. for pll enables, I would prefer to see this patch dropped in favour of fixing the users.

I tend to agree.
I would like to know any such case where enabling/disabling of some clock is so urgent as to require doing from ISR and not from a tasklet scheduled from ISR.
Also let us not make it impossible in future to switch to common clock api by Jeremy Kerr, that doesn't allow calls from IRQ context.

#3 lore raw

PATCH 1/1 ARM: SAMSUNG: Bug fix spin_lock recursion in clk_enable() and clk_disable()

Quoted 67 lines · Jassi Brar
Quoted 59 lines · Ben Dooks
Quoted 43 lines · Jaecheol Lee

From: Minho Ban <[email protected]>

The clk_enable() and clk_disable() can be used process and ISR either. So spin_lock_irqsave should be used instead.

---
 arch/arm/plat-samsung/clock.c |   12 ++++++++----
 1 files changed, 8 insertions(+), 4 deletions(-)
arch/arm/plat-samsung/clock.c +6 -2
index e8d20b0..2a991a5 100644
@@ -138,31 +138,35 @@ void clk_put(struct clk *clk)
 
 int clk_enable(struct clk *clk)
 {
+     unsigned long flags;
+
      if (IS_ERR(clk) || clk == NULL)
              return -EINVAL;
 
      clk_enable(clk->parent);
 
-     spin_lock(&clocks_lock);
+     spin_lock_irqsave(&clocks_lock, flags);
 
      if ((clk->usage++) == 0)
              (clk->enable)(clk, 1);
 
-     spin_unlock(&clocks_lock);
+     spin_unlock_irqrestore(&clocks_lock, flags);
      return 0;
 }
 
 void clk_disable(struct clk *clk)
 {
+     unsigned long flags;
+
      if (IS_ERR(clk) || clk == NULL)
              return;

arch/arm/plat-samsung/clock.c void clk_put(struct clk *clk) · line 170

Quoted 10 lines
arch/arm/plat-samsung/clock.c +2 -2
 
-     spin_lock(&clocks_lock);
+     spin_lock_irqsave(&clocks_lock, flags);
 
      if ((--clk->usage) == 0)
              (clk->enable)(clk, 0);
 
-     spin_unlock(&clocks_lock);
+     spin_unlock_irqrestore(&clocks_lock, flags);
      clk_disable(clk->parent);

I'm not sure, but I don't belive that the clk_ api has ever been callable from non-sleepable contexts such as interrupt handlers.

Quoted 6 lines

I would welcome RMK's response (or any other response) about this issue?

Personally, given that some clk_ calls may sleep esp. for pll enables, I would prefer to see this patch dropped in favour of fixing the users.

Quoted 10 lines

I tend to agree.
I would like to know any such case where enabling/disabling of some clock is so urgent as to require doing from ISR and not from a tasklet scheduled from ISR.
Also let us not make it impossible in future to switch to common clock api by Jeremy Kerr, that doesn't allow calls from IRQ context.
--
To unsubscribe from this list: send the line "unsubscribe linux-samsung-soc" in the body of a message to majordomo at vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html

Hello,

Mostly you are right but it would require some explicit comment in clk.h (eg, clk_get/put) not to be confusing to the driver developer.

Thanks.

Regards,
Minho Ban

Signature · 4 lines
Minho Ban (mhban at samsung.com)
Senior Engineer
Platform R&D Team,  Mobile Comm.
SAMSUNG ELECTRONICS Co.Ltd.
#4 lore raw

PATCH 1/1 ARM: SAMSUNG: Bug fix spin_lock recursion in clk_enable() and clk_disable()

In reply to Ben Dooks
Quoted 6 lines · Ben Dooks

I'm not sure, but I don't belive that the clk_ api has ever been callable from non-sleepable contexts such as interrupt handlers.

I would welcome RMK's response (or any other response) about this issue?

We have traditionally allowed it from non-sleepable contexts, mainly because the platforms I've been using it with have had such simple clock implementations that it's not worth artifically restricting it, or the implementation only requires a few us delay after enabling a clock.

#5 lore raw

PATCH 1/1 ARM: SAMSUNG: Bug fix spin_lock recursion in clk_enable() and clk_disable()

Quoted 5 lines · Russell King - ARM Linux
Quoted 4 lines · Ben Dooks

I'm not sure, but I don't belive that the clk_ api has ever been callable from non-sleepable contexts such as interrupt handlers.

Quoted 8 lines
Quoted 2 lines

I would welcome RMK's response (or any other response) about this issue?

We have traditionally allowed it from non-sleepable contexts, mainly because the platforms I've been using it with have had such simple clock implementations that it's not worth artifically restricting it, or the implementation only requires a few us delay after enabling a clock.

Some on chip PLLs can be 150-300uS to get running.

Signature · 1 line
Ben