Skip to contentSkip to conversation

mfd: qcom-pm8xxx: Fix potential deadlock on &chip->pm_irq_lock

PATCH Superseded Chengfeng Ye 3 participants 6 messages Jun 28, 2023 – Jul 20, 2023

Subsystems (2)

v1 of 2 Superseded Latest is v2 →

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 mfd: qcom-pm8xxx: Fix potential deadlock on &chip->pm_irq_lock

As &chip->pm_irq_lock is acquired by pm8xxx_irq_handler() under irq context, other process context code should disable irq before acquiring the lock.

I think .irq_set_type and .irq_get_irqchip_state callbacks should be executed from process context without irq disabled by default. Thus the same lock acquision should disable irq.

Possible deadlock scenario
pm8xxx_irq_set_type()

    -> pm8xxx_config_irq()
    -> spin_lock(&chip->pm_irq_lock)
        <irq interrupt>
        -> pm8xxx_irq_handler()
        -> pm8xxx_irq_master_handler()
        -> pm8xxx_irq_block_handler()
        -> pm8xxx_read_block_irq()
        -> spin_lock(&chip->pm_irq_lock) (deadlock here)

This flaw was found using an experimental static analysis tool we are developing for irq-related deadlock.

The tentative patch fix the potential deadlock by spin_lock_irqsave().

---
 drivers/mfd/qcom-pm8xxx.c | 10 ++++++----
 1 file changed, 6 insertions(+), 4 deletions(-)
drivers/mfd/qcom-pm8xxx.c +6 -4
index 9a948df8c28d..07c531bd1236 100644
@@ -103,8 +103,9 @@ static int
 pm8xxx_config_irq(struct pm_irq_chip *chip, unsigned int bp, unsigned int cp)
 {
 	int	rc;
+	unsigned long flags;
 
-	spin_lock(&chip->pm_irq_lock);
+	spin_lock_irqsave(&chip->pm_irq_lock, flags);
 	rc = regmap_write(chip->regmap, SSBI_REG_ADDR_IRQ_BLK_SEL, bp);
 	if (rc) {
 		pr_err("Failed Selecting Block %d rc=%d\n", bp, rc);
@@ -116,7 +117,7 @@ pm8xxx_config_irq(struct pm_irq_chip *chip, unsigned int bp, unsigned int cp)
 	if (rc)
 		pr_err("Failed Configuring IRQ rc=%d\n", rc);
 bail:
-	spin_unlock(&chip->pm_irq_lock);
+	spin_unlock_irqrestore(&chip->pm_irq_lock, flags);
 	return rc;
 }
 
@@ -321,6 +322,7 @@ static int pm8xxx_irq_get_irqchip_state(struct irq_data *d,
 	struct pm_irq_chip *chip = irq_data_get_irq_chip_data(d);
 	unsigned int pmirq = irqd_to_hwirq(d);
 	unsigned int bits;
+	unsigned long flags;
 	int irq_bit;
 	u8 block;
 	int rc;
@@ -331,7 +333,7 @@ static int pm8xxx_irq_get_irqchip_state(struct irq_data *d,
 	block = pmirq / 8;
 	irq_bit = pmirq % 8;
 
-	spin_lock(&chip->pm_irq_lock);
+	spin_lock_irqsave(&chip->pm_irq_lock, flags);
 	rc = regmap_write(chip->regmap, SSBI_REG_ADDR_IRQ_BLK_SEL, block);
 	if (rc) {
 		pr_err("Failed Selecting Block %d rc=%d\n", block, rc);
@@ -346,7 +348,7 @@ static int pm8xxx_irq_get_irqchip_state(struct irq_data *d,
 
 	*state = !!(bits & BIT(irq_bit));
 bail:
-	spin_unlock(&chip->pm_irq_lock);
+	spin_unlock_irqrestore(&chip->pm_irq_lock, flags);
 
 	return rc;
 }
Signature · 1 line
2.17.1
maintainer
#1 lore raw

I'd like some additional eyes-on please.

This is very old code that you're changing and some of the people who made the largest contributions are not on Cc.

Quoted 82 lines · Chengfeng Ye

As &chip->pm_irq_lock is acquired by pm8xxx_irq_handler() under irq context, other process context code should disable irq before acquiring the lock.

I think .irq_set_type and .irq_get_irqchip_state callbacks should be executed from process context without irq disabled by default. Thus the same lock acquision should disable irq.

Possible deadlock scenario
pm8xxx_irq_set_type()

    -> pm8xxx_config_irq()
    -> spin_lock(&chip->pm_irq_lock)
        <irq interrupt>
        -> pm8xxx_irq_handler()
        -> pm8xxx_irq_master_handler()
        -> pm8xxx_irq_block_handler()
        -> pm8xxx_read_block_irq()
        -> spin_lock(&chip->pm_irq_lock) (deadlock here)

This flaw was found using an experimental static analysis tool we are developing for irq-related deadlock.

The tentative patch fix the potential deadlock by spin_lock_irqsave().

---
 drivers/mfd/qcom-pm8xxx.c | 10 ++++++----
 1 file changed, 6 insertions(+), 4 deletions(-)
drivers/mfd/qcom-pm8xxx.c +6 -4
index 9a948df8c28d..07c531bd1236 100644
@@ -103,8 +103,9 @@ static int
 pm8xxx_config_irq(struct pm_irq_chip *chip, unsigned int bp, unsigned int cp)
 {
 	int	rc;
+	unsigned long flags;
 
-	spin_lock(&chip->pm_irq_lock);
+	spin_lock_irqsave(&chip->pm_irq_lock, flags);
 	rc = regmap_write(chip->regmap, SSBI_REG_ADDR_IRQ_BLK_SEL, bp);
 	if (rc) {
 		pr_err("Failed Selecting Block %d rc=%d\n", bp, rc);
@@ -116,7 +117,7 @@ pm8xxx_config_irq(struct pm_irq_chip *chip, unsigned int bp, unsigned int cp)
 	if (rc)
 		pr_err("Failed Configuring IRQ rc=%d\n", rc);
 bail:
-	spin_unlock(&chip->pm_irq_lock);
+	spin_unlock_irqrestore(&chip->pm_irq_lock, flags);
 	return rc;
 }
 
@@ -321,6 +322,7 @@ static int pm8xxx_irq_get_irqchip_state(struct irq_data *d,
 	struct pm_irq_chip *chip = irq_data_get_irq_chip_data(d);
 	unsigned int pmirq = irqd_to_hwirq(d);
 	unsigned int bits;
+	unsigned long flags;
 	int irq_bit;
 	u8 block;
 	int rc;
@@ -331,7 +333,7 @@ static int pm8xxx_irq_get_irqchip_state(struct irq_data *d,
 	block = pmirq / 8;
 	irq_bit = pmirq % 8;
 
-	spin_lock(&chip->pm_irq_lock);
+	spin_lock_irqsave(&chip->pm_irq_lock, flags);
 	rc = regmap_write(chip->regmap, SSBI_REG_ADDR_IRQ_BLK_SEL, block);
 	if (rc) {
 		pr_err("Failed Selecting Block %d rc=%d\n", block, rc);
@@ -346,7 +348,7 @@ static int pm8xxx_irq_get_irqchip_state(struct irq_data *d,
 
 	*state = !!(bits & BIT(irq_bit));
 bail:
-	spin_unlock(&chip->pm_irq_lock);
+	spin_unlock_irqrestore(&chip->pm_irq_lock, flags);
 
 	return rc;
 }
Signature · 1 line
2.17.1
Signature · 1 line
Lee Jones [李琼斯]
#2 lore raw
Quoted 2 lines

This is very old code that you're changing and some of the people who made the largest contributions are not on Cc.

Thanks for letting know. Will be patient while the patch is being reviewed.

Best,
Chengfeng

maintainer
#3 lore raw
Quoted 4 lines · Chengfeng Ye
Quoted 2 lines

This is very old code that you're changing and some of the people who made the largest contributions are not on Cc.

Thanks for letting know. Will be patient while the patch is being reviewed.

No, I need you to resubmit it with more historical people on Cc please.

git log --full-diff should help you trace back through name changes.

Signature · 1 line
Lee Jones [李琼斯]
Reviewed-by on v1 1/1
#4 lore raw
In reply to Chengfeng Ye
Quoted 5 lines · Chengfeng Ye

As &chip->pm_irq_lock is acquired by pm8xxx_irq_handler() under irq context, other process context code should disable irq before acquiring the lock.

I think .irq_set_type and .irq_get_irqchip_state callbacks should be

You are correct, so please drop "I think", and change "should be" to "are generally".

Quoted 14 lines

executed from process context without irq disabled by default. Thus the same lock acquision should disable irq.

Possible deadlock scenario
pm8xxx_irq_set_type()

    -> pm8xxx_config_irq()
    -> spin_lock(&chip->pm_irq_lock)
        <irq interrupt>
        -> pm8xxx_irq_handler()
        -> pm8xxx_irq_master_handler()
        -> pm8xxx_irq_block_handler()
        -> pm8xxx_read_block_irq()
        -> spin_lock(&chip->pm_irq_lock) (deadlock here)
Quoted 4 lines

This flaw was found using an experimental static analysis tool we are developing for irq-related deadlock.

The tentative patch fix the potential deadlock by spin_lock_irqsave().

I don't think this is a "tentative patch fix", it is the patch to fix the issue. I think you can omit this line, because you already described your problem, and the solution above.

Quoted 2 lines

Regards,
Bjorn

#5 lore raw
Quoted 5 lines

You are correct, so please drop "I think", and change "should be" to "are generally"
I don't think this is a "tentative patch fix", it is the patch to fix
the issue. I think you can omit this line, because you already described your problem, and the solution above.

The V2 patch was just sent to address the problem.

Quoted 1 line

Also thanks for your time in reviewing the patch :)

Quoted 1 line

No, I need you to resubmit it with more historical people on Cc please.

Thanks for the reminder, I Cc more people with large contributions to the file in the V2 patch.

Best Regards,
Chengfeng