diff mbox series

[PATCHv2] pinctrl-bcm2835.c: fix race condition when setting gpio dir

Message ID 4302b66b-ca20-0f19-d2aa-ee8661118863@xs4all.nl
State Accepted
Commit b7badd752de05312fdb1aeb388480f706d0c087f
Headers show
Series [PATCHv2] pinctrl-bcm2835.c: fix race condition when setting gpio dir | expand

Commit Message

Hans Verkuil April 20, 2023, 12:47 p.m. UTC 
In the past setting the pin direction called pinctrl_gpio_direction()
which uses a mutex to serialize this. That was changed to set the
direction directly in the pin controller driver, but that lost the
serialization mechanism. Since the direction of multiple pins are in
the same register you can have a race condition, something that was
in fact observed with the cec-gpio driver.

Add a new spinlock to serialize writing to the FSEL registers.

Signed-off-by: Hans Verkuil <hverkuil-cisco@xs4all.nl>
Fixes: 1a4541b68e25 ("pinctrl-bcm2835: don't call pinctrl_gpio_direction()")
---
Changes since v1:
- use 'goto unlock' as per Stefan's suggestion
---
 drivers/pinctrl/bcm/pinctrl-bcm2835.c | 19 +++++++++++++++----
 1 file changed, 15 insertions(+), 4 deletions(-)

Comments

Linus Walleij April 21, 2023, 9:01 a.m. UTC | #1 
On Thu, Apr 20, 2023 at 2:47 PM Hans Verkuil <hverkuil-cisco@xs4all.nl> wrote:

> In the past setting the pin direction called pinctrl_gpio_direction()
> which uses a mutex to serialize this. That was changed to set the
> direction directly in the pin controller driver, but that lost the
> serialization mechanism. Since the direction of multiple pins are in
> the same register you can have a race condition, something that was
> in fact observed with the cec-gpio driver.
>
> Add a new spinlock to serialize writing to the FSEL registers.
>
> Signed-off-by: Hans Verkuil <hverkuil-cisco@xs4all.nl>
> Fixes: 1a4541b68e25 ("pinctrl-bcm2835: don't call pinctrl_gpio_direction()")

I just applied this patch now so it can go in with the rest of the patches
for the merge window. I don't especially fancy having to immediately
start a fixes branch before I've even sent the bulk of patches to Torvalds.

Yours,
Linus Walleij
diff mbox series

Patch

diff --git a/drivers/pinctrl/bcm/pinctrl-bcm2835.c b/drivers/pinctrl/bcm/pinctrl-bcm2835.c
index 8e2551a08c37..7435173e10f4 100644
--- a/drivers/pinctrl/bcm/pinctrl-bcm2835.c
+++ b/drivers/pinctrl/bcm/pinctrl-bcm2835.c
@@ -90,6 +90,8 @@  struct bcm2835_pinctrl {
 	struct pinctrl_gpio_range gpio_range;

 	raw_spinlock_t irq_lock[BCM2835_NUM_BANKS];
+	/* Protect FSEL registers */
+	spinlock_t fsel_lock;
 };

 /* pins are just named GPIO0..GPIO53 */
@@ -284,14 +286,19 @@  static inline void bcm2835_pinctrl_fsel_set(
 		struct bcm2835_pinctrl *pc, unsigned pin,
 		enum bcm2835_fsel fsel)
 {
-	u32 val = bcm2835_gpio_rd(pc, FSEL_REG(pin));
-	enum bcm2835_fsel cur = (val >> FSEL_SHIFT(pin)) & BCM2835_FSEL_MASK;
+	u32 val;
+	enum bcm2835_fsel cur;
+	unsigned long flags;
+
+	spin_lock_irqsave(&pc->fsel_lock, flags);
+	val = bcm2835_gpio_rd(pc, FSEL_REG(pin));
+	cur = (val >> FSEL_SHIFT(pin)) & BCM2835_FSEL_MASK;

 	dev_dbg(pc->dev, "read %08x (%u => %s)\n", val, pin,
-			bcm2835_functions[cur]);
+		bcm2835_functions[cur]);

 	if (cur == fsel)
-		return;
+		goto unlock;

 	if (cur != BCM2835_FSEL_GPIO_IN && fsel != BCM2835_FSEL_GPIO_IN) {
 		/* always transition through GPIO_IN */
@@ -309,6 +316,9 @@  static inline void bcm2835_pinctrl_fsel_set(
 	dev_dbg(pc->dev, "write %08x (%u <= %s)\n", val, pin,
 			bcm2835_functions[fsel]);
 	bcm2835_gpio_wr(pc, FSEL_REG(pin), val);
+
+unlock:
+	spin_unlock_irqrestore(&pc->fsel_lock, flags);
 }

 static int bcm2835_gpio_direction_input(struct gpio_chip *chip, unsigned offset)
@@ -1248,6 +1258,7 @@  static int bcm2835_pinctrl_probe(struct platform_device *pdev)
 	pc->gpio_chip = *pdata->gpio_chip;
 	pc->gpio_chip.parent = dev;

+	spin_lock_init(&pc->fsel_lock);
 	for (i = 0; i < BCM2835_NUM_BANKS; i++) {
 		unsigned long events;
 		unsigned offset;