[PATCH 1/2] drivers: thermal: tsens: fix wrong check for tzd in irq handlers

[PATCH 1/2] drivers: thermal: tsens: fix wrong check for tzd in irq handlers

[Ansuel Smith]

Some device can have some thermal sensor disabled from the factory. The
current 2 irq handler functions check all the sensor by default and the
check if the sensor was actually registered is wrong. The tzd is
actually never set if the registration fail hence the IS_ERR check is
wrong.

Signed-off-by: Ansuel Smith <ansuelsmth@gmail.com>
---
 drivers/thermal/qcom/tsens.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/drivers/thermal/qcom/tsens.c b/drivers/thermal/qcom/tsens.c
index 4c7ebd1d3f9c..b1162e566a70 100644
--- a/drivers/thermal/qcom/tsens.c
+++ b/drivers/thermal/qcom/tsens.c
@@ -417,7 +417,7 @@ static irqreturn_t tsens_critical_irq_thread(int irq, void *data)
 		const struct tsens_sensor *s = &priv->sensor[i];
 		u32 hw_id = s->hw_id;
 
-		if (IS_ERR(s->tzd))
+		if (!s->tzd)
 			continue;
 		if (!tsens_threshold_violated(priv, hw_id, &d))
 			continue;
@@ -467,7 +467,7 @@ static irqreturn_t tsens_irq_thread(int irq, void *data)
 		const struct tsens_sensor *s = &priv->sensor[i];
 		u32 hw_id = s->hw_id;
 
-		if (IS_ERR(s->tzd))
+		if (!s->tzd)
 			continue;
 		if (!tsens_threshold_violated(priv, hw_id, &d))
 			continue;
-- 
2.32.0

[PATCH 2/2] drivers: thermal: tsens: add timeout to get_tem_tsens_valid

[Ansuel Smith]

The function can loop and lock the system if for whatever reason the bit
for the target sensor is NEVER valid. This is the case if a sensor is
disabled by the factory and the valid bit is never reported as actually
valid. Add a timeout check and exit if a timeout occurs. As this is
a very rare condition, handle the timeout only if the first read fails.

Signed-off-by: Ansuel Smith <ansuelsmth@gmail.com>
---
 drivers/thermal/qcom/tsens.c | 23 ++++++++++++++++-------
 1 file changed, 16 insertions(+), 7 deletions(-)

diff --git a/drivers/thermal/qcom/tsens.c b/drivers/thermal/qcom/tsens.c
index b1162e566a70..38afde1a599f 100644
--- a/drivers/thermal/qcom/tsens.c
+++ b/drivers/thermal/qcom/tsens.c
@@ -599,6 +599,7 @@ int get_temp_tsens_valid(const struct tsens_sensor *s, int *temp)
 	int hw_id = s->hw_id;
 	u32 temp_idx = LAST_TEMP_0 + hw_id;
 	u32 valid_idx = VALID_0 + hw_id;
+	unsigned long timeout;
 	u32 valid;
 	int ret;
 
@@ -607,13 +608,21 @@ int get_temp_tsens_valid(const struct tsens_sensor *s, int *temp)
 		ret = regmap_field_read(priv->rf[valid_idx], &valid);
 		if (ret)
 			return ret;
-		while (!valid) {
-			/* Valid bit is 0 for 6 AHB clock cycles.
-			 * At 19.2MHz, 1 AHB clock is ~60ns.
-			 * We should enter this loop very, very rarely.
-			 */
-			ndelay(400);
-			ret = regmap_field_read(priv->rf[valid_idx], &valid);
+
+		if (!valid) {
+			timeout = jiffies + msecs_to_jiffies(20);
+
+			do {
+				/* Valid bit is 0 for 6 AHB clock cycles.
+				 * At 19.2MHz, 1 AHB clock is ~60ns.
+				 * We should enter this loop very, very rarely.
+				 */
+				ndelay(400);
+				ret = regmap_field_read(priv->rf[valid_idx], &valid);
+				if (valid || ret)
+					break;
+			} while (!(ret = time_after_eq(jiffies, timeout)));
+
 			if (ret)
 				return ret;
 		}
-- 
2.32.0

Re: [PATCH 1/2] drivers: thermal: tsens: fix wrong check for tzd in irq handlers

[Matthias Kaehlcke]

On Sun, Sep 05, 2021 at 07:47:07PM +0200, Ansuel Smith wrote:
> Some device can have some thermal sensor disabled from the factory. The
> current 2 irq handler functions check all the sensor by default and the
> check if the sensor was actually registered is wrong. The tzd is
> actually never set if the registration fail hence the IS_ERR check is
> wrong.

Indeed, tsens_register() doesn't assign ->tzd when the registration of
the sensor fails.

> Signed-off-by: Ansuel Smith <ansuelsmth@gmail.com>

Reviewed-by: Matthias Kaehlcke <mka@chromium.org>

Re: [PATCH 2/2] drivers: thermal: tsens: add timeout to get_tem_tsens_valid

[Matthias Kaehlcke]

On Sun, Sep 05, 2021 at 07:47:08PM +0200, Ansuel Smith wrote:
> The function can loop and lock the system if for whatever reason the bit
> for the target sensor is NEVER valid. This is the case if a sensor is
> disabled by the factory and the valid bit is never reported as actually
> valid. Add a timeout check and exit if a timeout occurs. As this is
> a very rare condition, handle the timeout only if the first read fails.
> 
> Signed-off-by: Ansuel Smith <ansuelsmth@gmail.com>
> ---
>  drivers/thermal/qcom/tsens.c | 23 ++++++++++++++++-------
>  1 file changed, 16 insertions(+), 7 deletions(-)
> 
> diff --git a/drivers/thermal/qcom/tsens.c b/drivers/thermal/qcom/tsens.c
> index b1162e566a70..38afde1a599f 100644
> --- a/drivers/thermal/qcom/tsens.c
> +++ b/drivers/thermal/qcom/tsens.c
> @@ -599,6 +599,7 @@ int get_temp_tsens_valid(const struct tsens_sensor *s, int *temp)
>  	int hw_id = s->hw_id;
>  	u32 temp_idx = LAST_TEMP_0 + hw_id;
>  	u32 valid_idx = VALID_0 + hw_id;
> +	unsigned long timeout;
>  	u32 valid;
>  	int ret;
>  
> @@ -607,13 +608,21 @@ int get_temp_tsens_valid(const struct tsens_sensor *s, int *temp)
>  		ret = regmap_field_read(priv->rf[valid_idx], &valid);
>  		if (ret)
>  			return ret;
> -		while (!valid) {
> -			/* Valid bit is 0 for 6 AHB clock cycles.
> -			 * At 19.2MHz, 1 AHB clock is ~60ns.
> -			 * We should enter this loop very, very rarely.
> -			 */
> -			ndelay(400);
> -			ret = regmap_field_read(priv->rf[valid_idx], &valid);
> +
> +		if (!valid) {
> +			timeout = jiffies + msecs_to_jiffies(20);
> +
> +			do {
> +				/* Valid bit is 0 for 6 AHB clock cycles.
> +				 * At 19.2MHz, 1 AHB clock is ~60ns.
> +				 * We should enter this loop very, very rarely.
> +				 */
> +				ndelay(400);
> +				ret = regmap_field_read(priv->rf[valid_idx], &valid);
> +				if (valid || ret)
> +					break;
> +			} while (!(ret = time_after_eq(jiffies, timeout)));
> +
>  			if (ret)
>  				return ret;

With the overloading of 'ret' the return logic is getting a bit more
convoluted. Also the function should probably return -ETIMEDOUT or
some other meaningful error if the bit is never valid.

How about keeping the 'while (!valid)' condition, and adding

	if (time_after_eq(jiffies, timeout))
		return -ETIMEDOUT;

inside the loop?

Re: [PATCH 2/2] drivers: thermal: tsens: add timeout to get_tem_tsens_valid

[Ansuel Smith]

On Tue, Sep 07, 2021 at 01:55:03PM -0700, Matthias Kaehlcke wrote:
> On Sun, Sep 05, 2021 at 07:47:08PM +0200, Ansuel Smith wrote:
> > The function can loop and lock the system if for whatever reason the bit
> > for the target sensor is NEVER valid. This is the case if a sensor is
> > disabled by the factory and the valid bit is never reported as actually
> > valid. Add a timeout check and exit if a timeout occurs. As this is
> > a very rare condition, handle the timeout only if the first read fails.
> > 
> > Signed-off-by: Ansuel Smith <ansuelsmth@gmail.com>
> > ---
> >  drivers/thermal/qcom/tsens.c | 23 ++++++++++++++++-------
> >  1 file changed, 16 insertions(+), 7 deletions(-)
> > 
> > diff --git a/drivers/thermal/qcom/tsens.c b/drivers/thermal/qcom/tsens.c
> > index b1162e566a70..38afde1a599f 100644
> > --- a/drivers/thermal/qcom/tsens.c
> > +++ b/drivers/thermal/qcom/tsens.c
> > @@ -599,6 +599,7 @@ int get_temp_tsens_valid(const struct tsens_sensor *s, int *temp)
> >  	int hw_id = s->hw_id;
> >  	u32 temp_idx = LAST_TEMP_0 + hw_id;
> >  	u32 valid_idx = VALID_0 + hw_id;
> > +	unsigned long timeout;
> >  	u32 valid;
> >  	int ret;
> >  
> > @@ -607,13 +608,21 @@ int get_temp_tsens_valid(const struct tsens_sensor *s, int *temp)
> >  		ret = regmap_field_read(priv->rf[valid_idx], &valid);
> >  		if (ret)
> >  			return ret;
> > -		while (!valid) {
> > -			/* Valid bit is 0 for 6 AHB clock cycles.
> > -			 * At 19.2MHz, 1 AHB clock is ~60ns.
> > -			 * We should enter this loop very, very rarely.
> > -			 */
> > -			ndelay(400);
> > -			ret = regmap_field_read(priv->rf[valid_idx], &valid);
> > +
> > +		if (!valid) {
> > +			timeout = jiffies + msecs_to_jiffies(20);
> > +
> > +			do {
> > +				/* Valid bit is 0 for 6 AHB clock cycles.
> > +				 * At 19.2MHz, 1 AHB clock is ~60ns.
> > +				 * We should enter this loop very, very rarely.
> > +				 */
> > +				ndelay(400);
> > +				ret = regmap_field_read(priv->rf[valid_idx], &valid);
> > +				if (valid || ret)
> > +					break;
> > +			} while (!(ret = time_after_eq(jiffies, timeout)));
> > +
> >  			if (ret)
> >  				return ret;
> 
> With the overloading of 'ret' the return logic is getting a bit more
> convoluted. Also the function should probably return -ETIMEDOUT or
> some other meaningful error if the bit is never valid.
> 
> How about keeping the 'while (!valid)' condition, and adding
> 
> 	if (time_after_eq(jiffies, timeout))
> 		return -ETIMEDOUT;
> 
> inside the loop?
>

I tried to refactor the code using some tag. Hope it's better that way
and thanks for the review. Sending v2.