[v2,5/5] Input: elants_i2c: Delay longer with reset asserted

Commit Message

Doug Anderson Dec. 9, 2022, 2:06 a.m. UTC

The elan touchscreen datasheet says that the reset GPIO only needs to
be asserted for 500us in order to reset the regulator. The problem is
that some boards need a level shifter between the signals on the GPIO
controller and the signals on the touchscreen. All of these extra
components on the line can slow the transition of the signals. On one
board, we measured the reset line and saw that it took almost 1.8ms to
go low. Even after we bumped up the "drive strength" of the signal
from the default 2mA to 8mA we still saw it take 421us for the signal
to go low.

In order to account for this let's lengthen the amount of time that we
keep the reset asserted. Let's bump it up from 500us to 5000us.
That's still a relatively short amount of time and is much safer.

It should be noted that this fixes real problems. Case in point:
1. The touchscreen power rail may be shared with another device (like
   an eDP panel). That means that at probe time power might already be
   on.
2. In probe we grab the reset GPIO and assert it (make it low).
3. We turn on power (a noop since it was already on).
4. We wait 500us.
5. We deassert the reset GPIO.

With the above case and only a 500us delay we saw only a partial reset
asserted, which is bad. Giving it 5ms is overkill but feels safer in
case someone else has a different level shifter setup.

Note that bumping up the delay to 5000 means that some configs yell
about using udelay(). We'll change to using usleep_range(). We give a
small range here because:
- This isn't a delay that happens very often so we don't need to worry
  about giving a big range to allow for power efficiency.
- usleep_range() is known to almost always pick the upper bound and
  delay that long and we really don't want to slow down the power on
  of the touchscreen that much.

Signed-off-by: Douglas Anderson <[e-mail]>
Reviewed-by: Matthias Kaehlcke <[e-mail]>
---

Changes in v2:
- Fix typo in commit message (Matthias)
- udelay -> usleep_range (Patches Robot, Dmitry)

 drivers/input/touchscreen/elants_i2c.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

Comments

Dmitry Torokhov Dec. 9, 2022, 2:20 a.m. UTC | #1

On Thu, Dec 08, 2022 at 06:06:12PM -0800, Douglas Anderson wrote:
> The elan touchscreen datasheet says that the reset GPIO only needs to
> be asserted for 500us in order to reset the regulator. The problem is
> that some boards need a level shifter between the signals on the GPIO
> controller and the signals on the touchscreen. All of these extra
> components on the line can slow the transition of the signals. On one
> board, we measured the reset line and saw that it took almost 1.8ms to
> go low. Even after we bumped up the "drive strength" of the signal
> from the default 2mA to 8mA we still saw it take 421us for the signal
> to go low.
> 
> In order to account for this let's lengthen the amount of time that we
> keep the reset asserted. Let's bump it up from 500us to 5000us.
> That's still a relatively short amount of time and is much safer.
> 
> It should be noted that this fixes real problems. Case in point:
> 1. The touchscreen power rail may be shared with another device (like
>    an eDP panel). That means that at probe time power might already be
>    on.
> 2. In probe we grab the reset GPIO and assert it (make it low).
> 3. We turn on power (a noop since it was already on).
> 4. We wait 500us.
> 5. We deassert the reset GPIO.
> 
> With the above case and only a 500us delay we saw only a partial reset
> asserted, which is bad. Giving it 5ms is overkill but feels safer in
> case someone else has a different level shifter setup.
> 
> Note that bumping up the delay to 5000 means that some configs yell
> about using udelay(). We'll change to using usleep_range(). We give a
> small range here because:
> - This isn't a delay that happens very often so we don't need to worry
>   about giving a big range to allow for power efficiency.
> - usleep_range() is known to almost always pick the upper bound and
>   delay that long and we really don't want to slow down the power on
>   of the touchscreen that much.
> 
> Signed-off-by: Douglas Anderson <[e-mail]>
> Reviewed-by: Matthias Kaehlcke <[e-mail]>
> ---
> 
> Changes in v2:
> - Fix typo in commit message (Matthias)
> - udelay -> usleep_range (Patches Robot, Dmitry)
> 
>  drivers/input/touchscreen/elants_i2c.c | 4 ++--
>  1 file changed, 2 insertions(+), 2 deletions(-)
> 
> diff --git a/drivers/input/touchscreen/elants_i2c.c b/drivers/input/touchscreen/elants_i2c.c
> index 879a4d984c90..192d543e5aa9 100644
> --- a/drivers/input/touchscreen/elants_i2c.c
> +++ b/drivers/input/touchscreen/elants_i2c.c
> @@ -114,7 +114,7 @@
>  /* calibration timeout definition */
>  #define ELAN_CALI_TIMEOUT_MSEC	12000
>  
> -#define ELAN_POWERON_DELAY_USEC	500
> +#define ELAN_POWERON_DELAY_USEC	5000
>  #define ELAN_RESET_DELAY_MSEC	20
>  
>  /* FW boot code version */
> @@ -1352,7 +1352,7 @@ static int elants_i2c_power_on(struct elants_data *ts)
>  	 * We need to wait a bit after powering on controller before
>  	 * we are allowed to release reset GPIO.
>  	 */
> -	udelay(ELAN_POWERON_DELAY_USEC);
> +	usleep_range(ELAN_POWERON_DELAY_USEC, ELAN_POWERON_DELAY_USEC + 100);
>  
>  release_reset_gpio:
>  	gpiod_set_value_cansleep(ts->reset_gpio, 0);

This gives me conflict because this label is gone in my tree, so I
adjusted for context and applied.

Thanks.

Doug Anderson Dec. 9, 2022, 2:24 a.m. UTC | #2

Hi,

On Thu, Dec 8, 2022 at 6:21 PM Dmitry Torokhov
<[e-mail]> wrote:
>
> On Thu, Dec 08, 2022 at 06:06:12PM -0800, Douglas Anderson wrote:
> > The elan touchscreen datasheet says that the reset GPIO only needs to
> > be asserted for 500us in order to reset the regulator. The problem is
> > that some boards need a level shifter between the signals on the GPIO
> > controller and the signals on the touchscreen. All of these extra
> > components on the line can slow the transition of the signals. On one
> > board, we measured the reset line and saw that it took almost 1.8ms to
> > go low. Even after we bumped up the "drive strength" of the signal
> > from the default 2mA to 8mA we still saw it take 421us for the signal
> > to go low.
> >
> > In order to account for this let's lengthen the amount of time that we
> > keep the reset asserted. Let's bump it up from 500us to 5000us.
> > That's still a relatively short amount of time and is much safer.
> >
> > It should be noted that this fixes real problems. Case in point:
> > 1. The touchscreen power rail may be shared with another device (like
> >    an eDP panel). That means that at probe time power might already be
> >    on.
> > 2. In probe we grab the reset GPIO and assert it (make it low).
> > 3. We turn on power (a noop since it was already on).
> > 4. We wait 500us.
> > 5. We deassert the reset GPIO.
> >
> > With the above case and only a 500us delay we saw only a partial reset
> > asserted, which is bad. Giving it 5ms is overkill but feels safer in
> > case someone else has a different level shifter setup.
> >
> > Note that bumping up the delay to 5000 means that some configs yell
> > about using udelay(). We'll change to using usleep_range(). We give a
> > small range here because:
> > - This isn't a delay that happens very often so we don't need to worry
> >   about giving a big range to allow for power efficiency.
> > - usleep_range() is known to almost always pick the upper bound and
> >   delay that long and we really don't want to slow down the power on
> >   of the touchscreen that much.
> >
> > Signed-off-by: Douglas Anderson <[e-mail]>
> > Reviewed-by: Matthias Kaehlcke <[e-mail]>
> > ---
> >
> > Changes in v2:
> > - Fix typo in commit message (Matthias)
> > - udelay -> usleep_range (Patches Robot, Dmitry)
> >
> >  drivers/input/touchscreen/elants_i2c.c | 4 ++--
> >  1 file changed, 2 insertions(+), 2 deletions(-)

Ah, right. I posted it against the Qualcomm tree which, of course,
doesn't have the previous patch I posted to this driver. Thanks for
fixing it up. :-)

Patch

diff --git a/drivers/input/touchscreen/elants_i2c.c b/drivers/input/touchscreen/elants_i2c.c
index 879a4d984c90..192d543e5aa9 100644
--- a/drivers/input/touchscreen/elants_i2c.c
+++ b/drivers/input/touchscreen/elants_i2c.c
@@ -114,7 +114,7 @@ 
 /* calibration timeout definition */
 #define ELAN_CALI_TIMEOUT_MSEC	12000
 
-#define ELAN_POWERON_DELAY_USEC	500
+#define ELAN_POWERON_DELAY_USEC	5000
 #define ELAN_RESET_DELAY_MSEC	20
 
 /* FW boot code version */
@@ -1352,7 +1352,7 @@  static int elants_i2c_power_on(struct elants_data *ts)
 	 * We need to wait a bit after powering on controller before
 	 * we are allowed to release reset GPIO.
 	 */
-	udelay(ELAN_POWERON_DELAY_USEC);
+	usleep_range(ELAN_POWERON_DELAY_USEC, ELAN_POWERON_DELAY_USEC + 100);
 
 release_reset_gpio:
 	gpiod_set_value_cansleep(ts->reset_gpio, 0);