Hi Heiko, Jonathan
we are adding the notifier for this patch, indeed for future consideration.
Thanks.
在 2021/8/5 下午8:07, Heiko Stübner 写道:
Hi David,
Am Donnerstag, 5. August 2021, 13:56:18 CEST schrieb David Wu:
Hi Jonathan,
在 2021/8/3 下午8:51, Jonathan Cameron 写道:
On Tue, 3 Aug 2021 11:09:47 +0800
David Wu <david.wu@xxxxxxxxxxxxxx> wrote:
Hi Jonathan,
在 2021/8/2 下午6:42, Jonathan Cameron 写道:
On Mon, 2 Aug 2021 17:09:29 +0800
Simon Xue <xxm@xxxxxxxxxxxxxx> wrote:
From: David Wu <david.wu@xxxxxxxxxxxxxx>
The referenced voltage is not changed after initiation, so just only
get referenced voltage once.
Hi David,
Isn't this an external reference voltage? If so how do you know
it is not changed at runtime? It might be unlikely and not happen
on particular platforms, but that's not he same as saying it can never
happen. Clearly it's racey anyway if that does happen, but we definitely
don't expect frequent voltage changes.
The current regulator is not changed and not subject to external
changes, this can reduce the getting voltage. Assuming that there will
be changes in the future, we then add the notify of the regulator, so
that the voltage change can be obtained.
If this patch added the notifier that would be a nice solution, but
right now it potentially introduced a regression. You have made me a little curious...
Are you seeing a significant cost to querying that regulator voltage?
If so, I'd imagine it's a lack of caching in the regulator driver or similar.
Scale readback via sysfs shouldn't be in a fast path anyway.
You can't depend on what boards today do, because someone with a board
built tomorrow may well use an old kernel which supports the voltage
changing, and then see a regression when they upgrade to the kernel
containing this patch.
For all current chips, the expected voltage is a fixed voltage, and
don't want to change it in any process.:-)
So, if the voltage here does not change, then it can be obtained once in
probe(), which can save the time of each acquisition. For example, the
voltage of this regulator is obtained through i2c, which will increase
some consumption every time.
Jonathans request was to not think about "all current chips" but the
general case, and as Jonathan said, adding that regulator notifier you
already mentioned would be the nicest solution.
So I think the easiest way is to just add the voltage notifier to your patch
to make everyone happy ;-) .
Heiko
Jonathan
Jonathan
Signed-off-by: Simon Xue <xxm@xxxxxxxxxxxxxx>
Signed-off-by: David Wu <david.wu@xxxxxxxxxxxxxx>
---
drivers/iio/adc/rockchip_saradc.c | 16 +++++++++-------
1 file changed, 9 insertions(+), 7 deletions(-)
diff --git a/drivers/iio/adc/rockchip_saradc.c b/drivers/iio/adc/rockchip_saradc.c
index f3eb8d2e50dc..cd33c0b9d3eb 100644
--- a/drivers/iio/adc/rockchip_saradc.c
+++ b/drivers/iio/adc/rockchip_saradc.c
@@ -49,6 +49,7 @@ struct rockchip_saradc {
struct clk *clk;
struct completion completion;
struct regulator *vref;
+ int uv_vref;
struct reset_control *reset;
const struct rockchip_saradc_data *data;
u16 last_val;
@@ -105,13 +106,7 @@ static int rockchip_saradc_read_raw(struct iio_dev *indio_dev,
mutex_unlock(&indio_dev->mlock);
return IIO_VAL_INT;
case IIO_CHAN_INFO_SCALE:
- ret = regulator_get_voltage(info->vref);
- if (ret < 0) {
- dev_err(&indio_dev->dev, "failed to get voltage\n");
- return ret;
- }
-
- *val = ret / 1000;
+ *val = info->uv_vref / 1000;
*val2 = chan->scan_type.realbits;
return IIO_VAL_FRACTIONAL_LOG2;
default:
@@ -410,6 +405,13 @@ static int rockchip_saradc_probe(struct platform_device *pdev)
return ret;
}
+ info->uv_vref = regulator_get_voltage(info->vref);
+ if (info->uv_vref < 0) {
+ dev_err(&pdev->dev, "failed to get voltage\n");
+ ret = info->uv_vref;
+ return ret;
+ }
+
ret = clk_prepare_enable(info->pclk);
if (ret < 0) {
dev_err(&pdev->dev, "failed to enable pclk\n");
