2011-03-29 21:29:14

by Mark Brown

[permalink] [raw]
Subject: [PATCH] regulator: When constraining modes fall back to higher power modes

If a mode requested by a consumer is not allowed by constraints
automatically fall back to a higher power mode if possible. This
ensures that consumers get at least the output they requested while
allowing machine drivers to transparently limit lower power modes
if required.

Signed-off-by: Mark Brown <[email protected]>
---

I've not actually had a chance to test this but throwing it out there
for comment and testing now; someone with an OMAP board can probably
test fairly quickly as the OMAP HSMMC driver is using set_mode().

drivers/regulator/core.c | 24 +++++++++++++++---------
1 files changed, 15 insertions(+), 9 deletions(-)

diff --git a/drivers/regulator/core.c b/drivers/regulator/core.c
index 3ffc697..a634946 100644
--- a/drivers/regulator/core.c
+++ b/drivers/regulator/core.c
@@ -197,9 +197,9 @@ static int regulator_check_current_limit(struct regulator_dev *rdev,
}

/* operating mode constraint check */
-static int regulator_check_mode(struct regulator_dev *rdev, int mode)
+static int regulator_mode_constrain(struct regulator_dev *rdev, int *mode)
{
- switch (mode) {
+ switch (*mode) {
case REGULATOR_MODE_FAST:
case REGULATOR_MODE_NORMAL:
case REGULATOR_MODE_IDLE:
@@ -217,11 +217,17 @@ static int regulator_check_mode(struct regulator_dev *rdev, int mode)
rdev_err(rdev, "operation not allowed\n");
return -EPERM;
}
- if (!(rdev->constraints->valid_modes_mask & mode)) {
- rdev_err(rdev, "invalid mode %x\n", mode);
- return -EINVAL;
+
+ /* The modes are bitmasks, the most power hungry modes having
+ * the lowest values. If the requested mode isn't supported
+ * try higher modes. */
+ while (*mode) {
+ if (rdev->constraints->valid_modes_mask & *mode)
+ return 0;
+ *mode /= 2;
}
- return 0;
+
+ return -EINVAL;
}

/* dynamic regulator mode switching constraint check */
@@ -612,7 +618,7 @@ static void drms_uA_update(struct regulator_dev *rdev)
output_uV, current_uA);

/* check the new mode is allowed */
- err = regulator_check_mode(rdev, mode);
+ err = regulator_mode_constrain(rdev, &mode);
if (err == 0)
rdev->desc->ops->set_mode(rdev, mode);
}
@@ -2005,7 +2011,7 @@ int regulator_set_mode(struct regulator *regulator, unsigned int mode)
}

/* constraints check */
- ret = regulator_check_mode(rdev, mode);
+ ret = regulator_mode_constrain(rdev, mode);
if (ret < 0)
goto out;

@@ -2116,7 +2122,7 @@ int regulator_set_optimum_mode(struct regulator *regulator, int uA_load)
mode = rdev->desc->ops->get_optimum_mode(rdev,
input_uV, output_uV,
total_uA_load);
- ret = regulator_check_mode(rdev, mode);
+ ret = regulator_mode_constrain(rdev, &mode);
if (ret < 0) {
rdev_err(rdev, "failed to get optimum mode @ %d uA %d -> %d uV\n",
total_uA_load, input_uV, output_uV);
--
1.7.4.1


2011-03-29 23:29:30

by Liam Girdwood

[permalink] [raw]
Subject: Re: [PATCH] regulator: When constraining modes fall back to higher power modes

On Wed, 2011-03-30 at 06:29 +0900, Mark Brown wrote:
> If a mode requested by a consumer is not allowed by constraints
> automatically fall back to a higher power mode if possible. This
> ensures that consumers get at least the output they requested while
> allowing machine drivers to transparently limit lower power modes
> if required.
>
> Signed-off-by: Mark Brown <[email protected]>
> ---
>
> I've not actually had a chance to test this but throwing it out there
> for comment and testing now; someone with an OMAP board can probably
> test fairly quickly as the OMAP HSMMC driver is using set_mode().
>
> drivers/regulator/core.c | 24 +++++++++++++++---------
> 1 files changed, 15 insertions(+), 9 deletions(-)
>
> diff --git a/drivers/regulator/core.c b/drivers/regulator/core.c
> index 3ffc697..a634946 100644
> --- a/drivers/regulator/core.c
> +++ b/drivers/regulator/core.c
> @@ -197,9 +197,9 @@ static int regulator_check_current_limit(struct regulator_dev *rdev,
> }
>
> /* operating mode constraint check */
> -static int regulator_check_mode(struct regulator_dev *rdev, int mode)
> +static int regulator_mode_constrain(struct regulator_dev *rdev, int *mode)
> {
> - switch (mode) {
> + switch (*mode) {
> case REGULATOR_MODE_FAST:
> case REGULATOR_MODE_NORMAL:
> case REGULATOR_MODE_IDLE:
> @@ -217,11 +217,17 @@ static int regulator_check_mode(struct regulator_dev *rdev, int mode)
> rdev_err(rdev, "operation not allowed\n");
> return -EPERM;
> }
> - if (!(rdev->constraints->valid_modes_mask & mode)) {
> - rdev_err(rdev, "invalid mode %x\n", mode);
> - return -EINVAL;
> +
> + /* The modes are bitmasks, the most power hungry modes having
> + * the lowest values. If the requested mode isn't supported
> + * try higher modes. */
> + while (*mode) {
> + if (rdev->constraints->valid_modes_mask & *mode)
> + return 0;
> + *mode /= 2;
> }
> - return 0;
> +
> + return -EINVAL;
> }

It's late and I'm wondering if it's cleaner here just to :-

if (mask & *mode)
return 0;

*mode = fls(mask);
if (*mode)
return 0;

return -EINVAL;


What do you think ?

Liam

2011-03-29 23:39:03

by Mark Brown

[permalink] [raw]
Subject: Re: [PATCH] regulator: When constraining modes fall back to higher power modes

On Wed, Mar 30, 2011 at 12:29:24AM +0100, Liam Girdwood wrote:

> It's late and I'm wondering if it's cleaner here just to :-
>
> if (mask & *mode)
> return 0;

> *mode = fls(mask);
> if (*mode)
> return 0;

This will end up selecting the highest mode in mask, even if it's lower
than the one asked for, you'd need to recheck against the original mask.
There's a few ways I can think of to do that but they're none of them
exceptionally pretty.

2011-03-30 19:14:52

by Liam Girdwood

[permalink] [raw]
Subject: Re: [PATCH] regulator: When constraining modes fall back to higher power modes

On Wed, 2011-03-30 at 06:29 +0900, Mark Brown wrote:
> If a mode requested by a consumer is not allowed by constraints
> automatically fall back to a higher power mode if possible. This
> ensures that consumers get at least the output they requested while
> allowing machine drivers to transparently limit lower power modes
> if required.
>
> Signed-off-by: Mark Brown <[email protected]>
> ---

Applied.

Thanks

Liam