lkml.org 
[lkml]   [2022]   [May]   [24]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
Patch in this message
/
From
Date
SubjectRe: [PATCH v11 2/3] phy: qcom-qmp: add regulator_set_load to dp phy
Hi,

On Fri, May 20, 2022 at 2:27 PM Kuogee Hsieh <quic_khsieh@quicinc.com> wrote:
>
> This patch add regulator_set_load() before enable regulator at
> DP phy driver.
>
> Signed-off-by: Kuogee Hsieh <quic_khsieh@quicinc.com>
> Reviewed-by: Stephen Boyd <swboyd@chromium.org>
> ---
> drivers/phy/qualcomm/phy-qcom-qmp.c | 25 ++++++++++++++++++++++++-
> 1 file changed, 24 insertions(+), 1 deletion(-)
>
> diff --git a/drivers/phy/qualcomm/phy-qcom-qmp.c b/drivers/phy/qualcomm/phy-qcom-qmp.c
> index b144ae1..a93e153 100644
> --- a/drivers/phy/qualcomm/phy-qcom-qmp.c
> +++ b/drivers/phy/qualcomm/phy-qcom-qmp.c
> @@ -3130,6 +3130,7 @@ struct qmp_phy_cfg {
> int num_resets;
> /* regulators to be requested */
> const char * const *vreg_list;
> + const unsigned int *vreg_enable_load;
> int num_vregs;
>
> /* array of registers with different offsets */
> @@ -3346,6 +3347,10 @@ static const char * const qmp_phy_vreg_l[] = {
> "vdda-phy", "vdda-pll",
> };
>
> +static const unsigned int qmp_phy_vreg_enable_load[] = {
> + 21800, 36000
> +};

I'm a little confused. Why make a new parallel structure? Don't you
want to set a load for everyone who's using "qmp_phy_vreg_l"? It seems
like it would be better to do something like this:

struct qmp_regulator_data {
const char *name;
unsigned int load;
};

struct qmp_regulator_data qmp_phy_vreg_l[] = {
{ .name = "vdda-phy", .load = 21800 },
{ .name = "vdda-pll", .load = 36000 },
};

Right now some random smattering of devices are setting the load but
not others...

> static const struct qmp_phy_cfg ipq8074_usb3phy_cfg = {
> .type = PHY_TYPE_USB3,
> .nlanes = 1,
> @@ -3711,6 +3716,7 @@ static const struct qmp_phy_cfg sc7180_usb3phy_cfg = {
> .reset_list = sc7180_usb3phy_reset_l,
> .num_resets = ARRAY_SIZE(sc7180_usb3phy_reset_l),
> .vreg_list = qmp_phy_vreg_l,
> + .vreg_enable_load = qmp_phy_vreg_enable_load,
> .num_vregs = ARRAY_SIZE(qmp_phy_vreg_l),

One downside of the parallel structures is that there's nothing
enforcing that ARRAY_SIZE(qmp_phy_vreg_l) ==
ARRAY_SIZE(qmp_phy_vreg_enable_load), though the code below relies on
it.


> @@ -6175,6 +6186,18 @@ static int qcom_qmp_phy_probe(struct platform_device *pdev)
> return ret;
> }
>
> + if (cfg->vreg_enable_load) {
> + for (i = 0; i < cfg->num_vregs; i++) {
> + ret = regulator_set_load(qmp->vregs[i].consumer,
> + cfg->vreg_enable_load[i]);
> + if (ret) {
> + dev_err(dev, "failed to set load at %s\n",
> + qmp->vregs[i].supply);

nit: indentation of the 2nd line seems a bit off?

> + return ret;
> + }
> + }
> + }

Feels like the above snippet belongs in qcom_qmp_phy_vreg_init() ?

\
 
 \ /
  Last update: 2022-05-25 02:17    [W:0.068 / U:0.552 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site