[<prev] [next>] [<thread-prev] [day] [month] [year] [list]
Message-ID: <aIGddOWlz6LTAxgV@yuanjiey.ap.qualcomm.com>
Date: Thu, 24 Jul 2025 10:41:56 +0800
From: yuanjiey <yuanjie.yang@....qualcomm.com>
To: Dmitry Baryshkov <dmitry.baryshkov@....qualcomm.com>
Cc: andersson@...nel.org, linus.walleij@...aro.org,
linux-arm-msm@...r.kernel.org, linux-gpio@...r.kernel.org,
linux-kernel@...r.kernel.org, kernel@....qualcomm.com,
quic_tingweiz@...cinc.com, quic_yuanjiey@...cinc.com
Subject: Re: [PATCH v2] pinctrl: qcom: Fix logic error when TLMM reg-names
property is missing
On Wed, Jul 23, 2025 at 02:27:46PM +0300, Dmitry Baryshkov wrote:
> On Wed, Jul 23, 2025 at 03:50:24PM +0800, yuanjie yang wrote:
> > From: Yuanjie Yang <yuanjie.yang@....qualcomm.com>
> >
> > Some Qualcomm platforms, such as sm8750.dtsi, define a single TLMM
> > region without the reg-names property. This is a valid and expected
> > configuration. However, the current code incorrectly treats the absence
> > of reg-names as an error, resulting in unintended behavior.
> >
> > Refactor the logic to handle both cases correctly:
> > If only the gpio parameter is provided, default to TLMM region 0.
> > If both gpio and name are provided, compare the reg-names entries in the
> > TLMM node with the given name to select the appropriate region.
> >
> > Fixes: 56ffb63749f4 ("pinctrl: qcom: add multi TLMM region option parameter")
> >
> > Signed-off-by: Yuanjie Yang <yuanjie.yang@....qualcomm.com>
> >
> > ---
> > - optimize commit message
> > - optimize code structure and fix kfree issue
> > - rebase on tag: next-20250723
> > - Link to v1: https://lore.kernel.org/all/20250722054446.3432-1-yuanjie.yang@oss.qualcomm.com/
> >
> > ---
> > drivers/pinctrl/qcom/tlmm-test.c | 22 +++++++++++-----------
> > 1 file changed, 11 insertions(+), 11 deletions(-)
> >
> > diff --git a/drivers/pinctrl/qcom/tlmm-test.c b/drivers/pinctrl/qcom/tlmm-test.c
> > index 7d7fff538755..e41fe7e76c80 100644
> > --- a/drivers/pinctrl/qcom/tlmm-test.c
> > +++ b/drivers/pinctrl/qcom/tlmm-test.c
> > @@ -581,10 +581,13 @@ static int tlmm_reg_base(struct device_node *tlmm, struct resource *res)
> > int ret;
> > int i;
> >
> > + if (!strcmp(tlmm_reg_name, "default_region"))
> > + return of_address_to_resource(tlmm, 0, res);
> > +
> > count = of_property_count_strings(tlmm, "reg-names");
> > if (count <= 0) {
> > pr_err("failed to find tlmm reg name\n");
> > - return count;
> > + return -EINVAL;
>
> Why? It's better to propagate the error instead of reinventing it.
of_property_count_strings often return value > 0 (count) or value < 0 (error code).
But theoretically the return value can equal 0 (value == 0), perhaps this situation
will not happen.
However, from the code logic perspective, this tool should early return when count<=0,
if return count == 0, code flow can not stop, this will lead to a mistake in the tool.
line:
639:ret = tlmm_reg_base(tlmm, &res);
640:if (ret < 0)
641: return ret;
With this consideration, I have reinvented the error code.
> > }
> >
> > reg_names = kcalloc(count, sizeof(char *), GFP_KERNEL);
> > @@ -597,19 +600,16 @@ static int tlmm_reg_base(struct device_node *tlmm, struct resource *res)
> > return -EINVAL;
> > }
> >
> > - if (!strcmp(tlmm_reg_name, "default_region")) {
> > - ret = of_address_to_resource(tlmm, 0, res);
> > - } else {
> > - for (i = 0; i < count; i++) {
> > - if (!strcmp(reg_names[i], tlmm_reg_name)) {
> > - ret = of_address_to_resource(tlmm, i, res);
> > - break;
> > - }
> > + for (i = 0; i < count; i++) {
> > + if (!strcmp(reg_names[i], tlmm_reg_name)) {
> > + ret = of_address_to_resource(tlmm, i, res);
> > + break;
> > }
> > - if (i == count)
> > - ret = -EINVAL;
> > }
> >
> > + if (i == count)
> > + ret = -EINVAL;
> > +
> > kfree(reg_names);
> >
> > return ret;
> >
> > base-commit: a933d3dc1968fcfb0ab72879ec304b1971ed1b9a
> > --
> > 2.34.1
> >
>
> --
> With best wishes
> Dmitry
Thanks,
Yuanjie
Powered by blists - more mailing lists