| Message ID | 20220818082316.2717095-1-daniel.lezcano@linaro.org (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
| Series | [v2,1/2] thermal/drivers/imx_sc: Rely on the platform data to get the resource id | expand |
On 18/08/2022 10:23, Daniel Lezcano wrote: > Currently the imx_sc driver is reimplementing part of the thermal zone > parsing from the thermal OF tree code to get the sensor id associated > with a thermal zone sensor. > > The driver platform specific code should know what sensor is present > and not rely on the thermal zone description to do a discovery. Well > that is arguable but all the other drivers have a per platform data > telling what sensor id to use. > > The imx_sc thermal driver is the only one using a different > approach. Not invalid but forcing to keep a specific function > 'thermal_zone_of_get_sensor_id()' to get the sensor id for a specific > thermal zone as the self-explanatory function tells and having device > tree code inside the driver. > > The thermal OF code had a rework and remains now self-encapsulated > with a register/unregister functions and their 'devm' variants, except > for the function mentioned above. > > After investigating, it appears the imx_sc sensor is defined in > arch/arm64/boot/dts/freescale/imx8qxp.dtsi: > > which defines the cpu-thermal zone with the id: IMX_SC_R_SYSTEM > > This dtsi is included by: > - imx8qxp-ai_ml.dts > - imx8qxp-colibri.dtsi > - imx8qxp-mek.dts > > The two first ones do not define more thermal zones > The third one adds the pmic-thermal0 zone with id: IMX_SC_R_PMIC_0 > > The thermal OF code returns -ENODEV if the thermal zone registration > with a specific id fails because the description is not available in > the DT for such a sensor id. In this case we continue with the other > ids without bailing out with an error. > > So we can build for the 'fsl,imx-sc-thermal' a compatible data, an > array of sensor ids containing IMX_SC_R_SYSTEM and IMX_SC_R_PMIC_0. > > The latter won't be found but that will not result in an error but a > normal case where we continue the initialization with other ids. > > Just to clarify, it is what the thermal framework does and what the > other drivers are expecting: when a registration fails with -ENODEV > this is not an error but a case where the description is not found in > the device tree, that be can the entire thermal zones description or a > specific thermal zone with an unknown id. > > There is one small functional change but without impact. When there is > no 'thermal-zones' description the probe function was returning > '-ENODEV', now it returns zero. When a thermal zone fails to register > with an error different from '-ENODEV', the error is detected and > returned. > > Change the code accordingly and remove the OF code from the driver. > > Signed-off-by: Daniel Lezcano <daniel.lezcano@linaro.org> > --- > The changes apply on the linux-next branch: > > https://git.kernel.org/pub/scm/linux/kernel/git/thermal/linux.git/log/?h=thermal/linux-next > > I don't have the platform, so I was not able to test the changes. > > Changelog: > v2: > - Clarified a bit more the changes description > > - Fix use-after-free and add a couple of comments to reflect the > patch description > > - Put back the error message in case there is an error with the > registration > Is there any comment on this change ?
Shawn, Sascha, if there is no comment on these two patches. I'll pick them up Thanks -- D. On 26/08/2022 00:28, Daniel Lezcano wrote: > On 18/08/2022 10:23, Daniel Lezcano wrote: >> Currently the imx_sc driver is reimplementing part of the thermal zone >> parsing from the thermal OF tree code to get the sensor id associated >> with a thermal zone sensor. >> >> The driver platform specific code should know what sensor is present >> and not rely on the thermal zone description to do a discovery. Well >> that is arguable but all the other drivers have a per platform data >> telling what sensor id to use. >> >> The imx_sc thermal driver is the only one using a different >> approach. Not invalid but forcing to keep a specific function >> 'thermal_zone_of_get_sensor_id()' to get the sensor id for a specific >> thermal zone as the self-explanatory function tells and having device >> tree code inside the driver. >> >> The thermal OF code had a rework and remains now self-encapsulated >> with a register/unregister functions and their 'devm' variants, except >> for the function mentioned above. >> >> After investigating, it appears the imx_sc sensor is defined in >> arch/arm64/boot/dts/freescale/imx8qxp.dtsi: >> >> which defines the cpu-thermal zone with the id: IMX_SC_R_SYSTEM >> >> This dtsi is included by: >> - imx8qxp-ai_ml.dts >> - imx8qxp-colibri.dtsi >> - imx8qxp-mek.dts >> >> The two first ones do not define more thermal zones >> The third one adds the pmic-thermal0 zone with id: IMX_SC_R_PMIC_0 >> >> The thermal OF code returns -ENODEV if the thermal zone registration >> with a specific id fails because the description is not available in >> the DT for such a sensor id. In this case we continue with the other >> ids without bailing out with an error. >> >> So we can build for the 'fsl,imx-sc-thermal' a compatible data, an >> array of sensor ids containing IMX_SC_R_SYSTEM and IMX_SC_R_PMIC_0. >> >> The latter won't be found but that will not result in an error but a >> normal case where we continue the initialization with other ids. >> >> Just to clarify, it is what the thermal framework does and what the >> other drivers are expecting: when a registration fails with -ENODEV >> this is not an error but a case where the description is not found in >> the device tree, that be can the entire thermal zones description or a >> specific thermal zone with an unknown id. >> >> There is one small functional change but without impact. When there is >> no 'thermal-zones' description the probe function was returning >> '-ENODEV', now it returns zero. When a thermal zone fails to register >> with an error different from '-ENODEV', the error is detected and >> returned. >> >> Change the code accordingly and remove the OF code from the driver. >> >> Signed-off-by: Daniel Lezcano <daniel.lezcano@linaro.org> >> --- >> The changes apply on the linux-next branch: >> >> >> https://git.kernel.org/pub/scm/linux/kernel/git/thermal/linux.git/log/?h=thermal/linux-next >> >> >> I don't have the platform, so I was not able to test the changes. >> >> Changelog: >> v2: >> - Clarified a bit more the changes description >> >> - Fix use-after-free and add a couple of comments to reflect the >> patch description >> >> - Put back the error message in case there is an error with the >> registration >> > > Is there any comment on this change ? > > >
diff --git a/drivers/thermal/imx_sc_thermal.c b/drivers/thermal/imx_sc_thermal.c index 10bfa6507eb4..bba5c730b855 100644 --- a/drivers/thermal/imx_sc_thermal.c +++ b/drivers/thermal/imx_sc_thermal.c @@ -76,59 +76,55 @@ static const struct thermal_zone_device_ops imx_sc_thermal_ops = { static int imx_sc_thermal_probe(struct platform_device *pdev) { - struct device_node *np, *child, *sensor_np; struct imx_sc_sensor *sensor; - int ret; + const int *resource_id; + int i, ret; ret = imx_scu_get_handle(&thermal_ipc_handle); if (ret) return ret; - np = of_find_node_by_name(NULL, "thermal-zones"); - if (!np) - return -ENODEV; + resource_id = of_device_get_match_data(&pdev->dev); + if (!resource_id) + return -EINVAL; - sensor_np = of_node_get(pdev->dev.of_node); + for (i = 0; resource_id[i] > 0; i++) { - for_each_available_child_of_node(np, child) { sensor = devm_kzalloc(&pdev->dev, sizeof(*sensor), GFP_KERNEL); - if (!sensor) { - of_node_put(child); - ret = -ENOMEM; - goto put_node; - } + if (!sensor) + return -ENOMEM; - ret = thermal_zone_of_get_sensor_id(child, - sensor_np, - &sensor->resource_id); - if (ret < 0) { - dev_err(&pdev->dev, - "failed to get valid sensor resource id: %d\n", - ret); - of_node_put(child); - break; - } + sensor->resource_id = resource_id[i]; - sensor->tzd = devm_thermal_of_zone_register(&pdev->dev, - sensor->resource_id, - sensor, - &imx_sc_thermal_ops); + sensor->tzd = devm_thermal_of_zone_register(&pdev->dev, sensor->resource_id, + sensor, &imx_sc_thermal_ops); if (IS_ERR(sensor->tzd)) { - dev_err(&pdev->dev, "failed to register thermal zone\n"); + /* + * Save the error value before freeing the + * sensor pointer, otherwise we endup with a + * use-after-free error + */ ret = PTR_ERR(sensor->tzd); - of_node_put(child); - break; + + devm_kfree(&pdev->dev, sensor); + + /* + * The thermal framework notifies us there is + * no thermal zone description for such a + * sensor id + */ + if (ret == -ENODEV) + continue; + + dev_err(&pdev->dev, "failed to register thermal zone\n"); + return ret; } if (devm_thermal_add_hwmon_sysfs(sensor->tzd)) dev_warn(&pdev->dev, "failed to add hwmon sysfs attributes\n"); } -put_node: - of_node_put(sensor_np); - of_node_put(np); - - return ret; + return 0; } static int imx_sc_thermal_remove(struct platform_device *pdev) @@ -136,8 +132,10 @@ static int imx_sc_thermal_remove(struct platform_device *pdev) return 0; } +static int imx_sc_sensors[] = { IMX_SC_R_SYSTEM, IMX_SC_R_PMIC_0, -1 }; + static const struct of_device_id imx_sc_thermal_table[] = { - { .compatible = "fsl,imx-sc-thermal", }, + { .compatible = "fsl,imx-sc-thermal", .data = imx_sc_sensors }, {} }; MODULE_DEVICE_TABLE(of, imx_sc_thermal_table);