Re: [PATCH v3] platform/x86: lenovo-wmi-other: Deconflict battery with thinkpad_acpi

From: Rong Zhang

Date: Wed Nov 18 2026 - 06:55:04 EST


Hi Aaron,

On Wed, 2026-11-18 at 14:27 +0800, Aaron Ma wrote:
> Hi,
>
> On Fri, Nov 13, 2026 at 7:04 PM Ilpo Järvinen
> <ilpo.jarvinen@xxxxxxxxxxxxxxx> wrote:
> >
> > On Fri, 13 Nov 2026, Aaron Ma wrote:
> >
> > > ThinkPads expose battery charge thresholds and charge_behaviour through
> > > thinkpad_acpi's ACPI battery hook using raw sysfs attribute groups. The
> > > Lenovo WMI "Other Mode" driver also registers a power_supply_ext for
> > > charge_behaviour and charge_types on systems advertising those WMI
> > > capabilities.
> > >
> > > When lenovo-wmi-other registers first, thinkpad_acpi fails to create a
> > > duplicate charge_behaviour sysfs attribute and unregisters its battery
> > > extension. Conversely, registering or updating any power_supply_ext on
> > > the same battery runs sysfs_update_groups(), which removes standard
> > > power-supply attributes not owned by power_supply_core and wipes out
> > > thinkpad_acpi's charge_control_start_threshold,
> > > charge_control_end_threshold, and charge_behaviour files.
> > >
> > > Deconflict both ideapad_laptop (VPC2004 with GBMD and SBMC) and
> > > thinkpad_acpi (LEN0268, LEN0068, or IBM0068 with BDSG and BICG) via a
> > > per-driver table guarded by IS_ENABLED() in Kconfig, and update the
> > > force_load_psy_ext description accordingly.
> > >
> > > Fixes: 9ca8fc065b88b ("platform/x86: lenovo-wmi-other: Add WMI battery charge limiting")
> > > Reviewed-by: Derek J. Clark <derekjohn.clark@xxxxxxxxx>
> > > Co-developed-by: Rong Zhang <i@xxxxxxxx>
> > > Signed-off-by: Rong Zhang <i@xxxxxxxx>
> > > Signed-off-by: Aaron Ma <aaron.ma@xxxxxxxxxxxxx>
> > > ---
> > > Changes in v3:
> > > - Wrap deconfliction table entries in #if IS_ENABLED(CONFIG_*) and drop
> > > driver_enabled, as suggested by Rong Zhang.
> > > - Group deconfliction entries per driver (HIDs list + methods list).
> > >
> > > Changes in v2:
> > > - Consolidate IdeaPad and ThinkPad deconfliction into a table of ACPI
> > > device IDs, method pairs, and driver configuration checks, as suggested
> > > by Derek J. Clark.
> > > - Check GBMD and SBMC for IdeaPads, and BDSG and BICG for ThinkPads,
> > > instead of relying only on ThinkPad ACPI device presence.
> > > - Check whether the corresponding driver is enabled before looking up
> > > its ACPI device, including for ideapad_laptop.
> > > - Update the force_load_psy_ext description to cover both drivers.
> > >
> > > v2: https://lore.kernel.org/platform-driver-x86/20261112074823.853819-1-aaron.ma@xxxxxxxxxxxxx/
> > > v1: https://lore.kernel.org/platform-driver-x86/20261111083529.727285-1-aaron.ma@xxxxxxxxxxxxx/
> > >
> > > drivers/platform/x86/lenovo/wmi-other.c | 66 ++++++++++++++++++++-----
> > > 1 file changed, 53 insertions(+), 13 deletions(-)
> > >
> > > diff --git a/drivers/platform/x86/lenovo/wmi-other.c b/drivers/platform/x86/lenovo/wmi-other.c
> > > index bd64e7b44669d..340159df5f7cc 100644
> > > --- a/drivers/platform/x86/lenovo/wmi-other.c
> > > +++ b/drivers/platform/x86/lenovo/wmi-other.c
> > > @@ -189,10 +189,10 @@ MODULE_PARM_DESC(relax_fan_constraint,
> > > static bool force_load_psy_ext;
> > > module_param(force_load_psy_ext, bool, 0444);
> > > MODULE_PARM_DESC(force_load_psy_ext,
> > > - "This option will skip checking if the ideapad_laptop driver will conflict "
> > > + "This option will skip checking if the ideapad_laptop or thinkpad_acpi driver will conflict "
> > > "with adding an extension to set the battery charge behavior and battery charge "
> > > "control end threshold. It will also skip checking if the BIOS reports that "
> > > - "those features are fully supported. It is recommended to blacklist the ideapad "
> > > + "those features are fully supported. It is recommended to blacklist the conflicting "
> > > "driver before using this option.");
> > >
> > > /* ======== HWMON (component: lenovo-wmi-capdata 00 & fan) ======== */
> > > @@ -905,17 +905,41 @@ static acpi_status lwmi_acpi_match(acpi_handle handle, u32 lvl,
> > > return AE_CTRL_TERMINATE;
> > > }
> > >
> > > +struct lwmi_psy_deconflict {
> > > + char * const *hids;
> >
> > Who is only the pointer const, not the char?
> >
>
> It only lacked const because it shared LWMI_PSY_DECONFLICT_STRS() with
> methods. I made hids const char * const * in v4.
>
> > > + char * const *methods;
> >
> > If acpi_has_method would input const char * (to me it seems convertable
> > as it only calls acpi_get_handle that takes const char *), this too could
> > be const char.
> >
>
> Agreed. acpi_has_method() still takes char *, so I kept methods as
> char * const * in v4.
>
> > > +};
> > > +
> > > +#if IS_ENABLED(CONFIG_IDEAPAD_LAPTOP) || IS_ENABLED(CONFIG_THINKPAD_ACPI)
> > > +#define LWMI_PSY_DECONFLICT_STRS(...) \
> > > + ((char * const[]) { __VA_ARGS__, NULL })
> > > +#endif
> >
> > What exactly does ifdef'ing this construct achieve? Does something hunt
> > unused macros?
> >
>
> Yes, W=2 enables -Wunused-macros. The guard avoids the warning when
> both drivers are disabled.

Usually we don't particularly care about W=2, otherwise our lives would
be too tough to continue.

W=2/3 only makes sense when you are looking for potential bugs or
optimizations. If the code doesn't fall into either category, the
warnings can be safely ignored.

Thanks,
Rong

>
> > > +
> > > +static const struct lwmi_psy_deconflict lwmi_psy_deconflict_list[] = {
> > > +#if IS_ENABLED(CONFIG_IDEAPAD_LAPTOP)
> > > + {
> > > + .hids = LWMI_PSY_DECONFLICT_STRS("VPC2004"),
> > > + .methods = LWMI_PSY_DECONFLICT_STRS("GBMD", "SBMC"),
> > > + },
> > > +#endif /* IS_ENABLED(CONFIG_IDEAPAD_LAPTOP) */
> > > +#if IS_ENABLED(CONFIG_THINKPAD_ACPI)
> > > + {
> > > + .hids = LWMI_PSY_DECONFLICT_STRS("LEN0268", "LEN0068", "IBM0068"),
> > > + .methods = LWMI_PSY_DECONFLICT_STRS("BDSG", "BICG"),
> > > + },
> > > +#endif /* IS_ENABLED(CONFIG_THINKPAD_ACPI) */
> > > +};
> > > +
> > > /**
> > > * lwmi_om_psy_ext_init() - Hooks power supply extension to device battery
> > > * @priv: Pointer to the lwmi_om_priv drvdata.
> > > *
> > > - * Checks if the ideapad_laptop driver is going to manage charge attributes first,
> > > - * then if not, hooks the battery to our WMI methods if they are supported.
> > > + * Checks if the ideapad_laptop or thinkpad_acpi driver is going to manage
> > > + * charge attributes first, then if not, hooks the battery to our WMI methods
> > > + * if they are supported.
> > > */
> > > static void lwmi_om_psy_ext_init(struct lwmi_om_priv *priv)
> > > {
> > > - static const char * const ideapad_hid = "VPC2004";
> > > - acpi_handle handle = NULL;
> > > unsigned int props = 0;
> > > int ret;
> > >
> > > @@ -928,14 +952,30 @@ static void lwmi_om_psy_ext_init(struct lwmi_om_priv *priv)
> > > if (force_load_psy_ext)
> > > goto load_psy_ext;
> > >
> > > - /* Deconflict ideapad_laptop driver */
> > > - ret = acpi_get_devices(ideapad_hid, lwmi_acpi_match, &handle, NULL);
> > > - if (ret)
> > > - return;
> > > + for (unsigned int i = 0; i < ARRAY_SIZE(lwmi_psy_deconflict_list); i++) {
> > > + const struct lwmi_psy_deconflict *entry = &lwmi_psy_deconflict_list[i];
> > >
> > > - if (handle && acpi_has_method(handle, "GBMD") && acpi_has_method(handle, "SBMC")) {
> > > - dev_dbg(&priv->wdev->dev, "ideapad_laptop driver manages battery for device\n");
> > > - return;
> > > + for (unsigned int j = 0; entry->hids[j]; j++) {
> > > + const char *hid = entry->hids[j];
> > > + acpi_handle handle = NULL;
> > > + bool has_methods = true;
> > > +
> > > + ret = acpi_get_devices(hid, lwmi_acpi_match, &handle, NULL);
> > > + if (ret || !handle)
> > > + continue;
> > > +
> > > + for (unsigned int k = 0; entry->methods[k]; k++) {
> > > + has_methods = acpi_has_method(handle, entry->methods[k]);
> > > + if (!has_methods)
> > > + break;
> >
> > So this goes to dev_dbg + return, why complicate the code flow with extra
> > break + check that is never true?
> >
> > If indentation is the worry here, the innermost bodies don't seem to
> > require much from the outer ones so they could be moved to a function
> > (j body + k loop).
> >
>
> When missing a method then try the next HID. In v4, I moved the lookup
> and method checks into a helper.
>
> Thanks,
> Aaron
>
> > > + }
> > > + if (!has_methods)
> > > + continue;
> > > +
> > > + dev_dbg(&priv->wdev->dev, "ACPI device %s manages battery for device\n",
> > > + hid);
> > > + return;
> > > + }
> > > }
> > >
> > > load_psy_ext:
> > >
> >
> > --
> > i.
> >