Skip to contentSkip to conversation

ACPI: dock: Retry probing battery docks after hotplug

PATCH Applied Vibhor Goel 2 participants 2 messages Oct 3 – Oct 12 ACPI

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 ACPI: dock: Retry probing battery docks after hotplug

On a ThinkPad T480 booted with the removable battery absent, reinserting that battery can leave BAT1 missing from /sys/class/power_supply even after firmware sends an ACPI add notification.

The battery dock platform device already exists, but its driver did not bind while the battery was absent. During hotplug, acpi_bus_scan() updates the ACPI device and attempts to attach it on the ACPI bus. acpi_create_platform_device() does not create another platform device because the existing physical node is still associated with the ACPI device. Consequently, the existing platform device remains unbound.

Retry probing the battery dock platform device after hotplug_dock_devices() has refreshed the ACPI state. Restrict this to battery docks. If the platform device is already bound, device_initial_probe() leaves that binding in place.

---
Tested on a dual-battery ThinkPad T480 with an aftermarket external pack.

On Fedora 7.2.8-200.fc44.x86_64, tracing showed the firmware add notification, a successful ACPI rescan, and no attachment attempt for the existing dock.0 platform device. A subsequent manual bind registered BAT1.

On patched upstream v7.2.8, booting with BAT1 absent and then triggering a firmware add notification automatically called device_initial_probe(), bound acpi-battery and registered BAT1. UPower detected the battery too. Removal and reinsertion after binding also worked, without a second driver probe. Both captures completed without trace-buffer loss.

Physical reinsertion initially produced no add notification on this pack; toggling charger mains while leaving the battery untouched triggered one. This patch addresses Linux handling after that notification. The initial notification behavior is outside its scope.

The baseline and patched kernels differ in configuration and Fedora patches. The traces identify the missing attachment operation and show the new call resolving it; this was not a same-source unpatched comparison.

Built the kernel and all 221 configured modules, booted with Secure Boot enabled, and tested on the physical laptop. The patch was developed with AI assistance and reviewed and tested by me.

 drivers/acpi/dock.c | 8 ++++++++
 1 file changed, 8 insertions(+)
drivers/acpi/dock.c +8 -0
index d7f3a8912..748df19d9 100644
@@ -445,6 +445,14 @@ int dock_notify(struct acpi_device *adev, u32 event)
 				break;
 			}
 			hotplug_dock_devices(ds, event);
+			/*
+			 * The dock platform device already exists when an empty battery
+			 * bay becomes occupied, but its driver may not have bound while
+			 * the battery was absent.  Retry after the ACPI scan updates its
+			 * status, since the scan only attaches the ACPI device.
+			 */
+			if (ds->flags & DOCK_IS_BAT)
+				device_initial_probe(&ds->dock_device->dev);
 			complete_dock(ds);
 			dock_event(ds, event, DOCK_EVENT);
 			acpi_evaluate_lck(ds->handle, 1);
Signature · 1 line
2.55.0
maintainer Applied
#1 lore raw
Quoted 57 lines · Vibhor Goel

On a ThinkPad T480 booted with the removable battery absent, reinserting that battery can leave BAT1 missing from /sys/class/power_supply even after firmware sends an ACPI add notification.

The battery dock platform device already exists, but its driver did not bind while the battery was absent. During hotplug, acpi_bus_scan() updates the ACPI device and attempts to attach it on the ACPI bus. acpi_create_platform_device() does not create another platform device because the existing physical node is still associated with the ACPI device. Consequently, the existing platform device remains unbound.

Retry probing the battery dock platform device after hotplug_dock_devices() has refreshed the ACPI state. Restrict this to battery docks. If the platform device is already bound, device_initial_probe() leaves that binding in place.

---
Tested on a dual-battery ThinkPad T480 with an aftermarket external pack.

On Fedora 7.2.8-200.fc44.x86_64, tracing showed the firmware add notification, a successful ACPI rescan, and no attachment attempt for the existing dock.0 platform device. A subsequent manual bind registered BAT1.

On patched upstream v7.2.8, booting with BAT1 absent and then triggering a firmware add notification automatically called device_initial_probe(), bound acpi-battery and registered BAT1. UPower detected the battery too. Removal and reinsertion after binding also worked, without a second driver probe. Both captures completed without trace-buffer loss.

Physical reinsertion initially produced no add notification on this pack; toggling charger mains while leaving the battery untouched triggered one. This patch addresses Linux handling after that notification. The initial notification behavior is outside its scope.

The baseline and patched kernels differ in configuration and Fedora patches. The traces identify the missing attachment operation and show the new call resolving it; this was not a same-source unpatched comparison.

Built the kernel and all 221 configured modules, booted with Secure Boot enabled, and tested on the physical laptop. The patch was developed with AI assistance and reviewed and tested by me.

 drivers/acpi/dock.c | 8 ++++++++
 1 file changed, 8 insertions(+)
drivers/acpi/dock.c +2 -0
index d7f3a8912..748df19d9 100644
@@ -445,6 +445,14 @@ int dock_notify(struct acpi_device *adev, u32 event)
                                break;
                        }
                        hotplug_dock_devices(ds, event);
+                       /*
+                        * The dock platform device already exists when an empty battery

drivers/acpi/dock.c int dock_notify(struct acpi_device *adev, u32 event)

Quoted 10 lines
drivers/acpi/dock.c +6 -1
+                        * bay becomes occupied, but its driver may not have bound while
+                        * the battery was absent.  Retry after the ACPI scan updates its
+                        * status, since the scan only attaches the ACPI device.
+                        */
+                       if (ds->flags & DOCK_IS_BAT)
+                               device_initial_probe(&ds->dock_device->dev);
                        complete_dock(ds);
                        dock_event(ds, event, DOCK_EVENT);
                        acpi_evaluate_lck(ds->handle, 1);
--

Applied as 7.4 material, thanks!