On 9/3/25 01:33, Cryolitia PukNgae wrote:
On 02/09/2025 23.38, Guenter Roeck wrote:
On Wed, Aug 20, 2025 at 05:50:38PM +0800, Cryolitia PukNgae wrote:
From: Cryolitia PukNgae <cryolitia@xxxxxxxxxxxxx>
Sensors driver for GPD Handhelds that expose fan reading and control via
hwmon sysfs.
Shenzhen GPD Technology Co., Ltd. manufactures a series of handheld
devices. This driver implements these functions through x86 port-mapped
IO.
Signed-off-by: Cryolitia PukNgae <cryolitia@xxxxxxxxxxxxx>
---
+static int gpd_win_mini_set_pwm_enable(enum FAN_PWM_ENABLE pwm_enable)
+{
+ const struct gpd_fan_drvdata *drvdata;
+
+ switch (pwm_enable) {
+ case DISABLE:
+ return gpd_generic_write_pwm(255);
+ case MANUAL:
+ return gpd_generic_write_pwm(gpd_driver_priv.pwm_value);
This means setting the enable status to MANUAL will set the pwm
speed to the maximum unless a different speed is written first.
That includes win_max_2 even if the value was read before.
Is this really what is intended ?
Thank you for raising this important question. After reconsidering the
design, I believe this approach is indeed intentional and justified for
this specific hardware platform.
GPD devices are extremely compact yet high-performance systems, and their
gaming handheld form factor means they're often subjected to rapidly
fluctuating workloads. The small thermal mass combined with high heat
generation creates a scenario where even minor CPU load changes can cause
temperature spikes of 10+ Kelvin within seconds—as I observed during
testing when briefly stopping the fan.
This behavior is designed with a specific safety philosophy: when userspace
announces it's taking manual control, we assume it has already prepared
proper thermal monitoring and fan curves. The brief moment of max fan
speed acts as a safety buffer until userspace writes its first intended
value—which should happen almost immediately if the userspace controller
is properly implemented. This prevents any gap in thermal protection or
frame drops caused by throttling if userspace fails to write a value
promptly after claiming control.
Please make sure to document this behavior.
Thanks,
Guenter
