]> git.ipfire.org Git - thirdparty/linux.git/commitdiff
hwmon: add driver for ARCTIC Fan Controller
authorAureo Serrano de Souza <aureo.serrano@arctic.de>
Fri, 8 May 2026 06:44:00 +0000 (14:44 +0800)
committerGuenter Roeck <linux@roeck-us.net>
Tue, 9 Jun 2026 15:22:59 +0000 (08:22 -0700)
Add hwmon driver for the ARCTIC Fan Controller, a USB HID device
(VID 0x3904, PID 0xF001) with 10 fan channels. Exposes fan speed in
RPM (read-only) and PWM duty cycle (0-255, read/write) via sysfs.

The device pushes IN reports at ~1 Hz containing RPM readings. PWM is
set via OUT reports; the device applies the new duty cycle and sends
back a 2-byte ACK (Report ID 0x02). The driver waits up to 1 s for
the ACK using a completion. Measured device latency: max ~563 ms over
500 iterations. PWM control is manual-only: the device never changes
duty cycle autonomously.

raw_event() may run in hardirq context, so fan_rpm[] is protected by
a spinlock with irq-save. pwm_duty[] is also protected by this spinlock
because reset_resume() clears it outside the hwmon core lock. The OUT
report buffer is built and write_pending is armed under the same lock so
that no reset_resume() can race with the pwm_duty[] snapshot. priv->buf
is exclusively accessed by write(), which the hwmon core serializes.

Signed-off-by: Aureo Serrano de Souza <aureo.serrano@arctic.de>
Link: https://lore.kernel.org/r/20260508064405.38676-1-aureo.serrano@arctic.de
Signed-off-by: Guenter Roeck <linux@roeck-us.net>
Documentation/hwmon/arctic_fan_controller.rst [new file with mode: 0644]
Documentation/hwmon/index.rst
MAINTAINERS
drivers/hwmon/Kconfig
drivers/hwmon/Makefile
drivers/hwmon/arctic_fan_controller.c [new file with mode: 0644]

diff --git a/Documentation/hwmon/arctic_fan_controller.rst b/Documentation/hwmon/arctic_fan_controller.rst
new file mode 100644 (file)
index 0000000..b5be88a
--- /dev/null
@@ -0,0 +1,56 @@
+.. SPDX-License-Identifier: GPL-2.0-or-later
+
+Kernel driver arctic_fan_controller
+=====================================
+
+Supported devices:
+
+* ARCTIC Fan Controller (USB HID, VID 0x3904, PID 0xF001)
+
+Author: Aureo Serrano de Souza <aureo.serrano@arctic.de>
+
+Description
+-----------
+
+This driver provides hwmon support for the ARCTIC Fan Controller, a USB
+Custom HID device with 10 fan channels. The device sends IN reports about
+once per second containing current RPM values (bytes 11-30, 10 x uint16 LE).
+Fan speed control is manual-only: the device does not change PWM
+autonomously; it only applies a new duty cycle when it receives an OUT
+report from the host.
+
+After the device applies an OUT report, it sends back a 2-byte ACK IN
+report (Report ID 0x02, byte 1 = 0x00 on success) confirming the command
+was applied.
+
+Usage notes
+-----------
+
+Since it is a USB device, hotplug is supported. The device is autodetected.
+
+The device does not support GET_REPORT, so the driver cannot read back the
+current hardware PWM state at probe time. The cached PWM values (readable
+via pwm[1-10]) start at 0 and reflect only values that have been
+successfully written. Because each OUT report carries all 10 channel values,
+writing a single channel also sends the cached values for all other channels.
+Users should set all channels to the desired values before relying on the
+cached state.
+
+On system suspend, the device may lose power and reset its PWM channels to
+hardware defaults. The driver clears its cached duty values on resume so
+that reads reflect the unknown hardware state rather than stale pre-suspend
+values. Userspace is responsible for re-applying the desired duty cycles
+after resume.
+
+Sysfs entries
+-------------
+
+================ ==============================================================
+fan[1-10]_input  Fan speed in RPM (read-only). Updated from IN reports at ~1 Hz.
+pwm[1-10]        PWM duty cycle (0-255). Write: sends an OUT report setting the
+                 duty cycle (scaled from 0-255 to 0-100% for the device);
+                 the cached value is updated only after the device ACKs the
+                 command with a success status. Read: returns the last
+                 successfully written value; initialized to 0 at driver load
+                 and after resume (hardware state unknown).
+================ ==============================================================
index f9cbdcb75128c713f9fb421e65b2b45c1d5ca57f..78c783f11e2503ac1c1f4ba6a2a5896d9be85d84 100644 (file)
@@ -43,6 +43,7 @@ Hardware Monitoring Kernel Drivers
    amc6821
    aps-379
    aquacomputer_d5next
+   arctic_fan_controller
    asb100
    asc7621
    aspeed-g6-pwm-tach
index 743e451760c9ccc3f23ceae1e310aba70c7f2c8f..5d4453414e772cb08f827375c63a6eca6e494814 100644 (file)
@@ -2070,6 +2070,13 @@ S:       Maintained
 F:     drivers/net/arcnet/
 F:     include/uapi/linux/if_arcnet.h
 
+ARCTIC FAN CONTROLLER DRIVER
+M:     Aureo Serrano de Souza <aureo.serrano@arctic.de>
+L:     linux-hwmon@vger.kernel.org
+S:     Maintained
+F:     Documentation/hwmon/arctic_fan_controller.rst
+F:     drivers/hwmon/arctic_fan_controller.c
+
 ARM AND ARM64 SoC SUB-ARCHITECTURES (COMMON PARTS)
 M:     Arnd Bergmann <arnd@arndb.de>
 M:     Krzysztof Kozlowski <krzk@kernel.org>
index 14e4cea48acc473b9801ee3d030e5570d74212e8..997bba56fedfd56a8d7a7976370e7217546bd90c 100644 (file)
@@ -388,6 +388,18 @@ config SENSORS_APPLESMC
          Say Y here if you have an applicable laptop and want to experience
          the awesome power of applesmc.
 
+config SENSORS_ARCTIC_FAN_CONTROLLER
+       tristate "ARCTIC Fan Controller"
+       depends on USB_HID
+       help
+         If you say yes here you get support for the ARCTIC Fan Controller,
+         a USB HID device (VID 0x3904, PID 0xF001) with 10 fan channels.
+         The driver exposes fan speed (RPM) and PWM control via the hwmon
+         sysfs interface.
+
+         This driver can also be built as a module. If so, the module
+         will be called arctic_fan_controller.
+
 config SENSORS_ARM_SCMI
        tristate "ARM SCMI Sensors"
        depends on ARM_SCMI_PROTOCOL
index 982ee2c6f9deb6c48894d1cc642bfd67cfe89c95..efbd1cb81816f12dc4db5fc3c115a28426b29ace 100644 (file)
@@ -49,6 +49,7 @@ obj-$(CONFIG_SENSORS_ADT7475) += adt7475.o
 obj-$(CONFIG_SENSORS_AHT10)    += aht10.o
 obj-$(CONFIG_SENSORS_APPLESMC) += applesmc.o
 obj-$(CONFIG_SENSORS_AQUACOMPUTER_D5NEXT) += aquacomputer_d5next.o
+obj-$(CONFIG_SENSORS_ARCTIC_FAN_CONTROLLER)    += arctic_fan_controller.o
 obj-$(CONFIG_SENSORS_ARM_SCMI) += scmi-hwmon.o
 obj-$(CONFIG_SENSORS_ARM_SCPI) += scpi-hwmon.o
 obj-$(CONFIG_SENSORS_AS370)    += as370-hwmon.o
diff --git a/drivers/hwmon/arctic_fan_controller.c b/drivers/hwmon/arctic_fan_controller.c
new file mode 100644 (file)
index 0000000..dbe84cd
--- /dev/null
@@ -0,0 +1,374 @@
+// SPDX-License-Identifier: GPL-2.0-or-later
+/*
+ * Linux hwmon driver for ARCTIC Fan Controller
+ *
+ * USB Custom HID device with 10 fan channels.
+ * Exposes fan RPM (input) and PWM (0-255) via hwmon. Device pushes IN reports
+ * at ~1 Hz; no GET_REPORT. OUT reports set PWM duty (bytes 1-10, 0-100%).
+ * PWM is manual-only: the device does not change duty autonomously, only
+ * when it receives an OUT report from the host.
+ */
+
+#include <linux/completion.h>
+#include <linux/dma-mapping.h>
+#include <linux/err.h>
+#include <linux/hid.h>
+#include <linux/hwmon.h>
+#include <linux/jiffies.h>
+#include <linux/minmax.h>
+#include <linux/module.h>
+#include <linux/spinlock.h>
+#include <linux/string.h>
+#include <linux/unaligned.h>
+
+#define ARCTIC_VID                     0x3904
+#define ARCTIC_PID                     0xF001
+#define ARCTIC_NUM_FANS                        10
+#define ARCTIC_OUTPUT_REPORT_ID                0x01
+#define ARCTIC_REPORT_LEN              32
+#define ARCTIC_RPM_OFFSET              11      /* bytes 11-30: 10 x uint16 LE */
+/* ACK report: device sends Report ID 0x02, 2 bytes (ID + status) after applying OUT report */
+#define ARCTIC_ACK_REPORT_ID           0x02
+#define ARCTIC_ACK_REPORT_LEN          2
+/*
+ * Time to wait for ACK report after send.
+ * Measured over 500 iterations: max ~563 ms. Keep 1 s as margin.
+ */
+#define ARCTIC_ACK_TIMEOUT_MS          1000
+
+struct arctic_fan_data {
+       struct hid_device *hdev;
+       struct device *hwmon_dev;       /* stored for explicit unregister in remove() */
+       spinlock_t in_report_lock;      /* protects fan_rpm, ack_status, write_pending, pwm_duty */
+       struct completion in_report_received; /* ACK (ID 0x02) received in raw_event */
+       int ack_status;                 /* 0 = OK, negative errno on device error */
+       bool write_pending;             /* true while an OUT report ACK is in flight */
+       u32 fan_rpm[ARCTIC_NUM_FANS];
+       u8 pwm_duty[ARCTIC_NUM_FANS];   /* 0-255 matching sysfs range; converted to 0-100 on send */
+       /*
+        * OUT report buffer passed to hid_hw_output_report(). Embedded in the
+        * devm_kzalloc'd struct so it is heap-allocated and passes
+        * usb_hcd_map_urb_for_dma(). Exclusively accessed by write(), which
+        * the hwmon core serializes.
+        */
+       __dma_from_device_group_begin();
+       u8 buf[ARCTIC_REPORT_LEN];
+       __dma_from_device_group_end();
+};
+
+/*
+ * Parse RPM values from the periodic status report (10 x uint16 LE at rpm_off).
+ * pwm_duty is not updated from the report: the device is manual-only, so the
+ * host cache is the authoritative source for PWM.
+ * Called from raw_event which may run in IRQ context; must not sleep.
+ */
+static void arctic_fan_parse_report(struct arctic_fan_data *priv, u8 *buf,
+                                   int len, int rpm_off)
+{
+       unsigned long flags;
+       int i;
+
+       if (len < rpm_off + 20)
+               return;
+
+       spin_lock_irqsave(&priv->in_report_lock, flags);
+       for (i = 0; i < ARCTIC_NUM_FANS; i++)
+               priv->fan_rpm[i] = get_unaligned_le16(&buf[rpm_off + i * 2]);
+       spin_unlock_irqrestore(&priv->in_report_lock, flags);
+}
+
+/*
+ * raw_event: IN reports.
+ *
+ * Status report: Report ID 0x01, 32 bytes:
+ *   byte 0 = report ID, bytes 1-10 = PWM 0-100%, bytes 11-30 = 10 x RPM uint16 LE.
+ *   Device pushes these at ~1 Hz; no GET_REPORT.
+ *
+ * ACK report: Report ID 0x02, 2 bytes:
+ *   byte 0 = 0x02, byte 1 = status (0x00 = OK, 0x01 = ERROR).
+ *   Sent once after accepting and applying an OUT report (ID 0x01).
+ */
+static int arctic_fan_raw_event(struct hid_device *hdev,
+                               struct hid_report *report, u8 *data, int size)
+{
+       struct arctic_fan_data *priv = hid_get_drvdata(hdev);
+       unsigned long flags;
+
+       hid_dbg(hdev, "arctic_fan: raw_event id=%u size=%d\n", report->id, size);
+
+       if (report->id == ARCTIC_ACK_REPORT_ID && size == ARCTIC_ACK_REPORT_LEN) {
+               spin_lock_irqsave(&priv->in_report_lock, flags);
+               /*
+                * Only deliver if a write is in flight. This prevents a
+                * late-arriving ACK from a timed-out write from erroneously
+                * satisfying a subsequent write's completion wait.
+                */
+               if (priv->write_pending) {
+                       priv->ack_status = data[1] == 0x00 ? 0 : -EIO;
+                       complete(&priv->in_report_received);
+               }
+               spin_unlock_irqrestore(&priv->in_report_lock, flags);
+               return 0;
+       }
+
+       if (report->id != ARCTIC_OUTPUT_REPORT_ID || size != ARCTIC_REPORT_LEN) {
+               hid_dbg(hdev, "arctic_fan: raw_event id=%u size=%d ignored\n",
+                       report->id, size);
+               return 0;
+       }
+
+       arctic_fan_parse_report(priv, data, size, ARCTIC_RPM_OFFSET);
+       return 0;
+}
+
+static umode_t arctic_fan_is_visible(const void *data,
+                                    enum hwmon_sensor_types type,
+                                    u32 attr, int channel)
+{
+       if (type == hwmon_fan && attr == hwmon_fan_input)
+               return 0444;
+       if (type == hwmon_pwm && attr == hwmon_pwm_input)
+               return 0644;
+       return 0;
+}
+
+static int arctic_fan_read(struct device *dev, enum hwmon_sensor_types type,
+                          u32 attr, int channel, long *val)
+{
+       struct arctic_fan_data *priv = dev_get_drvdata(dev);
+       unsigned long flags;
+
+       if (type == hwmon_fan && attr == hwmon_fan_input) {
+               spin_lock_irqsave(&priv->in_report_lock, flags);
+               *val = priv->fan_rpm[channel];
+               spin_unlock_irqrestore(&priv->in_report_lock, flags);
+               return 0;
+       }
+       if (type == hwmon_pwm && attr == hwmon_pwm_input) {
+               spin_lock_irqsave(&priv->in_report_lock, flags);
+               *val = priv->pwm_duty[channel];
+               spin_unlock_irqrestore(&priv->in_report_lock, flags);
+               return 0;
+       }
+       return -EINVAL;
+}
+
+static int arctic_fan_write(struct device *dev, enum hwmon_sensor_types type,
+                           u32 attr, int channel, long val)
+{
+       struct arctic_fan_data *priv = dev_get_drvdata(dev);
+       u8 new_duty = (u8)clamp_val(val, 0, 255);
+       unsigned long flags;
+       unsigned long t;
+       int i, ret;
+
+       /*
+        * Build the buffer and arm write_pending under in_report_lock so that
+        * reset_resume() cannot clear pwm_duty[] between the pwm_duty[] read
+        * and the buffer write, and raw_event() cannot deliver a stale ACK
+        * from a previous write into this write's completion.
+        *
+        * priv->buf is heap-allocated (embedded in the devm_kzalloc'd struct),
+        * satisfying usb_hcd_map_urb_for_dma(). Exclusively accessed by
+        * write() which the hwmon core serializes.
+        *
+        * pwm_duty[channel] is committed only after a positive device ACK so a
+        * failed or timed-out write does not corrupt the cached state.
+        *
+        * Residual theoretical race: if write A times out (write_pending
+        * cleared), write B sets write_pending = true, and a late ACK from
+        * write A—delayed beyond ARCTIC_ACK_TIMEOUT_MS—arrives during write
+        * B's pending window, it would falsely satisfy write B's completion.
+        * This cannot be prevented in driver code without protocol support
+        * (for example, a correlation ID echoed in the device ACK report).
+        * In testing, observed ACK latency stayed below the 1 s timeout
+        * (maximum ~563 ms over 500 iterations).
+        *
+        * The wait is non-interruptible so that a signal cannot cause write()
+        * to return early while the OUT report is already in flight; an
+        * interruptible early return would create the same late-ACK window
+        * without even the timeout guard.
+        * Serialized by the hwmon core: only one arctic_fan_write() at a time.
+        * Use irqsave to match the IRQ context in which raw_event may run.
+        */
+       spin_lock_irqsave(&priv->in_report_lock, flags);
+       priv->buf[0] = ARCTIC_OUTPUT_REPORT_ID;
+       for (i = 0; i < ARCTIC_NUM_FANS; i++) {
+               u8 d = i == channel ? new_duty : priv->pwm_duty[i];
+
+               priv->buf[1 + i] = DIV_ROUND_CLOSEST((unsigned int)d * 100, 255);
+       }
+       priv->ack_status = -ETIMEDOUT;
+       priv->write_pending = true;
+       reinit_completion(&priv->in_report_received);
+       spin_unlock_irqrestore(&priv->in_report_lock, flags);
+
+       ret = hid_hw_output_report(priv->hdev, priv->buf, ARCTIC_REPORT_LEN);
+       if (ret < 0) {
+               spin_lock_irqsave(&priv->in_report_lock, flags);
+               priv->write_pending = false;
+               spin_unlock_irqrestore(&priv->in_report_lock, flags);
+               return ret;
+       }
+
+       t = wait_for_completion_timeout(&priv->in_report_received,
+                                       msecs_to_jiffies(ARCTIC_ACK_TIMEOUT_MS));
+       spin_lock_irqsave(&priv->in_report_lock, flags);
+       priv->write_pending = false;
+       /* Commit inside the lock so reset_resume() cannot race with this write */
+       if (t && priv->ack_status == 0)
+               priv->pwm_duty[channel] = new_duty;
+       spin_unlock_irqrestore(&priv->in_report_lock, flags);
+
+       if (!t)
+               return -ETIMEDOUT;
+       return priv->ack_status; /* 0=OK, -EIO=device error */
+}
+
+static const struct hwmon_ops arctic_fan_ops = {
+       .is_visible = arctic_fan_is_visible,
+       .read = arctic_fan_read,
+       .write = arctic_fan_write,
+};
+
+static const struct hwmon_channel_info *arctic_fan_info[] = {
+       HWMON_CHANNEL_INFO(fan,
+                          HWMON_F_INPUT, HWMON_F_INPUT, HWMON_F_INPUT,
+                          HWMON_F_INPUT, HWMON_F_INPUT, HWMON_F_INPUT,
+                          HWMON_F_INPUT, HWMON_F_INPUT, HWMON_F_INPUT,
+                          HWMON_F_INPUT),
+       HWMON_CHANNEL_INFO(pwm,
+                          HWMON_PWM_INPUT, HWMON_PWM_INPUT, HWMON_PWM_INPUT,
+                          HWMON_PWM_INPUT, HWMON_PWM_INPUT, HWMON_PWM_INPUT,
+                          HWMON_PWM_INPUT, HWMON_PWM_INPUT, HWMON_PWM_INPUT,
+                          HWMON_PWM_INPUT),
+       NULL
+};
+
+static const struct hwmon_chip_info arctic_fan_chip_info = {
+       .ops = &arctic_fan_ops,
+       .info = arctic_fan_info,
+};
+
+static int arctic_fan_reset_resume(struct hid_device *hdev)
+{
+       struct arctic_fan_data *priv = hid_get_drvdata(hdev);
+       unsigned long flags;
+
+       /*
+        * The device resets its PWM channels to hardware defaults on power
+        * loss during suspend. Clear the cached duty values so they reflect
+        * the unknown hardware state, consistent with probe-time behaviour
+        * (the device has no GET_REPORT support). Hold in_report_lock so
+        * this does not race with a concurrent pwm read or write callback.
+        */
+       spin_lock_irqsave(&priv->in_report_lock, flags);
+       memset(priv->pwm_duty, 0, sizeof(priv->pwm_duty));
+       spin_unlock_irqrestore(&priv->in_report_lock, flags);
+       return 0;
+}
+
+static int arctic_fan_probe(struct hid_device *hdev,
+                           const struct hid_device_id *id)
+{
+       struct arctic_fan_data *priv;
+       int ret;
+
+       if (!hid_is_usb(hdev))
+               return -ENODEV;
+
+       ret = hid_parse(hdev);
+       if (ret)
+               return ret;
+
+       priv = devm_kzalloc(&hdev->dev, sizeof(*priv), GFP_KERNEL);
+       if (!priv)
+               return -ENOMEM;
+
+       priv->hdev = hdev;
+       spin_lock_init(&priv->in_report_lock);
+       init_completion(&priv->in_report_received);
+       hid_set_drvdata(hdev, priv);
+
+       ret = hid_hw_start(hdev, HID_CONNECT_DRIVER);
+       if (ret)
+               return ret;
+
+       ret = hid_hw_open(hdev);
+       if (ret)
+               goto out_stop;
+
+       /*
+        * Start IO before registering with hwmon. If IO were started after
+        * hwmon registration, a sysfs write arriving in that narrow window
+        * would send an OUT report but the ACK could not be delivered (the HID
+        * core discards events until io_started), causing a spurious timeout.
+        */
+       hid_device_io_start(hdev);
+
+       /*
+        * Use the non-devm variant and store the pointer so remove() can
+        * call hwmon_device_unregister() before tearing down the HID
+        * transport. devm_hwmon_device_register_with_info() would defer
+        * unregistration until after remove() returns, leaving a window
+        * where a concurrent sysfs write could call hid_hw_output_report()
+        * on an already-stopped device (use-after-free).
+        */
+       priv->hwmon_dev = hwmon_device_register_with_info(&hdev->dev, "arctic_fan",
+                                                         priv, &arctic_fan_chip_info,
+                                                         NULL);
+       if (IS_ERR(priv->hwmon_dev)) {
+               ret = PTR_ERR(priv->hwmon_dev);
+               goto out_close;
+       }
+
+       return 0;
+
+out_close:
+       hid_device_io_stop(hdev);
+       hid_hw_close(hdev);
+out_stop:
+       hid_hw_stop(hdev);
+       return ret;
+}
+
+static void arctic_fan_remove(struct hid_device *hdev)
+{
+       struct arctic_fan_data *priv = hid_get_drvdata(hdev);
+
+       /*
+        * Unregister hwmon before stopping the HID transport. This removes
+        * the sysfs files and waits for any in-progress write() callback to
+        * return, so no hwmon op can call hid_hw_output_report() after
+        * hid_hw_stop() frees the underlying USB resources.
+        * Matches the pattern used by nzxt-smart2 and aquacomputer_d5next.
+        *
+        * The HID core clears hdev->io_started before invoking ->remove(),
+        * so hid_device_io_stop() is not called here; doing so would emit
+        * a spurious "io already stopped" warning.
+        */
+       hwmon_device_unregister(priv->hwmon_dev);
+       hid_hw_close(hdev);
+       hid_hw_stop(hdev);
+}
+
+static const struct hid_device_id arctic_fan_id_table[] = {
+       { HID_USB_DEVICE(ARCTIC_VID, ARCTIC_PID) },
+       { }
+};
+MODULE_DEVICE_TABLE(hid, arctic_fan_id_table);
+
+static struct hid_driver arctic_fan_driver = {
+       .name = "arctic_fan",
+       .id_table = arctic_fan_id_table,
+       .probe = arctic_fan_probe,
+       .remove = arctic_fan_remove,
+       .raw_event = arctic_fan_raw_event,
+       .reset_resume = arctic_fan_reset_resume,
+};
+
+module_hid_driver(arctic_fan_driver);
+
+MODULE_AUTHOR("Aureo Serrano de Souza <aureo.serrano@arctic.de>");
+MODULE_DESCRIPTION("HID hwmon driver for ARCTIC Fan Controller");
+MODULE_LICENSE("GPL");