A driver's probe calls hid_device_io_start() to enable input delivery,
then fails at a later initialization step and unwinds via hid_hw_stop().
The unwind frees struct hidraw via hidraw_disconnect() while in-flight
HID reports may still be running on another CPU, dereferencing the
freed object through hidraw_report_event(). syzbot reports the
resulting use-after-free for the corsair-psu HID driver.

Edward Adam Davis posted a per-driver fix for corsair-psu that adds
an explicit hid_device_io_stop() before hid_hw_stop() in the probe
error path ("hwmon: prevent packets from going to driver for probe",
2026-04-28). Auditing the tree shows 15 drivers call
hid_device_io_start(); 7 also call hid_device_io_stop() and 8 do not:

  drivers calling hid_device_io_start() without a matching
  hid_device_io_stop() before hid_hw_stop():
    drivers/hwmon/corsair-psu.c       (fix posted by Edward)
    drivers/hwmon/corsair-cpro.c
    drivers/hwmon/nzxt-kraken3.c
    drivers/hwmon/nzxt-smart2.c
    drivers/hwmon/gigabyte_waterforce.c
    drivers/hid/hid-logitech-dj.c
    drivers/hid/hid-nintendo.c
    drivers/hid/hid-mcp2221.c

Roughly half of all callers of the API are exposed. Centralize the
quiesce in hid_hw_stop() so callers do not have to remember the
matching stop: if a driver has left hdev->io_started true on entry,
call hid_device_io_stop() before hid_disconnect().

For the 7 drivers that already call hid_device_io_stop() correctly,
hdev->io_started is false on entry, the guard short-circuits, and
behavior is unchanged.

No Fixes: tag because the affected drivers gained their
hid_device_io_start() calls independently over years; the bug is a
class-wide API misuse rather than a regression from one commit.

Reported-by: syzbot+9eebf5f6544c5e873858@syzkaller.appspotmail.com
Closes: https://syzkaller.appspot.com/bug?extid=9eebf5f6544c5e873858
Signed-off-by: Philipp Weber <kernel@phwe.de>
---
 drivers/hid/hid-core.c | 7 +++++++
 1 file changed, 7 insertions(+)

diff --git a/drivers/hid/hid-core.c b/drivers/hid/hid-core.c
index 41a79e43c82b..6b024118d983 100644
--- a/drivers/hid/hid-core.c
+++ b/drivers/hid/hid-core.c
@@ -2440,9 +2440,16 @@ EXPORT_SYMBOL_GPL(hid_hw_start);
  *
  * This is usually called from remove function or from probe when something
  * failed and hid_hw_start was called already.
+ *
+ * If the caller enabled HID input via hid_device_io_start() and is unwinding
+ * without an explicit hid_device_io_stop(), quiesce input first so that
+ * in-flight reports cannot reach handlers (e.g. hidraw_report_event) whose
+ * backing objects hid_disconnect() is about to free.
  */
 void hid_hw_stop(struct hid_device *hdev)
 {
+	if (hdev->io_started)
+		hid_device_io_stop(hdev);
 	hid_disconnect(hdev);
 	hdev->ll_driver->stop(hdev);
 }

base-commit: ab5fce87a778cb780a05984a2ca448f2b41aafbf
-- 
2.53.0