Skip to contentSkip to conversation

HID: nintendo: Add preliminary Switch 2 controller driver

PATCH Superseded Vicki Pfau 5 participants 15 messages May 12 – Jun 22

Subsystems (3)

v5 of 16 Superseded Latest is v16 →

lore
Share & tools
lore

Vicki Pfau

author maintainer
#0 lore raw

v5 cover HID: nintendo: Add preliminary Switch 2 controller driver

This series adds preliminary support for Switch 2 controllers using the same split-driver model as previous versions. This is a minor iteration on v4, fixing some leaked input_dev objects in error paths as well as fixing a merge/rebase error when CONFIG_PM ifdefs got removed in 7.0.

Vicki Pfau (3):
HID: nintendo: Add preliminary Switch 2 controller driver
HID: nintendo: Add rumble support for Switch 2 controllers
HID: nintendo: Add unified report format support

 MAINTAINERS                                   |    1 +
 drivers/hid/Kconfig                           |   19 +-
 drivers/hid/hid-ids.h                         |    4 +
 drivers/hid/hid-nintendo.c                    | 1504 ++++++++++++++++-
 drivers/hid/hid-nintendo.h                    |   72 +
 drivers/input/joystick/Kconfig                |   11 +
 drivers/input/joystick/Makefile               |    1 +
 drivers/input/joystick/nintendo-switch2-usb.c |  353 ++++
 8 files changed, 1951 insertions(+), 14 deletions(-)
 create mode 100644 drivers/hid/hid-nintendo.h
 create mode 100644 drivers/input/joystick/nintendo-switch2-usb.c
Signature · 1 line
2.54.0
author maintainer
#1 lore raw

PATCH v5 1/3 HID: nintendo: Add preliminary Switch 2 controller driver

This adds a new driver for the Switch 2 controllers. The Switch 2 uses an unusual split-interface design such that input and rumble occur on the main HID interface, but all other communication occurs over a "configuration" interface. This is the case on both USB and Bluetooth, so this new driver uses a split-driver design with the HID interface being the "main" driver and the configuration interface is a secondary driver that looks up to the HID interface, sharing resources on a common struct.

Due to using a non-standard pairing interface as well as Bluetooth communications being extremely limited in the kernel, a custom interface between userspace and the kernel will need to be designed, along with bringup in BlueZ. That is beyond the scope of this initial patch, which only contains the generic HID and USB configuration interface drivers.

This initial work supports general input for the Joy-Con 2, Pro Controller 2, and GameCube NSO controllers. IMU, rumble and battery support is not yet present.

---
 MAINTAINERS                                   |    1 +
 drivers/hid/Kconfig                           |   11 +-
 drivers/hid/hid-ids.h                         |    4 +
 drivers/hid/hid-nintendo.c                    | 1191 ++++++++++++++++-
 drivers/hid/hid-nintendo.h                    |   72 +
 drivers/input/joystick/Kconfig                |   11 +
 drivers/input/joystick/Makefile               |    1 +
 drivers/input/joystick/nintendo-switch2-usb.c |  353 +++++
 8 files changed, 1634 insertions(+), 10 deletions(-)
 create mode 100644 drivers/hid/hid-nintendo.h
 create mode 100644 drivers/input/joystick/nintendo-switch2-usb.c
MAINTAINERS +1 -0
index d317e2ca65c1..9ebecc0297a9 100644
@@ -18882,6 +18882,7 @@ F:	drivers/scsi/nsp32*
 
 NINTENDO HID DRIVER
 M:	Daniel J. Ogorchock <[email protected]>
+M:	Vicki Pfau <[email protected]>
 L:	[email protected]
 S:	Maintained
 F:	drivers/hid/hid-nintendo*
drivers/hid/Kconfig +7 -4
index f9bcaeb66385..19c77c323ec9 100644
@@ -843,10 +843,13 @@ config HID_NINTENDO
 	depends on LEDS_CLASS
 	select POWER_SUPPLY
 	help
-	Adds support for the Nintendo Switch Joy-Cons, NSO, Pro Controller.
-	All controllers support bluetooth, and the Pro Controller also supports
-	its USB mode. This also includes support for the Nintendo Switch Online
-	Controllers which include the NES, Genesis, SNES, and N64 controllers.
+	Adds support for the Nintendo Switch Joy-Cons, NSO, Pro Controller, as
+	well as Nintendo Switch 2 Joy-Cons, Pro Controller, and NSO GameCube
+	controllers. All Switch controllers support bluetooth, and the Pro
+	Controller also supports its USB mode. This also includes support for
+	the Nintendo Switch Online Controllers which include the NES, Genesis,
+	SNES, and N64 controllers. Switch 2 controllers currently only support
+	USB mode.
 
 	To compile this driver as a module, choose M here: the
 	module will be called hid-nintendo.
drivers/hid/hid-ids.h +4 -0
index 20e7d2dbbf7c..0ac906fbbf1c 100644
@@ -1098,6 +1098,10 @@
 #define USB_DEVICE_ID_NINTENDO_SNESCON	0x2017
 #define USB_DEVICE_ID_NINTENDO_GENCON	0x201e
 #define USB_DEVICE_ID_NINTENDO_N64CON	0x2019
+#define USB_DEVICE_ID_NINTENDO_NS2_JOYCONR	0x2066
+#define USB_DEVICE_ID_NINTENDO_NS2_JOYCONL	0x2067
+#define USB_DEVICE_ID_NINTENDO_NS2_PROCON	0x2069
+#define USB_DEVICE_ID_NINTENDO_NS2_GCCON	0x2073
 
 #define USB_VENDOR_ID_NOVATEK		0x0603
 #define USB_DEVICE_ID_NOVATEK_PCT	0x0600
drivers/hid/hid-nintendo.c +1185 -6

1,270 lines not on this page · raw message

drivers/hid/hid-nintendo.h new file +72 -0

72 lines not on this page · raw message

drivers/input/joystick/Kconfig +11 -0
index 7755e5b454d2..868262c6ccd9 100644
@@ -422,4 +422,15 @@ config JOYSTICK_SEESAW
 	  To compile this driver as a module, choose M here: the module will be
 	  called adafruit-seesaw.
 
+config JOYSTICK_NINTENDO_SWITCH2_USB
+	tristate "Wired Nintendo Switch 2 controller support"
+	depends on HID_NINTENDO
+	depends on USB
+	help
+	  Say Y here if you want to enable support for wired Nintendo Switch 2
+	  controllers.
+
+	  To compile this driver as a module, choose M here: the
+	  module will be called nintendo-switch2-usb.
+
 endif
drivers/input/joystick/Makefile +1 -0
index 9976f596a920..8f92900ae885 100644
@@ -34,6 +34,7 @@ obj-$(CONFIG_JOYSTICK_SIDEWINDER)	+= sidewinder.o
 obj-$(CONFIG_JOYSTICK_SPACEBALL)	+= spaceball.o
 obj-$(CONFIG_JOYSTICK_SPACEORB)		+= spaceorb.o
 obj-$(CONFIG_JOYSTICK_STINGER)		+= stinger.o
+obj-$(CONFIG_JOYSTICK_NINTENDO_SWITCH2_USB)	+= nintendo-switch2-usb.o
 obj-$(CONFIG_JOYSTICK_TMDC)		+= tmdc.o
 obj-$(CONFIG_JOYSTICK_TURBOGRAFX)	+= turbografx.o
 obj-$(CONFIG_JOYSTICK_TWIDJOY)		+= twidjoy.o
drivers/input/joystick/nintendo-switch2-usb.c new file +353 -0

353 lines not on this page · raw message

Signature · 1 line
2.54.0
#2 lore raw
Thank you for your contribution! Sashiko AI review found 11 potential issue(s) to consider: - [High] Sleeping in atomic context and AB-BA deadlock in `switch2_usb_disconnect` - [High] Race condition…

Thank you for your contribution! Sashiko AI review found 11 potential issue(s) to consider:
- [High] Sleeping in atomic context and AB-BA deadlock in switch2_usb_disconnect
- [High] Race condition in IN URB tracking causes permanent starvation
- [High] Dangling pointers and use-after-free on probe failure
- [High] Memory corruption and double free due to incorrect devm input lifecycle management
- [High] NULL pointer dereference in USB probe error path
- [High] Division by zero crash from unvalidated trigger calibration data
- [High] BUG_ON kernel panic when ida_free is called with U32_MAX
- [High] Out-of-bounds read when parsing NS2_CMD_FW_INFO response
- [High] Race condition due to URB submission before initialization of synchronization primitives
- [Medium] Infinite USB retry storm on protocol errors
- [Medium] Unaligned memory access on byte array casts
--

commit 00cb7265ad5983144c6923702e2ee2c3c047d29e
Author: Vicki Pfau <[email protected]>
Subject: HID: nintendo: Add preliminary Switch 2 controller driver

This adds a new driver for the Switch 2 controllers. The Switch 2 uses an unusual split-interface design such that input and rumble occur on the main HID interface, but all other communication occurs over a "configuration" interface.

Quoted 4 lines
drivers/hid/hid-nintendo.c +0 -0
index 29008c2cc5304..2193afcdefabc 100644

[ ... ]

Quoted 2 lines
drivers/hid/hid-nintendo.c +2 -0
+static int switch2_init_input(struct switch2_controller *ns2)
+{

[ ... ]

Quoted 3 lines
drivers/hid/hid-nintendo.c +3 -0
+	input = devm_input_allocate_device(&hdev->dev);
+	if (!input)
+		return -ENOMEM;

[ ... ]

drivers/hid/hid-nintendo.c static int nintendo_hid_suspend(struct hid_device *hdev, pm_message_t message) · line 3238

Quoted 4 lines
drivers/hid/hid-nintendo.c +4 -0
+	ret = input_register_device(input);
+	if (ret < 0) {
+		hid_err(ns2->hdev, "Failed to register input; ret=%d\n", ret);
+		input_free_device(input);

Calling input_free_device or input_unregister_device on a devm-managed input device can cause double frees when devres cleanup fires. Can this lead to memory corruption?

Quoted 2 lines
drivers/hid/hid-nintendo.c +2 -0
+		return ret;
+	}

[ ... ]

Quoted 4 lines
drivers/hid/hid-nintendo.c +4 -0
+static void switch2_controller_put(struct switch2_controller *ns2)
+{
+	struct input_dev *input;
+	bool do_free;

[ ... ]

drivers/hid/hid-nintendo.c static int nintendo_hid_suspend(struct hid_device *hdev, pm_message_t message) · line 3309

Quoted 2 lines
drivers/hid/hid-nintendo.c +2 -0
+	if (input)
+		input_unregister_device(input);

Calling input_unregister_device on a devm-managed input device can cause issues. If the input device remains alive (tied to the HID device devres) and its child LEDs are not released, will reconnecting the USB cable allocate a new input device and corrupt the LED lists by re-registering the exact same ns2->leds structures?

Quoted 2 lines
drivers/hid/hid-nintendo.c +2 -0
+
+	if (do_free) {

[ ... ]

drivers/hid/hid-nintendo.c static int nintendo_hid_suspend(struct hid_device *hdev, pm_message_t message) · line 3502

Quoted 3 lines
drivers/hid/hid-nintendo.c +3 -0
+static void switch2_report_trigger(struct input_dev *input, uint8_t zero, int abs, uint8_t data)
+{
+	int value = (NS2_TRIGGER_RANGE + 1) * (data - zero) / (232 - zero);

The zero variable is loaded directly from the controller's factory calibration flash. If a device reports a zero point of exactly 232, will this trigger a division by zero exception and crash the kernel?

Quoted 3 lines
drivers/hid/hid-nintendo.c +3 -0
+
+	input_report_abs(input, abs, clamp(value, 0, NS2_TRIGGER_RANGE));
+}

[ ... ]

drivers/hid/hid-nintendo.c static int nintendo_hid_suspend(struct hid_device *hdev, pm_message_t message) · line 3597

Quoted 8 lines
drivers/hid/hid-nintendo.c +8 -0
+static int switch2_read_flash(struct switch2_controller *ns2, uint32_t address,
+	uint8_t size)
+{
+	uint8_t message[8] = { size, 0x7e };
+
+	if (!ns2->cfg)
+		return -ENOTCONN;
+	*(__le32 *)&message[4] = __cpu_to_le32(address);

The driver directly casts &message[4] to __le32. Since message is a uint8_t array, it is only guaranteed to be 1-byte aligned. On strict architectures, will this unaligned memory write trigger a hardware alignment fault?

Quoted 3 lines
drivers/hid/hid-nintendo.c +3 -0
+	return ns2->cfg->send_command(NS2_CMD_FLASH, NS2_SUBCMD_FLASH_READ, message,
+		sizeof(message), ns2->cfg);
+}

[ ... ]

Quoted 3 lines
drivers/hid/hid-nintendo.c +3 -0
+int switch2_receive_command(struct switch2_controller *ns2,
+	const uint8_t *message, size_t length)
+{

[ ... ]

Quoted 7 lines
drivers/hid/hid-nintendo.c +7 -0
+	header = (const struct switch2_cmd_header *)message;
+	if (!(header->flags & NS2_FLAG_OK)) {
+		ret = -EIO;
+		goto exit;
+	}
+	message = &message[8];
+	switch (header->command) {

[ ... ]

drivers/hid/hid-nintendo.c static int nintendo_hid_suspend(struct hid_device *hdev, pm_message_t message) · line 3773

Quoted 7 lines
drivers/hid/hid-nintendo.c +7 -0
+	case NS2_CMD_FW_INFO:
+		if (header->subcommand == NS2_SUBCMD_FW_INFO_GET) {
+			if (length < sizeof(ns2->version)) {
+				ret = -EINVAL;
+				goto exit;
+			}
+			memcpy(&ns2->version, message, sizeof(ns2->version));

The message pointer is advanced by 8 bytes to skip the header, but length is not decremented. The bounds check compares length against sizeof(ns2->version). If length is exactly sizeof(ns2->version), the check passes, but the advanced message pointer only has length - 8 valid bytes remaining. Will the memcpy over-read into uninitialized memory?

drivers/hid/hid-nintendo.c static int nintendo_hid_suspend(struct hid_device *hdev, pm_message_t message) · line 3784

Quoted 11 lines
drivers/hid/hid-nintendo.c +11 -0
+			ns2->ctlr_type = ns2->version.ctlr_type;
+			switch2_init_step_done(ns2, NS2_INIT_GET_FIRMWARE_INFO);
+		}
+		break;
+	default:
+		break;
+	}
+
+exit:
+	if (ns2->init_step < NS2_INIT_DONE)
+		switch2_init_controller(ns2);

If a command response has the NACK flag set or encounters another error, the code jumps to exit and invokes switch2_init_controller. Since ns2->init_step has not advanced, the state machine will immediately re-send the exact same command. Can this cause an infinite tight feedback loop of failures and retries, flooding the USB bus?

Quoted 3 lines
drivers/hid/hid-nintendo.c +3 -0
+
+	return ret;
+}

[ ... ]

Quoted 2 lines
drivers/hid/hid-nintendo.c +2 -0
+int switch2_controller_attach_cfg(const char *phys, struct switch2_cfg_intf *cfg)
+{

[ ... ]

drivers/hid/hid-nintendo.c static int nintendo_hid_suspend(struct hid_device *hdev, pm_message_t message) · line 3804

Quoted 6 lines
drivers/hid/hid-nintendo.c +6 -0
+	guard(mutex)(&ns2->lock);
+	WARN_ON(ns2->cfg);
+	ns2->cfg = cfg;
+
+	if (ns2->hdev)
+		return switch2_init_controller(ns2);

If switch2_init_controller fails here or in switch2_probe, the error is returned immediately without unrolling the state. Will ns2->cfg or ns2->hdev remain set, potentially leading to a use-after-free when subsequent device events occur?

Quoted 2 lines
drivers/hid/hid-nintendo.c +2 -0
+	return 0;
+}

[ ... ]

drivers/hid/hid-nintendo.c static int nintendo_hid_suspend(struct hid_device *hdev, pm_message_t message) · line 3908

Quoted 10 lines
drivers/hid/hid-nintendo.c +10 -0
+static void switch2_remove(struct hid_device *hdev)
+{
+	struct switch2_controller *ns2 = hid_get_drvdata(hdev);
+
+	hid_hw_close(hdev);
+	mutex_lock(&ns2->lock);
+	WARN_ON(ns2->hdev != hdev);
+	ns2->hdev = NULL;
+	mutex_unlock(&ns2->lock);
+	ida_free(&nintendo_player_id_allocator, ns2->player_id);

If ida_alloc fails during probe, ns2->player_id remains at its default value of U32_MAX. When ida_free is called, the IDA framework enforces a check on the ID. Casting U32_MAX to a signed integer yields -1. Will this trigger the BUG_ON and cause a kernel panic?

Quoted 3 lines
drivers/hid/hid-nintendo.c +3 -0
+	switch2_controller_put(ns2);
+	hid_hw_stop(hdev);
+}

[ ... ]

Quoted 5 lines
drivers/input/joystick/nintendo-switch2-usb.c new file +0 -0
new file mode 100644
index 0000000000000..ebd89d852e21a

[ ... ]

Quoted 2 lines
drivers/input/joystick/nintendo-switch2-usb.c new file +2 -0
+static void switch2_bulk_in(struct urb *urb)
+{

[ ... ]

drivers/input/joystick/nintendo-switch2-usb.c line 72

Quoted 7 lines
drivers/input/joystick/nintendo-switch2-usb.c new file +7 -0
+	spin_lock_irqsave(&ns2_usb->bulk_in_lock, flags);
+	for (i = 0; i < NS2_IN_URBS; i++) {
+		int err;
+		struct switch2_urb *ns2_urb;
+
+		if (ns2_usb->bulk_in[i].urb == urb) {
+			ns2_usb->message_in = i;

If multiple URBs complete in quick succession before the work item executes, the message_in index is overwritten. Will this cause the work item to only process the last completed URB, leaving earlier URBs indefinitely active and leading to starvation of the URB pool?

Quoted 2 lines
drivers/input/joystick/nintendo-switch2-usb.c new file +2 -0
+			continue;
+		}

[ ... ]

Quoted 2 lines
drivers/input/joystick/nintendo-switch2-usb.c new file +2 -0
+static int switch2_usb_probe(struct usb_interface *intf, const struct usb_device_id *id)
+{

[ ... ]

drivers/input/joystick/nintendo-switch2-usb.c line 271

Quoted 10 lines
drivers/input/joystick/nintendo-switch2-usb.c new file +10 -0
+	ns2_usb->bulk_in[0].active = true;
+	ret = usb_submit_urb(ns2_usb->bulk_in[0].urb, GFP_ATOMIC);
+	if (ret < 0)
+		goto err_free_out;
+
+	init_usb_anchor(&ns2_usb->bulk_out_anchor);
+	spin_lock_init(&ns2_usb->bulk_out_lock);
+	init_usb_anchor(&ns2_usb->bulk_in_anchor);
+	spin_lock_init(&ns2_usb->bulk_in_lock);
+	INIT_WORK(&ns2_usb->message_in_work, switch2_usb_message_in_work);

usb_submit_urb is invoked for bulk_in[0] before the initialization of the spinlocks, anchors, and work struct. If the URB completes immediately, the completion handler will execute and attempt to use uninitialized synchronization primitives. Can this cause a crash?

Quoted 2 lines
drivers/input/joystick/nintendo-switch2-usb.c new file +2 -0
+
+	usb_set_intfdata(intf, ns2_usb);

[ ... ]

drivers/input/joystick/nintendo-switch2-usb.c line 295

Quoted 12 lines
drivers/input/joystick/nintendo-switch2-usb.c new file +12 -0
+err_kill_urb:
+	usb_kill_urb(ns2_usb->bulk_in[0].urb);
+err_free_out:
+	for (i = 0; i < NS2_OUT_URBS; i++) {
+		usb_free_coherent(ns2_usb->udev, NS2_BULK_SIZE, ns2_usb->bulk_out[i].data,
+			ns2_usb->bulk_out[i].urb->transfer_dma);
+		usb_free_urb(ns2_usb->bulk_out[i].urb);
+	}
+err_free_in:
+	for (i = 0; i < NS2_IN_URBS; i++) {
+		usb_free_coherent(ns2_usb->udev, NS2_BULK_SIZE, ns2_usb->bulk_in[i].data,
+			ns2_usb->bulk_in[i].urb->transfer_dma);

If usb_alloc_urb fails during the loop, it jumps to err_free_in. During cleanup, it accesses ns2_usb->bulk_in[i].urb->transfer_dma. Since urb is NULL for the failed index, will this trigger a NULL pointer dereference?

Quoted 5 lines
drivers/input/joystick/nintendo-switch2-usb.c new file +5 -0
+		usb_free_urb(ns2_usb->bulk_in[i].urb);
+	}
+	devm_kfree(&intf->dev, ns2_usb);
+
+	return ret;

drivers/input/joystick/nintendo-switch2-usb.c line 310

Quoted 10 lines
drivers/input/joystick/nintendo-switch2-usb.c new file +10 -0
+}
+
+static void switch2_usb_disconnect(struct usb_interface *intf)
+{
+	struct switch2_usb *ns2_usb = usb_get_intfdata(intf);
+	unsigned long flags;
+	int i;
+
+	spin_lock_irqsave(&ns2_usb->bulk_out_lock, flags);
+	usb_kill_anchored_urbs(&ns2_usb->bulk_out_anchor);

Both usb_kill_anchored_urbs and cancel_work_sync can sleep. Is it safe to call them while holding a spinlock?

drivers/input/joystick/nintendo-switch2-usb.c line 320

Quoted 10 lines
drivers/input/joystick/nintendo-switch2-usb.c new file +10 -0
+	for (i = 0; i < NS2_OUT_URBS; i++) {
+		usb_free_coherent(ns2_usb->udev, NS2_BULK_SIZE, ns2_usb->bulk_out[i].data,
+			ns2_usb->bulk_out[i].urb->transfer_dma);
+		usb_free_urb(ns2_usb->bulk_out[i].urb);
+	}
+	spin_unlock_irqrestore(&ns2_usb->bulk_out_lock, flags);
+
+	spin_lock_irqsave(&ns2_usb->bulk_in_lock, flags);
+	usb_kill_anchored_urbs(&ns2_usb->bulk_in_anchor);
+	cancel_work_sync(&ns2_usb->message_in_work);

Furthermore, cancel_work_sync waits for message_in_work which attempts to acquire bulk_in_lock. Does this cause a deadlock since the lock is already held by the disconnecting thread?

Quoted 1 line
drivers/input/joystick/nintendo-switch2-usb.c new file +1 -0
+	for (i = 0; i < NS2_IN_URBS; i++) {
Signature · 1 line
Sashiko AI review · https://sashiko.dev/#/patchset/[email protected]?part=1
#3 lore raw

Heyhey!

Below some comments regarding the findings (note that I am not an expert on neither Linux kernel drivers in general nor USB HID specifically).

Quoted 35 lines · [email protected] wrote:

Thank you for your contribution! Sashiko AI review found 11 potential issue(s) to consider:
- [High] Sleeping in atomic context and AB-BA deadlock in switch2_usb_disconnect
- [High] Race condition in IN URB tracking causes permanent starvation
- [High] Dangling pointers and use-after-free on probe failure
- [High] Memory corruption and double free due to incorrect devm input lifecycle management
- [High] NULL pointer dereference in USB probe error path
- [High] Division by zero crash from unvalidated trigger calibration data
- [High] BUG_ON kernel panic when ida_free is called with U32_MAX
- [High] Out-of-bounds read when parsing NS2_CMD_FW_INFO response
- [High] Race condition due to URB submission before initialization of synchronization primitives
- [Medium] Infinite USB retry storm on protocol errors
- [Medium] Unaligned memory access on byte array casts
--

commit 00cb7265ad5983144c6923702e2ee2c3c047d29e
Author: Vicki Pfau <[email protected]>
Subject: HID: nintendo: Add preliminary Switch 2 controller driver

This adds a new driver for the Switch 2 controllers. The Switch 2 uses an unusual split-interface design such that input and rumble occur on the main HID interface, but all other communication occurs over a "configuration" interface.

Quoted 4 lines
drivers/hid/hid-nintendo.c +0 -0
index 29008c2cc5304..2193afcdefabc 100644

[ ... ]

Quoted 2 lines
drivers/hid/hid-nintendo.c +2 -0
+static int switch2_init_input(struct switch2_controller *ns2)
+{

[ ... ]

Quoted 3 lines
drivers/hid/hid-nintendo.c +3 -0
+	input = devm_input_allocate_device(&hdev->dev);
+	if (!input)
+		return -ENOMEM;

[ ... ]

Quoted 8 lines

drivers/hid/hid-nintendo.c static int nintendo_hid_suspend(struct hid_device *hdev, pm_message_t message) · line 3238

Quoted 4 lines
drivers/hid/hid-nintendo.c +4 -0
+	ret = input_register_device(input);
+	if (ret < 0) {
+		hid_err(ns2->hdev, "Failed to register input; ret=%d\n", ret);
+		input_free_device(input);

Calling input_free_device or input_unregister_device on a devm-managed input device can cause double frees when devres cleanup fires. Can this lead to memory corruption?

I think sashiko is actually correct here, unfortunately. The documentation for devm_input_allocate_device states that

Managed input devices do not need to be explicitly unregistered or freed as it will be done automatically when owner device unbinds from its driver (or binding fails). ...

Looks like we can remove all the input_unregister_device and input_free_device calls here.

Quoted 9 lines
Quoted 3 lines
drivers/hid/hid-nintendo.c +2 -0
+		return ret;
+	}

[ ... ]

Quoted 4 lines
drivers/hid/hid-nintendo.c +4 -0
+static void switch2_controller_put(struct switch2_controller *ns2)
+{
+	struct input_dev *input;
+	bool do_free;

[ ... ]

Quoted 8 lines

drivers/hid/hid-nintendo.c static int nintendo_hid_suspend(struct hid_device *hdev, pm_message_t message) · line 3309

Quoted 2 lines
drivers/hid/hid-nintendo.c +2 -0
+	if (input)
+		input_unregister_device(input);

Calling input_unregister_device on a devm-managed input device can cause issues. If the input device remains alive (tied to the HID device devres) and its child LEDs are not released, will reconnecting the USB cable allocate a new input device and corrupt the LED lists by re-registering the exact same ns2->leds structures?

This input_unregister_device can also be removed.

Quoted 11 lines
Quoted 3 lines
drivers/hid/hid-nintendo.c +2 -0
+
+	if (do_free) {

[ ... ]

drivers/hid/hid-nintendo.c static int nintendo_hid_suspend(struct hid_device *hdev, pm_message_t message) · line 3502

Quoted 3 lines
drivers/hid/hid-nintendo.c +3 -0
+static void switch2_report_trigger(struct input_dev *input, uint8_t zero, int abs, uint8_t data)
+{
+	int value = (NS2_TRIGGER_RANGE + 1) * (data - zero) / (232 - zero);

The zero variable is loaded directly from the controller's factory calibration flash. If a device reports a zero point of exactly 232, will this trigger a division by zero exception and crash the kernel?

Not sure this is an issue in practice, but we might want to protect the code in any case by doing something like the below perhaps (untested).

int value = (NS2_TRIGGER_RANGE + 1) * (data - zero) / max_t(int, (232 - zero), 1);

Quoted 5 lines
Quoted 4 lines
drivers/hid/hid-nintendo.c +3 -0
+
+	input_report_abs(input, abs, clamp(value, 0, NS2_TRIGGER_RANGE));
+}

[ ... ]

Quoted 12 lines

drivers/hid/hid-nintendo.c static int nintendo_hid_suspend(struct hid_device *hdev, pm_message_t message) · line 3597

Quoted 8 lines
drivers/hid/hid-nintendo.c +8 -0
+static int switch2_read_flash(struct switch2_controller *ns2, uint32_t address,
+	uint8_t size)
+{
+	uint8_t message[8] = { size, 0x7e };
+
+	if (!ns2->cfg)
+		return -ENOTCONN;
+	*(__le32 *)&message[4] = __cpu_to_le32(address);

The driver directly casts &message[4] to __le32. Since message is a uint8_t array, it is only guaranteed to be 1-byte aligned. On strict architectures, will this unaligned memory write trigger a hardware alignment fault?

I assume this could be an issue and we should just use something like the below (only compile-tested).

put_unaligned_le32(__cpu_to_le32(address), &message[4]);

Quoted 17 lines
Quoted 4 lines
drivers/hid/hid-nintendo.c +3 -0
+	return ns2->cfg->send_command(NS2_CMD_FLASH, NS2_SUBCMD_FLASH_READ, message,
+		sizeof(message), ns2->cfg);
+}

[ ... ]

Quoted 3 lines
drivers/hid/hid-nintendo.c +3 -0
+int switch2_receive_command(struct switch2_controller *ns2,
+	const uint8_t *message, size_t length)
+{

[ ... ]

Quoted 7 lines
drivers/hid/hid-nintendo.c +7 -0
+	header = (const struct switch2_cmd_header *)message;
+	if (!(header->flags & NS2_FLAG_OK)) {
+		ret = -EIO;
+		goto exit;
+	}
+	message = &message[8];
+	switch (header->command) {

[ ... ]

Quoted 13 lines

drivers/hid/hid-nintendo.c static int nintendo_hid_suspend(struct hid_device *hdev, pm_message_t message) · line 3773

Quoted 7 lines
drivers/hid/hid-nintendo.c +7 -0
+	case NS2_CMD_FW_INFO:
+		if (header->subcommand == NS2_SUBCMD_FW_INFO_GET) {
+			if (length < sizeof(ns2->version)) {
+				ret = -EINVAL;
+				goto exit;
+			}
+			memcpy(&ns2->version, message, sizeof(ns2->version));

The message pointer is advanced by 8 bytes to skip the header, but length is not decremented. The bounds check compares length against sizeof(ns2->version). If length is exactly sizeof(ns2->version), the check passes, but the advanced message pointer only has length - 8 valid bytes remaining. Will the memcpy over-read into uninitialized memory?

Sounds like something that we should fix by either decreasing length or increasing the size in the check.

Quoted 10 lines
Quoted 10 lines
drivers/hid/hid-nintendo.c +9 -0
+			ns2->ctlr_type = ns2->version.ctlr_type;
+			switch2_init_step_done(ns2, NS2_INIT_GET_FIRMWARE_INFO);
+		}
+		break;
+	default:
+		break;
+	}
+
+exit:
Quoted 8 lines

drivers/hid/hid-nintendo.c static int nintendo_hid_suspend(struct hid_device *hdev, pm_message_t message) · line 3784

Quoted 2 lines
drivers/hid/hid-nintendo.c +2 -0
+	if (ns2->init_step < NS2_INIT_DONE)
+		switch2_init_controller(ns2);

If a command response has the NACK flag set or encounters another error, the code jumps to exit and invokes switch2_init_controller. Since ns2->init_step has not advanced, the state machine will immediately re-send the exact same command. Can this cause an infinite tight feedback loop of failures and retries, flooding the USB bus?

We have discussed this before and it doesn't seem to be an issue in practice.

If we want to be on the safe side, we could pass a retry count and abort on too many error retries, perhaps?

Quoted 7 lines
Quoted 3 lines
drivers/hid/hid-nintendo.c +3 -0
+
+	return ret;
+}

[ ... ]

Quoted 2 lines
drivers/hid/hid-nintendo.c +2 -0
+int switch2_controller_attach_cfg(const char *phys, struct switch2_cfg_intf *cfg)
+{

[ ... ]

Quoted 11 lines

drivers/hid/hid-nintendo.c static int nintendo_hid_suspend(struct hid_device *hdev, pm_message_t message) · line 3804

Quoted 6 lines
drivers/hid/hid-nintendo.c +6 -0
+	guard(mutex)(&ns2->lock);
+	WARN_ON(ns2->cfg);
+	ns2->cfg = cfg;
+
+	if (ns2->hdev)
+		return switch2_init_controller(ns2);

If switch2_init_controller fails here or in switch2_probe, the error is returned immediately without unrolling the state. Will ns2->cfg or ns2->hdev remain set, potentially leading to a use-after-free when subsequent device events occur?

AFAICT, since we are always checking for both ns2->cfg and ns2->hdev before doing any processing in switch2_init_controller, there should never be a use-after-free

It *is* true that we don't call hid_hw_{close,stop} in switch2_probe, if the switch2_init_controller call fails though. It might make sense to return a specific error and use it to check whether we want to give up and call these?

Quoted 4 lines
Quoted 3 lines
drivers/hid/hid-nintendo.c +2 -0
+	return 0;
+}

[ ... ]

Quoted 15 lines

drivers/hid/hid-nintendo.c static int nintendo_hid_suspend(struct hid_device *hdev, pm_message_t message) · line 3908

Quoted 10 lines
drivers/hid/hid-nintendo.c +10 -0
+static void switch2_remove(struct hid_device *hdev)
+{
+	struct switch2_controller *ns2 = hid_get_drvdata(hdev);
+
+	hid_hw_close(hdev);
+	mutex_lock(&ns2->lock);
+	WARN_ON(ns2->hdev != hdev);
+	ns2->hdev = NULL;
+	mutex_unlock(&ns2->lock);
+	ida_free(&nintendo_player_id_allocator, ns2->player_id);

If ida_alloc fails during probe, ns2->player_id remains at its default value of U32_MAX. When ida_free is called, the IDA framework enforces a check on the ID. Casting U32_MAX to a signed integer yields -1. Will this trigger the BUG_ON and cause a kernel panic?

ida_free seems to return early on

if ((int)id < 0)

	return;

so that should be fine as is.

Quoted 33 lines
Quoted 4 lines
drivers/hid/hid-nintendo.c +3 -0
+	switch2_controller_put(ns2);
+	hid_hw_stop(hdev);
+}

[ ... ]

Quoted 5 lines
drivers/input/joystick/nintendo-switch2-usb.c new file +0 -0
new file mode 100644
index 0000000000000..ebd89d852e21a

[ ... ]

Quoted 2 lines
drivers/input/joystick/nintendo-switch2-usb.c new file +2 -0
+static void switch2_bulk_in(struct urb *urb)
+{

[ ... ]

drivers/input/joystick/nintendo-switch2-usb.c line 72

Quoted 7 lines
drivers/input/joystick/nintendo-switch2-usb.c new file +7 -0
+	spin_lock_irqsave(&ns2_usb->bulk_in_lock, flags);
+	for (i = 0; i < NS2_IN_URBS; i++) {
+		int err;
+		struct switch2_urb *ns2_urb;
+
+		if (ns2_usb->bulk_in[i].urb == urb) {
+			ns2_usb->message_in = i;

If multiple URBs complete in quick succession before the work item executes, the message_in index is overwritten. Will this cause the work item to only process the last completed URB, leaving earlier URBs indefinitely active and leading to starvation of the URB pool?

Quoted 2 lines
drivers/input/joystick/nintendo-switch2-usb.c new file +2 -0
+			continue;
+		}

[ ... ]

Quoted 2 lines
drivers/input/joystick/nintendo-switch2-usb.c new file +2 -0
+static int switch2_usb_probe(struct usb_interface *intf, const struct usb_device_id *id)
+{

[ ... ]

Quoted 15 lines

drivers/input/joystick/nintendo-switch2-usb.c line 271

Quoted 10 lines
drivers/input/joystick/nintendo-switch2-usb.c new file +10 -0
+	ns2_usb->bulk_in[0].active = true;
+	ret = usb_submit_urb(ns2_usb->bulk_in[0].urb, GFP_ATOMIC);
+	if (ret < 0)
+		goto err_free_out;
+
+	init_usb_anchor(&ns2_usb->bulk_out_anchor);
+	spin_lock_init(&ns2_usb->bulk_out_lock);
+	init_usb_anchor(&ns2_usb->bulk_in_anchor);
+	spin_lock_init(&ns2_usb->bulk_in_lock);
+	INIT_WORK(&ns2_usb->message_in_work, switch2_usb_message_in_work);

usb_submit_urb is invoked for bulk_in[0] before the initialization of the spinlocks, anchors, and work struct. If the URB completes immediately, the completion handler will execute and attempt to use uninitialized synchronization primitives. Can this cause a crash?

As the documentation for usb_submit_urb mentions that the call is async, I assume that this is indeed a theoretical possibility.

Quoted 4 lines
Quoted 3 lines
drivers/input/joystick/nintendo-switch2-usb.c new file +2 -0
+
+	usb_set_intfdata(intf, ns2_usb);

[ ... ]

Quoted 16 lines

drivers/input/joystick/nintendo-switch2-usb.c line 295

Quoted 12 lines
drivers/input/joystick/nintendo-switch2-usb.c new file +12 -0
+err_kill_urb:
+	usb_kill_urb(ns2_usb->bulk_in[0].urb);
+err_free_out:
+	for (i = 0; i < NS2_OUT_URBS; i++) {
+		usb_free_coherent(ns2_usb->udev, NS2_BULK_SIZE, ns2_usb->bulk_out[i].data,
+			ns2_usb->bulk_out[i].urb->transfer_dma);
+		usb_free_urb(ns2_usb->bulk_out[i].urb);
+	}
+err_free_in:
+	for (i = 0; i < NS2_IN_URBS; i++) {
+		usb_free_coherent(ns2_usb->udev, NS2_BULK_SIZE, ns2_usb->bulk_in[i].data,
+			ns2_usb->bulk_in[i].urb->transfer_dma);

If usb_alloc_urb fails during the loop, it jumps to err_free_in. During cleanup, it accesses ns2_usb->bulk_in[i].urb->transfer_dma. Since urb is NULL for the failed index, will this trigger a NULL pointer dereference?

Yes, most likely we would need a new label for this case.

Quoted 11 lines
Quoted 11 lines
drivers/input/joystick/nintendo-switch2-usb.c new file +10 -0
+		usb_free_urb(ns2_usb->bulk_in[i].urb);
+	}
+	devm_kfree(&intf->dev, ns2_usb);
+
+	return ret;
+}
+
+static void switch2_usb_disconnect(struct usb_interface *intf)
+{
+	struct switch2_usb *ns2_usb = usb_get_intfdata(intf);
Quoted 8 lines

drivers/input/joystick/nintendo-switch2-usb.c line 310

Quoted 5 lines
drivers/input/joystick/nintendo-switch2-usb.c new file +5 -0
+	unsigned long flags;
+	int i;
+
+	spin_lock_irqsave(&ns2_usb->bulk_out_lock, flags);
+	usb_kill_anchored_urbs(&ns2_usb->bulk_out_anchor);

Both usb_kill_anchored_urbs and cancel_work_sync can sleep. Is it safe to call them while holding a spinlock?

According to some googling, this indeed seems to be an isuse. The non-sleeping way to cancel transfer requests would be to use
https://www.kernel.org/doc/html/v6.1/driver-api/usb/usb.html#c.usb_unlink_anchored_urbs

Quoted 15 lines

drivers/input/joystick/nintendo-switch2-usb.c line 320

Quoted 11 lines
drivers/input/joystick/nintendo-switch2-usb.c new file +10 -0
+	for (i = 0; i < NS2_OUT_URBS; i++) {
+		usb_free_coherent(ns2_usb->udev, NS2_BULK_SIZE, ns2_usb->bulk_out[i].data,
+			ns2_usb->bulk_out[i].urb->transfer_dma);
+		usb_free_urb(ns2_usb->bulk_out[i].urb);
+	}
+	spin_unlock_irqrestore(&ns2_usb->bulk_out_lock, flags);
+
+	spin_lock_irqsave(&ns2_usb->bulk_in_lock, flags);
+	usb_kill_anchored_urbs(&ns2_usb->bulk_in_anchor);
+	cancel_work_sync(&ns2_usb->message_in_work);

Furthermore, cancel_work_sync waits for message_in_work which attempts to acquire bulk_in_lock. Does this cause a deadlock since the lock is already held by the disconnecting thread?

Yes, this looks like a real concern as well.

Cheers,
Silvan

Quoted 2 lines
Quoted 2 lines
drivers/input/joystick/nintendo-switch2-usb.c new file +1 -0
+	for (i = 0; i < NS2_IN_URBS; i++) {
author maintainer
#4 lore raw
Quoted 53 lines · Silvan Jegen

Heyhey!

Below some comments regarding the findings (note that I am not an expert on neither Linux kernel drivers in general nor USB HID specifically).

Quoted 35 lines · [email protected] wrote:

Thank you for your contribution! Sashiko AI review found 11 potential issue(s) to consider:
- [High] Sleeping in atomic context and AB-BA deadlock in switch2_usb_disconnect
- [High] Race condition in IN URB tracking causes permanent starvation
- [High] Dangling pointers and use-after-free on probe failure
- [High] Memory corruption and double free due to incorrect devm input lifecycle management
- [High] NULL pointer dereference in USB probe error path
- [High] Division by zero crash from unvalidated trigger calibration data
- [High] BUG_ON kernel panic when ida_free is called with U32_MAX
- [High] Out-of-bounds read when parsing NS2_CMD_FW_INFO response
- [High] Race condition due to URB submission before initialization of synchronization primitives
- [Medium] Infinite USB retry storm on protocol errors
- [Medium] Unaligned memory access on byte array casts
--

commit 00cb7265ad5983144c6923702e2ee2c3c047d29e
Author: Vicki Pfau <[email protected]>
Subject: HID: nintendo: Add preliminary Switch 2 controller driver

This adds a new driver for the Switch 2 controllers. The Switch 2 uses an unusual split-interface design such that input and rumble occur on the main HID interface, but all other communication occurs over a "configuration" interface.

Quoted 4 lines
drivers/hid/hid-nintendo.c +0 -0
index 29008c2cc5304..2193afcdefabc 100644

[ ... ]

Quoted 2 lines
drivers/hid/hid-nintendo.c +2 -0
+static int switch2_init_input(struct switch2_controller *ns2)
+{

[ ... ]

Quoted 3 lines
drivers/hid/hid-nintendo.c +3 -0
+	input = devm_input_allocate_device(&hdev->dev);
+	if (!input)
+		return -ENOMEM;

[ ... ]

Quoted 8 lines

drivers/hid/hid-nintendo.c static int nintendo_hid_suspend(struct hid_device *hdev, pm_message_t message) · line 3238

Quoted 4 lines
drivers/hid/hid-nintendo.c +4 -0
+	ret = input_register_device(input);
+	if (ret < 0) {
+		hid_err(ns2->hdev, "Failed to register input; ret=%d\n", ret);
+		input_free_device(input);

Calling input_free_device or input_unregister_device on a devm-managed input device can cause double frees when devres cleanup fires. Can this lead to memory corruption?

I think sashiko is actually correct here, unfortunately. The documentation for devm_input_allocate_device states that

Quoted 6 lines

Managed input devices do not need to be explicitly unregistered or freed as it will be done automatically when owner device unbinds from its driver (or binding fails). ...

Looks like we can remove all the input_unregister_device and input_free_device calls here.

sashiko is not correct. See my previous reply.

Quoted 33 lines
Quoted 10 lines
Quoted 4 lines
drivers/hid/hid-nintendo.c +2 -0
+		return ret;
+	}

[ ... ]

Quoted 4 lines
drivers/hid/hid-nintendo.c +4 -0
+static void switch2_controller_put(struct switch2_controller *ns2)
+{
+	struct input_dev *input;
+	bool do_free;

[ ... ]

Quoted 8 lines

drivers/hid/hid-nintendo.c static int nintendo_hid_suspend(struct hid_device *hdev, pm_message_t message) · line 3309

Quoted 2 lines
drivers/hid/hid-nintendo.c +2 -0
+	if (input)
+		input_unregister_device(input);

Calling input_unregister_device on a devm-managed input device can cause issues. If the input device remains alive (tied to the HID device devres) and its child LEDs are not released, will reconnecting the USB cable allocate a new input device and corrupt the LED lists by re-registering the exact same ns2->leds structures?

This input_unregister_device can also be removed.

Quoted 11 lines
Quoted 3 lines
drivers/hid/hid-nintendo.c +2 -0
+
+	if (do_free) {

[ ... ]

drivers/hid/hid-nintendo.c static int nintendo_hid_suspend(struct hid_device *hdev, pm_message_t message) · line 3502

Quoted 3 lines
drivers/hid/hid-nintendo.c +3 -0
+static void switch2_report_trigger(struct input_dev *input, uint8_t zero, int abs, uint8_t data)
+{
+	int value = (NS2_TRIGGER_RANGE + 1) * (data - zero) / (232 - zero);

The zero variable is loaded directly from the controller's factory calibration flash. If a device reports a zero point of exactly 232, will this trigger a division by zero exception and crash the kernel?

Quoted 6 lines

Not sure this is an issue in practice, but we might want to protect the code in any case by doing something like the below perhaps (untested).

int value = (NS2_TRIGGER_RANGE + 1) * (data - zero) / max_t(int, (232 - zero), 1);

It won't be an issue in practice but it's harmless to fix.

Quoted 18 lines
Quoted 5 lines
Quoted 4 lines
drivers/hid/hid-nintendo.c +3 -0
+
+	input_report_abs(input, abs, clamp(value, 0, NS2_TRIGGER_RANGE));
+}

[ ... ]

Quoted 12 lines

drivers/hid/hid-nintendo.c static int nintendo_hid_suspend(struct hid_device *hdev, pm_message_t message) · line 3597

Quoted 8 lines
drivers/hid/hid-nintendo.c +8 -0
+static int switch2_read_flash(struct switch2_controller *ns2, uint32_t address,
+	uint8_t size)
+{
+	uint8_t message[8] = { size, 0x7e };
+
+	if (!ns2->cfg)
+		return -ENOTCONN;
+	*(__le32 *)&message[4] = __cpu_to_le32(address);

The driver directly casts &message[4] to __le32. Since message is a uint8_t array, it is only guaranteed to be 1-byte aligned. On strict architectures, will this unaligned memory write trigger a hardware alignment fault?

Quoted 5 lines

I assume this could be an issue and we should just use something like the below (only compile-tested).

put_unaligned_le32(__cpu_to_le32(address), &message[4]);

Correct.

Quoted 54 lines
Quoted 18 lines
Quoted 5 lines
drivers/hid/hid-nintendo.c +3 -0
+	return ns2->cfg->send_command(NS2_CMD_FLASH, NS2_SUBCMD_FLASH_READ, message,
+		sizeof(message), ns2->cfg);
+}

[ ... ]

Quoted 3 lines
drivers/hid/hid-nintendo.c +3 -0
+int switch2_receive_command(struct switch2_controller *ns2,
+	const uint8_t *message, size_t length)
+{

[ ... ]

Quoted 7 lines
drivers/hid/hid-nintendo.c +7 -0
+	header = (const struct switch2_cmd_header *)message;
+	if (!(header->flags & NS2_FLAG_OK)) {
+		ret = -EIO;
+		goto exit;
+	}
+	message = &message[8];
+	switch (header->command) {

[ ... ]

Quoted 13 lines

drivers/hid/hid-nintendo.c static int nintendo_hid_suspend(struct hid_device *hdev, pm_message_t message) · line 3773

Quoted 7 lines
drivers/hid/hid-nintendo.c +7 -0
+	case NS2_CMD_FW_INFO:
+		if (header->subcommand == NS2_SUBCMD_FW_INFO_GET) {
+			if (length < sizeof(ns2->version)) {
+				ret = -EINVAL;
+				goto exit;
+			}
+			memcpy(&ns2->version, message, sizeof(ns2->version));

The message pointer is advanced by 8 bytes to skip the header, but length is not decremented. The bounds check compares length against sizeof(ns2->version). If length is exactly sizeof(ns2->version), the check passes, but the advanced message pointer only has length - 8 valid bytes remaining. Will the memcpy over-read into uninitialized memory?

Sounds like something that we should fix by either decreasing length or increasing the size in the check.

Quoted 10 lines
Quoted 10 lines
drivers/hid/hid-nintendo.c +9 -0
+			ns2->ctlr_type = ns2->version.ctlr_type;
+			switch2_init_step_done(ns2, NS2_INIT_GET_FIRMWARE_INFO);
+		}
+		break;
+	default:
+		break;
+	}
+
+exit:
Quoted 8 lines

drivers/hid/hid-nintendo.c static int nintendo_hid_suspend(struct hid_device *hdev, pm_message_t message) · line 3784

Quoted 2 lines
drivers/hid/hid-nintendo.c +2 -0
+	if (ns2->init_step < NS2_INIT_DONE)
+		switch2_init_controller(ns2);

If a command response has the NACK flag set or encounters another error, the code jumps to exit and invokes switch2_init_controller. Since ns2->init_step has not advanced, the state machine will immediately re-send the exact same command. Can this cause an infinite tight feedback loop of failures and retries, flooding the USB bus?

Quoted 5 lines

We have discussed this before and it doesn't seem to be an issue in practice.

If we want to be on the safe side, we could pass a retry count and abort on too many error retries, perhaps?

I had considered a retry count but getting it to work properly was fiddly. This should only go wrong if there's a malicious device attached, as in practice the Switch 2 controllers won't do this, and bus abuse is trivial with a malicious device regardless.

Quoted 21 lines
Quoted 9 lines
Quoted 5 lines
drivers/hid/hid-nintendo.c +3 -0
+
+	return ret;
+}

[ ... ]

Quoted 2 lines
drivers/hid/hid-nintendo.c +2 -0
+int switch2_controller_attach_cfg(const char *phys, struct switch2_cfg_intf *cfg)
+{

[ ... ]

Quoted 11 lines

drivers/hid/hid-nintendo.c static int nintendo_hid_suspend(struct hid_device *hdev, pm_message_t message) · line 3804

Quoted 6 lines
drivers/hid/hid-nintendo.c +6 -0
+	guard(mutex)(&ns2->lock);
+	WARN_ON(ns2->cfg);
+	ns2->cfg = cfg;
+
+	if (ns2->hdev)
+		return switch2_init_controller(ns2);

If switch2_init_controller fails here or in switch2_probe, the error is returned immediately without unrolling the state. Will ns2->cfg or ns2->hdev remain set, potentially leading to a use-after-free when subsequent device events occur?

Quoted 8 lines

AFAICT, since we are always checking for both ns2->cfg and ns2->hdev before doing any processing in switch2_init_controller, there should never be a use-after-free

It *is* true that we don't call hid_hw_{close,stop} in switch2_probe, if the switch2_init_controller call fails though. It might make sense to return a specific error and use it to check whether we want to give up and call these?

The problem was that it will return an error after setting ns2->cfg, which causes the caller to free it without removing it. That's a real bug that can cause UAF.

Quoted 123 lines
Quoted 5 lines
Quoted 4 lines
drivers/hid/hid-nintendo.c +2 -0
+	return 0;
+}

[ ... ]

Quoted 15 lines

drivers/hid/hid-nintendo.c static int nintendo_hid_suspend(struct hid_device *hdev, pm_message_t message) · line 3908

Quoted 10 lines
drivers/hid/hid-nintendo.c +10 -0
+static void switch2_remove(struct hid_device *hdev)
+{
+	struct switch2_controller *ns2 = hid_get_drvdata(hdev);
+
+	hid_hw_close(hdev);
+	mutex_lock(&ns2->lock);
+	WARN_ON(ns2->hdev != hdev);
+	ns2->hdev = NULL;
+	mutex_unlock(&ns2->lock);
+	ida_free(&nintendo_player_id_allocator, ns2->player_id);

If ida_alloc fails during probe, ns2->player_id remains at its default value of U32_MAX. When ida_free is called, the IDA framework enforces a check on the ID. Casting U32_MAX to a signed integer yields -1. Will this trigger the BUG_ON and cause a kernel panic?

ida_free seems to return early on

if ((int)id < 0)

	return;

so that should be fine as is.

Quoted 33 lines
Quoted 4 lines
drivers/hid/hid-nintendo.c +3 -0
+	switch2_controller_put(ns2);
+	hid_hw_stop(hdev);
+}

[ ... ]

Quoted 5 lines
drivers/input/joystick/nintendo-switch2-usb.c new file +0 -0
new file mode 100644
index 0000000000000..ebd89d852e21a

[ ... ]

Quoted 2 lines
drivers/input/joystick/nintendo-switch2-usb.c new file +2 -0
+static void switch2_bulk_in(struct urb *urb)
+{

[ ... ]

drivers/input/joystick/nintendo-switch2-usb.c line 72

Quoted 7 lines
drivers/input/joystick/nintendo-switch2-usb.c new file +7 -0
+	spin_lock_irqsave(&ns2_usb->bulk_in_lock, flags);
+	for (i = 0; i < NS2_IN_URBS; i++) {
+		int err;
+		struct switch2_urb *ns2_urb;
+
+		if (ns2_usb->bulk_in[i].urb == urb) {
+			ns2_usb->message_in = i;

If multiple URBs complete in quick succession before the work item executes, the message_in index is overwritten. Will this cause the work item to only process the last completed URB, leaving earlier URBs indefinitely active and leading to starvation of the URB pool?

Quoted 2 lines
drivers/input/joystick/nintendo-switch2-usb.c new file +2 -0
+			continue;
+		}

[ ... ]

Quoted 2 lines
drivers/input/joystick/nintendo-switch2-usb.c new file +2 -0
+static int switch2_usb_probe(struct usb_interface *intf, const struct usb_device_id *id)
+{

[ ... ]

Quoted 15 lines

drivers/input/joystick/nintendo-switch2-usb.c line 271

Quoted 10 lines
drivers/input/joystick/nintendo-switch2-usb.c new file +10 -0
+	ns2_usb->bulk_in[0].active = true;
+	ret = usb_submit_urb(ns2_usb->bulk_in[0].urb, GFP_ATOMIC);
+	if (ret < 0)
+		goto err_free_out;
+
+	init_usb_anchor(&ns2_usb->bulk_out_anchor);
+	spin_lock_init(&ns2_usb->bulk_out_lock);
+	init_usb_anchor(&ns2_usb->bulk_in_anchor);
+	spin_lock_init(&ns2_usb->bulk_in_lock);
+	INIT_WORK(&ns2_usb->message_in_work, switch2_usb_message_in_work);

usb_submit_urb is invoked for bulk_in[0] before the initialization of the spinlocks, anchors, and work struct. If the URB completes immediately, the completion handler will execute and attempt to use uninitialized synchronization primitives. Can this cause a crash?

As the documentation for usb_submit_urb mentions that the call is async, I assume that this is indeed a theoretical possibility.

Quoted 4 lines
Quoted 3 lines
drivers/input/joystick/nintendo-switch2-usb.c new file +2 -0
+
+	usb_set_intfdata(intf, ns2_usb);

[ ... ]

Quoted 16 lines

drivers/input/joystick/nintendo-switch2-usb.c line 295

Quoted 12 lines
drivers/input/joystick/nintendo-switch2-usb.c new file +12 -0
+err_kill_urb:
+	usb_kill_urb(ns2_usb->bulk_in[0].urb);
+err_free_out:
+	for (i = 0; i < NS2_OUT_URBS; i++) {
+		usb_free_coherent(ns2_usb->udev, NS2_BULK_SIZE, ns2_usb->bulk_out[i].data,
+			ns2_usb->bulk_out[i].urb->transfer_dma);
+		usb_free_urb(ns2_usb->bulk_out[i].urb);
+	}
+err_free_in:
+	for (i = 0; i < NS2_IN_URBS; i++) {
+		usb_free_coherent(ns2_usb->udev, NS2_BULK_SIZE, ns2_usb->bulk_in[i].data,
+			ns2_usb->bulk_in[i].urb->transfer_dma);

If usb_alloc_urb fails during the loop, it jumps to err_free_in. During cleanup, it accesses ns2_usb->bulk_in[i].urb->transfer_dma. Since urb is NULL for the failed index, will this trigger a NULL pointer dereference?

Yes, most likely we would need a new label for this case.

Quoted 11 lines
Quoted 11 lines
drivers/input/joystick/nintendo-switch2-usb.c new file +10 -0
+		usb_free_urb(ns2_usb->bulk_in[i].urb);
+	}
+	devm_kfree(&intf->dev, ns2_usb);
+
+	return ret;
+}
+
+static void switch2_usb_disconnect(struct usb_interface *intf)
+{
+	struct switch2_usb *ns2_usb = usb_get_intfdata(intf);
Quoted 8 lines

drivers/input/joystick/nintendo-switch2-usb.c line 310

Quoted 5 lines
drivers/input/joystick/nintendo-switch2-usb.c new file +5 -0
+	unsigned long flags;
+	int i;
+
+	spin_lock_irqsave(&ns2_usb->bulk_out_lock, flags);
+	usb_kill_anchored_urbs(&ns2_usb->bulk_out_anchor);

Both usb_kill_anchored_urbs and cancel_work_sync can sleep. Is it safe to call them while holding a spinlock?

Quoted 3 lines

According to some googling, this indeed seems to be an isuse. The non-sleeping way to cancel transfer requests would be to use
https://www.kernel.org/doc/html/v6.1/driver-api/usb/usb.html#c.usb_unlink_anchored_urbs

This was removed in 6.17 and has been unused since 2009.

Quoted 25 lines
Quoted 17 lines

drivers/input/joystick/nintendo-switch2-usb.c line 320

Quoted 13 lines
drivers/input/joystick/nintendo-switch2-usb.c new file +10 -0
+	for (i = 0; i < NS2_OUT_URBS; i++) {
+		usb_free_coherent(ns2_usb->udev, NS2_BULK_SIZE, ns2_usb->bulk_out[i].data,
+			ns2_usb->bulk_out[i].urb->transfer_dma);
+		usb_free_urb(ns2_usb->bulk_out[i].urb);
+	}
+	spin_unlock_irqrestore(&ns2_usb->bulk_out_lock, flags);
+
+	spin_lock_irqsave(&ns2_usb->bulk_in_lock, flags);
+	usb_kill_anchored_urbs(&ns2_usb->bulk_in_anchor);
+	cancel_work_sync(&ns2_usb->message_in_work);

Furthermore, cancel_work_sync waits for message_in_work which attempts to acquire bulk_in_lock. Does this cause a deadlock since the lock is already held by the disconnecting thread?

Yes, this looks like a real concern as well.

Cheers,
Silvan

Quoted 2 lines
Quoted 2 lines
drivers/input/joystick/nintendo-switch2-usb.c new file +1 -0
+	for (i = 0; i < NS2_IN_URBS; i++) {
author maintainer
#5 lore raw
In reply to [email protected]

Guess I'm doing LLM review replies now.

Quoted 35 lines · On 5/13/26 5:15 PM, [email protected] wrote:

Thank you for your contribution! Sashiko AI review found 11 potential issue(s) to consider:
- [High] Sleeping in atomic context and AB-BA deadlock in switch2_usb_disconnect
- [High] Race condition in IN URB tracking causes permanent starvation
- [High] Dangling pointers and use-after-free on probe failure
- [High] Memory corruption and double free due to incorrect devm input lifecycle management
- [High] NULL pointer dereference in USB probe error path
- [High] Division by zero crash from unvalidated trigger calibration data
- [High] BUG_ON kernel panic when ida_free is called with U32_MAX
- [High] Out-of-bounds read when parsing NS2_CMD_FW_INFO response
- [High] Race condition due to URB submission before initialization of synchronization primitives
- [Medium] Infinite USB retry storm on protocol errors
- [Medium] Unaligned memory access on byte array casts
--

commit 00cb7265ad5983144c6923702e2ee2c3c047d29e
Author: Vicki Pfau <[email protected]>
Subject: HID: nintendo: Add preliminary Switch 2 controller driver

This adds a new driver for the Switch 2 controllers. The Switch 2 uses an unusual split-interface design such that input and rumble occur on the main HID interface, but all other communication occurs over a "configuration" interface.

Quoted 4 lines
drivers/hid/hid-nintendo.c +0 -0
index 29008c2cc5304..2193afcdefabc 100644

[ ... ]

Quoted 2 lines
drivers/hid/hid-nintendo.c +2 -0
+static int switch2_init_input(struct switch2_controller *ns2)
+{

[ ... ]

Quoted 3 lines
drivers/hid/hid-nintendo.c +3 -0
+	input = devm_input_allocate_device(&hdev->dev);
+	if (!input)
+		return -ENOMEM;

[ ... ]

Quoted 8 lines

drivers/hid/hid-nintendo.c static int nintendo_hid_suspend(struct hid_device *hdev, pm_message_t message) · line 3238

Quoted 4 lines
drivers/hid/hid-nintendo.c +4 -0
+	ret = input_register_device(input);
+	if (ret < 0) {
+		hid_err(ns2->hdev, "Failed to register input; ret=%d\n", ret);
+		input_free_device(input);

Calling input_free_device or input_unregister_device on a devm-managed input device can cause double frees when devres cleanup fires. Can this lead to memory corruption?

False positive. The documentation clearly states there is no devm equivalent and input_unregister_device works fine on devm-managed devices. The source of input_free_device also clearly shows logic for handling devm, and the source of input_device_unregister calls into a static devm_input_device_unregister if needed.

Quoted 9 lines
Quoted 3 lines
drivers/hid/hid-nintendo.c +2 -0
+		return ret;
+	}

[ ... ]

Quoted 4 lines
drivers/hid/hid-nintendo.c +4 -0
+static void switch2_controller_put(struct switch2_controller *ns2)
+{
+	struct input_dev *input;
+	bool do_free;

[ ... ]

Quoted 8 lines

drivers/hid/hid-nintendo.c static int nintendo_hid_suspend(struct hid_device *hdev, pm_message_t message) · line 3309

Quoted 2 lines
drivers/hid/hid-nintendo.c +2 -0
+	if (input)
+		input_unregister_device(input);

Calling input_unregister_device on a devm-managed input device can cause issues. If the input device remains alive (tied to the HID device devres) and its child LEDs are not released, will reconnecting the USB cable allocate a new input device and corrupt the LED lists by re-registering the exact same ns2->leds structures?

Same false positive

Quoted 11 lines
Quoted 3 lines
drivers/hid/hid-nintendo.c +2 -0
+
+	if (do_free) {

[ ... ]

drivers/hid/hid-nintendo.c static int nintendo_hid_suspend(struct hid_device *hdev, pm_message_t message) · line 3502

Quoted 3 lines
drivers/hid/hid-nintendo.c +3 -0
+static void switch2_report_trigger(struct input_dev *input, uint8_t zero, int abs, uint8_t data)
+{
+	int value = (NS2_TRIGGER_RANGE + 1) * (data - zero) / (232 - zero);

The zero variable is loaded directly from the controller's factory calibration flash. If a device reports a zero point of exactly 232, will this trigger a division by zero exception and crash the kernel?

Ok, an actual finding. I'll have this fixed in v6.

Quoted 5 lines
Quoted 4 lines
drivers/hid/hid-nintendo.c +3 -0
+
+	input_report_abs(input, abs, clamp(value, 0, NS2_TRIGGER_RANGE));
+}

[ ... ]

Quoted 12 lines

drivers/hid/hid-nintendo.c static int nintendo_hid_suspend(struct hid_device *hdev, pm_message_t message) · line 3597

Quoted 8 lines
drivers/hid/hid-nintendo.c +8 -0
+static int switch2_read_flash(struct switch2_controller *ns2, uint32_t address,
+	uint8_t size)
+{
+	uint8_t message[8] = { size, 0x7e };
+
+	if (!ns2->cfg)
+		return -ENOTCONN;
+	*(__le32 *)&message[4] = __cpu_to_le32(address);

The driver directly casts &message[4] to __le32. Since message is a uint8_t array, it is only guaranteed to be 1-byte aligned. On strict architectures, will this unaligned memory write trigger a hardware alignment fault?

Good catch. Fixed in v6.

Quoted 17 lines
Quoted 4 lines
drivers/hid/hid-nintendo.c +3 -0
+	return ns2->cfg->send_command(NS2_CMD_FLASH, NS2_SUBCMD_FLASH_READ, message,
+		sizeof(message), ns2->cfg);
+}

[ ... ]

Quoted 3 lines
drivers/hid/hid-nintendo.c +3 -0
+int switch2_receive_command(struct switch2_controller *ns2,
+	const uint8_t *message, size_t length)
+{

[ ... ]

Quoted 7 lines
drivers/hid/hid-nintendo.c +7 -0
+	header = (const struct switch2_cmd_header *)message;
+	if (!(header->flags & NS2_FLAG_OK)) {
+		ret = -EIO;
+		goto exit;
+	}
+	message = &message[8];
+	switch (header->command) {

[ ... ]

Quoted 13 lines

drivers/hid/hid-nintendo.c static int nintendo_hid_suspend(struct hid_device *hdev, pm_message_t message) · line 3773

Quoted 7 lines
drivers/hid/hid-nintendo.c +7 -0
+	case NS2_CMD_FW_INFO:
+		if (header->subcommand == NS2_SUBCMD_FW_INFO_GET) {
+			if (length < sizeof(ns2->version)) {
+				ret = -EINVAL;
+				goto exit;
+			}
+			memcpy(&ns2->version, message, sizeof(ns2->version));

The message pointer is advanced by 8 bytes to skip the header, but length is not decremented. The bounds check compares length against sizeof(ns2->version). If length is exactly sizeof(ns2->version), the check passes, but the advanced message pointer only has length - 8 valid bytes remaining. Will the memcpy over-read into uninitialized memory?

Fixed in v6.

Quoted 10 lines
Quoted 10 lines
drivers/hid/hid-nintendo.c +9 -0
+			ns2->ctlr_type = ns2->version.ctlr_type;
+			switch2_init_step_done(ns2, NS2_INIT_GET_FIRMWARE_INFO);
+		}
+		break;
+	default:
+		break;
+	}
+
+exit:
Quoted 8 lines

drivers/hid/hid-nintendo.c static int nintendo_hid_suspend(struct hid_device *hdev, pm_message_t message) · line 3784

Quoted 2 lines
drivers/hid/hid-nintendo.c +2 -0
+	if (ns2->init_step < NS2_INIT_DONE)
+		switch2_init_controller(ns2);

If a command response has the NACK flag set or encounters another error, the code jumps to exit and invokes switch2_init_controller. Since ns2->init_step has not advanced, the state machine will immediately re-send the exact same command. Can this cause an infinite tight feedback loop of failures and retries, flooding the USB bus?

This doesn't happen for problems already described in the thread previously.

Quoted 8 lines
Quoted 4 lines
drivers/hid/hid-nintendo.c +3 -0
+
+	return ret;
+}

[ ... ]

Quoted 2 lines
drivers/hid/hid-nintendo.c +2 -0
+int switch2_controller_attach_cfg(const char *phys, struct switch2_cfg_intf *cfg)
+{

[ ... ]

Quoted 11 lines

drivers/hid/hid-nintendo.c static int nintendo_hid_suspend(struct hid_device *hdev, pm_message_t message) · line 3804

Quoted 6 lines
drivers/hid/hid-nintendo.c +6 -0
+	guard(mutex)(&ns2->lock);
+	WARN_ON(ns2->cfg);
+	ns2->cfg = cfg;
+
+	if (ns2->hdev)
+		return switch2_init_controller(ns2);

If switch2_init_controller fails here or in switch2_probe, the error is returned immediately without unrolling the state. Will ns2->cfg or ns2->hdev remain set, potentially leading to a use-after-free when subsequent device events occur?

This is a good catch. I'm not sure what the right fix is though; obviously a half-configured state is bad, but I'm not sure a full teardown is the best approach; it can leave it in a weird state, even if it's not a UAF risk. I'll go with a full teardown for now at least.

Quoted 4 lines
Quoted 3 lines
drivers/hid/hid-nintendo.c +2 -0
+	return 0;
+}

[ ... ]

Quoted 15 lines

drivers/hid/hid-nintendo.c static int nintendo_hid_suspend(struct hid_device *hdev, pm_message_t message) · line 3908

Quoted 10 lines
drivers/hid/hid-nintendo.c +10 -0
+static void switch2_remove(struct hid_device *hdev)
+{
+	struct switch2_controller *ns2 = hid_get_drvdata(hdev);
+
+	hid_hw_close(hdev);
+	mutex_lock(&ns2->lock);
+	WARN_ON(ns2->hdev != hdev);
+	ns2->hdev = NULL;
+	mutex_unlock(&ns2->lock);
+	ida_free(&nintendo_player_id_allocator, ns2->player_id);

If ida_alloc fails during probe, ns2->player_id remains at its default value of U32_MAX. When ida_free is called, the IDA framework enforces a check on the ID. Casting U32_MAX to a signed integer yields -1. Will this trigger the BUG_ON and cause a kernel panic?

I can't find any evidence of such a BUG_ON. A negative number will just early return. This seems like a hallucination.

Quoted 14 lines
Quoted 4 lines
drivers/hid/hid-nintendo.c +3 -0
+	switch2_controller_put(ns2);
+	hid_hw_stop(hdev);
+}

[ ... ]

Quoted 5 lines
drivers/input/joystick/nintendo-switch2-usb.c new file +0 -0
new file mode 100644
index 0000000000000..ebd89d852e21a

[ ... ]

Quoted 2 lines
drivers/input/joystick/nintendo-switch2-usb.c new file +2 -0
+static void switch2_bulk_in(struct urb *urb)
+{

[ ... ]

Quoted 13 lines

drivers/input/joystick/nintendo-switch2-usb.c line 72

Quoted 7 lines
drivers/input/joystick/nintendo-switch2-usb.c new file +7 -0
+	spin_lock_irqsave(&ns2_usb->bulk_in_lock, flags);
+	for (i = 0; i < NS2_IN_URBS; i++) {
+		int err;
+		struct switch2_urb *ns2_urb;
+
+		if (ns2_usb->bulk_in[i].urb == urb) {
+			ns2_usb->message_in = i;

If multiple URBs complete in quick succession before the work item executes, the message_in index is overwritten. Will this cause the work item to only process the last completed URB, leaving earlier URBs indefinitely active and leading to starvation of the URB pool?

In practice this probably won't happen but I guess it's worth fixing.

Quoted 6 lines
Quoted 2 lines
drivers/input/joystick/nintendo-switch2-usb.c new file +2 -0
+			continue;
+		}

[ ... ]

Quoted 2 lines
drivers/input/joystick/nintendo-switch2-usb.c new file +2 -0
+static int switch2_usb_probe(struct usb_interface *intf, const struct usb_device_id *id)
+{

[ ... ]

Quoted 15 lines

drivers/input/joystick/nintendo-switch2-usb.c line 271

Quoted 10 lines
drivers/input/joystick/nintendo-switch2-usb.c new file +10 -0
+	ns2_usb->bulk_in[0].active = true;
+	ret = usb_submit_urb(ns2_usb->bulk_in[0].urb, GFP_ATOMIC);
+	if (ret < 0)
+		goto err_free_out;
+
+	init_usb_anchor(&ns2_usb->bulk_out_anchor);
+	spin_lock_init(&ns2_usb->bulk_out_lock);
+	init_usb_anchor(&ns2_usb->bulk_in_anchor);
+	spin_lock_init(&ns2_usb->bulk_in_lock);
+	INIT_WORK(&ns2_usb->message_in_work, switch2_usb_message_in_work);

usb_submit_urb is invoked for bulk_in[0] before the initialization of the spinlocks, anchors, and work struct. If the URB completes immediately, the completion handler will execute and attempt to use uninitialized synchronization primitives. Can this cause a crash?

I don't know if this is possible but I guess I'll fix it.

Quoted 4 lines
Quoted 3 lines
drivers/input/joystick/nintendo-switch2-usb.c new file +2 -0
+
+	usb_set_intfdata(intf, ns2_usb);

[ ... ]

Quoted 16 lines

drivers/input/joystick/nintendo-switch2-usb.c line 295

Quoted 12 lines
drivers/input/joystick/nintendo-switch2-usb.c new file +12 -0
+err_kill_urb:
+	usb_kill_urb(ns2_usb->bulk_in[0].urb);
+err_free_out:
+	for (i = 0; i < NS2_OUT_URBS; i++) {
+		usb_free_coherent(ns2_usb->udev, NS2_BULK_SIZE, ns2_usb->bulk_out[i].data,
+			ns2_usb->bulk_out[i].urb->transfer_dma);
+		usb_free_urb(ns2_usb->bulk_out[i].urb);
+	}
+err_free_in:
+	for (i = 0; i < NS2_IN_URBS; i++) {
+		usb_free_coherent(ns2_usb->udev, NS2_BULK_SIZE, ns2_usb->bulk_in[i].data,
+			ns2_usb->bulk_in[i].urb->transfer_dma);

If usb_alloc_urb fails during the loop, it jumps to err_free_in. During cleanup, it accesses ns2_usb->bulk_in[i].urb->transfer_dma. Since urb is NULL for the failed index, will this trigger a NULL pointer dereference?

I removed the NULL check due to checkpatch telling me I didn't need it but I guess I'll put it back.

Quoted 11 lines
Quoted 11 lines
drivers/input/joystick/nintendo-switch2-usb.c new file +10 -0
+		usb_free_urb(ns2_usb->bulk_in[i].urb);
+	}
+	devm_kfree(&intf->dev, ns2_usb);
+
+	return ret;
+}
+
+static void switch2_usb_disconnect(struct usb_interface *intf)
+{
+	struct switch2_usb *ns2_usb = usb_get_intfdata(intf);
Quoted 8 lines

drivers/input/joystick/nintendo-switch2-usb.c line 310

Quoted 5 lines
drivers/input/joystick/nintendo-switch2-usb.c new file +5 -0
+	unsigned long flags;
+	int i;
+
+	spin_lock_irqsave(&ns2_usb->bulk_out_lock, flags);
+	usb_kill_anchored_urbs(&ns2_usb->bulk_out_anchor);

Both usb_kill_anchored_urbs and cancel_work_sync can sleep. Is it safe to call them while holding a spinlock?

Fixing in v6.

Quoted 15 lines

drivers/input/joystick/nintendo-switch2-usb.c line 320

Quoted 11 lines
drivers/input/joystick/nintendo-switch2-usb.c new file +10 -0
+	for (i = 0; i < NS2_OUT_URBS; i++) {
+		usb_free_coherent(ns2_usb->udev, NS2_BULK_SIZE, ns2_usb->bulk_out[i].data,
+			ns2_usb->bulk_out[i].urb->transfer_dma);
+		usb_free_urb(ns2_usb->bulk_out[i].urb);
+	}
+	spin_unlock_irqrestore(&ns2_usb->bulk_out_lock, flags);
+
+	spin_lock_irqsave(&ns2_usb->bulk_in_lock, flags);
+	usb_kill_anchored_urbs(&ns2_usb->bulk_in_anchor);
+	cancel_work_sync(&ns2_usb->message_in_work);

Furthermore, cancel_work_sync waits for message_in_work which attempts to acquire bulk_in_lock. Does this cause a deadlock since the lock is already held by the disconnecting thread?

Likewise.

Quoted 3 lines
Quoted 2 lines
drivers/input/joystick/nintendo-switch2-usb.c new file +1 -0
+	for (i = 0; i < NS2_IN_URBS; i++) {
#6 lore raw

PATCH 1/1 HID: nintendo: fix clang build of switch2_probe guard/goto

In reply to Vicki Pfau

clang rejects the forward goto err_close/err_stop in switch2_probe() because they jump across the scope of the guard(mutex)(&ns2->lock) cleanup variable ("cannot jump from this goto statement to its label"). All three early error paths run before the lock is acquired, so replace the gotos with explicit hid_hw_close()/hid_hw_stop() cleanup and drop the now-unused labels. No functional change; gcc accepted the original.

---

Hi Vicki, I've been testing your Switch 2 series on a local machine here with JC2, GC, and Pro all over USB and hits this clang build break, Patch below , feel free to just squash it in.

drivers/hid/hid-nintendo.c:4147:3: error: cannot jump from this goto statement to its label
drivers/hid/hid-nintendo.c:4153:3: error: cannot jump from this goto statement to its label

with the explanatory notes pointing at:
note: jump bypasses initialization of variable with __attribute__((cleanup))
guard(mutex)(&ns2->lock);

 drivers/hid/hid-nintendo.c | 14 +++++---------
 1 file changed, 5 insertions(+), 9 deletions(-)
drivers/hid/hid-nintendo.c +5 -9
index 40f57036d810..6a286d49e5bd 100644
@@ -4144,13 +4144,16 @@ static int switch2_probe(struct hid_device *hdev, const struct hid_device_id *id
 	ret = hid_hw_open(hdev);
 	if (ret) {
 		hid_err(hdev, "hw_open failed %d\n", ret);
-		goto err_stop;
+		hid_hw_stop(hdev);
+		return ret;
 	}
 
 	ns2 = switch2_get_controller(phys);
 	if (IS_ERR(ns2)) {
 		ret = PTR_ERR(ns2);
-		goto err_close;
+		hid_hw_close(hdev);
+		hid_hw_stop(hdev);
+		return ret;
 	}
 
 	guard(mutex)(&ns2->lock);
@@ -4194,13 +4197,6 @@ static int switch2_probe(struct hid_device *hdev, const struct hid_device_id *id
 		return switch2_init_controller(ns2);
 
 	return 0;
-
-err_close:
-	hid_hw_close(hdev);
-err_stop:
-	hid_hw_stop(hdev);
-
-	return ret;
 }
 
 static void switch2_remove(struct hid_device *hdev)
Signature · 1 line
2.54.0
author maintainer
#7 lore raw

PATCH v5 2/3 HID: nintendo: Add rumble support for Switch 2 controllers

In reply to Vicki Pfau

This adds rumble support for both the "HD Rumble" linear resonant actuator type as used in the Joy-Cons and Pro Controller, as well as the eccentric rotating mass type used in the GameCube controller. Note that since there's currently no API for exposing full control of LRAs with evdev, it only simulates a basic rumble for now.

---
 drivers/hid/Kconfig        |   8 +-
 drivers/hid/hid-nintendo.c | 179 ++++++++++++++++++++++++++++++++++++-
 2 files changed, 181 insertions(+), 6 deletions(-)
drivers/hid/Kconfig +4 -4
index 19c77c323ec9..851eed76c236 100644
@@ -859,10 +859,10 @@ config NINTENDO_FF
 	depends on HID_NINTENDO
 	select INPUT_FF_MEMLESS
 	help
-	Say Y here if you have a Nintendo Switch controller and want to enable
-	force feedback support for it. This works for both joy-cons, the pro
-	controller, and the NSO N64 controller. For the pro controller, both
-	rumble motors can be controlled individually.
+	Say Y here if you have a Nintendo Switch or Switch 2 controller and want
+	to enable force feedback support for it. This works for Joy-Cons, the Pro
+	Controllers, and the NSO N64 and GameCube controller. For the Pro
+	Controller, both rumble motors can be controlled individually.
 
 config HID_NTI
 	tristate "NTI keyboard adapters"
drivers/hid/hid-nintendo.c +177 -2

231 lines not on this page · raw message

Signature · 1 line
2.54.0
#8 lore raw

Re: [PATCH v5 2/3] HID: nintendo: Add rumble support for Switch 2 controllers

Thank you for your contribution! Sashiko AI review found 5 potential issue(s) to consider: - [High] Mixing jiffies and milliseconds in delayed work scheduling logic causes an integer underflow…

Thank you for your contribution! Sashiko AI review found 5 potential issue(s) to consider:
- [High] Mixing jiffies and milliseconds in delayed work scheduling logic causes an integer underflow, permanently stalling the haptic rumble.
- [High] A guaranteed NULL pointer dereference occurs in the error logging path of switch2_rumble_work, as well as a TOCTOU race condition.
- [Medium] ns2->ctlr_type is checked before it is dynamically initialized, causing union corruption for GameCube controllers.
- [Low] Implicit type promotion in the max() macro causes a compiler warning.
- [Low] Kconfig documentation claims individual motor control for the Pro Controller, but the implementation identicalizes the rumble data for both motors. --

commit d72f0f8e0dadde2cacc27050b92088c2e5cbea0e
Author: Vicki Pfau <[email protected]>
Subject: HID: nintendo: Add rumble support for Switch 2 controllers

This commit introduces rumble support for Switch 2 controllers, handling both HD Rumble linear resonant actuators and eccentric rotating mass haptics. It adds state machines for rumble scheduling and output reporting via hdev.

Quoted 6 lines
drivers/hid/Kconfig +0 -0
index 26ef589f652c6..a7fe74d982475 100644
@@ -850,10 +850,10 @@ config NINTENDO_FF
 	depends on HID_NINTENDO

drivers/hid/Kconfig config NINTENDO_FF · line 856

Quoted 10 lines
drivers/hid/Kconfig +4 -4
 	select INPUT_FF_MEMLESS
 	help
-	Say Y here if you have a Nintendo Switch controller and want to enable
-	force feedback support for it. This works for both joy-cons, the pro
-	controller, and the NSO N64 controller. For the pro controller, both
-	rumble motors can be controlled individually.
+	Say Y here if you have a Nintendo Switch or Switch 2 controller and want
+	to enable force feedback support for it. This works for Joy-Cons, the Pro
+	Controllers, and the NSO N64 and GameCube controller. For the Pro
+	Controller, both rumble motors can be controlled individually.

This isn't a bug, but is the help text accurate regarding individual motor control? The implementation in switch2_rumble_work() appears to encode the exact same rumble configuration to both the left and right motor offsets, which would make both motors vibrate identically.

Quoted 4 lines
drivers/hid/hid-nintendo.c +0 -0
index 2193afcdefabc..9801799a90758 100644

[ ... ]

Quoted 15 lines
drivers/hid/hid-nintendo.c +11 -0
@@ -3103,6 +3155,107 @@ static inline bool switch2_ctlr_is_joycon(enum switch2_ctlr_type type)
 	return type == NS2_CTLR_TYPE_JCL || type == NS2_CTLR_TYPE_JCR;
 }
 
+#if IS_ENABLED(CONFIG_NINTENDO_FF)
+static void switch2_encode_rumble(struct switch2_hd_rumble *rumble, uint8_t buffer[5])
+{
+	buffer[0] = rumble->hi_freq;
+	buffer[1] = (rumble->hi_freq >> 8) | (rumble->hi_amp << 2);
+	buffer[2] = (rumble->hi_amp >> 6) | (rumble->lo_freq << 4);
+	buffer[3] = (rumble->lo_freq >> 4) | (rumble->lo_amp << 6);
+	buffer[4] = rumble->lo_amp >> 2;
+}
+
+static int switch2_play_effect(struct input_dev *dev, void *data, struct ff_effect *effect)

drivers/hid/hid-nintendo.c static inline bool switch2_ctlr_is_joycon(enum switch2_ctlr_type type) · line 3178

Quoted 10 lines
drivers/hid/hid-nintendo.c +10 -0
+{
+	struct switch2_controller *ns2 = input_get_drvdata(dev);
+
+	if (effect->type != FF_RUMBLE)
+		return 0;
+
+	guard(spinlock_irqsave)(&ns2->rumble_lock);
+	if (ns2->ctlr_type == NS2_CTLR_TYPE_GC) {
+		ns2->rumble.sd.amplitude = max(effect->u.rumble.strong_magnitude,
+			effect->u.rumble.weak_magnitude >> 1);

Will this max calculation trigger a compiler warning? The strict type checking in the max macro might fail because effect->u.rumble.weak_magnitude >> 1 promotes to an int, while strong_magnitude is a u16.

Quoted 19 lines
drivers/hid/hid-nintendo.c +19 -0
+	} else {
+		ns2->rumble.hd.hi_amp = effect->u.rumble.weak_magnitude * RUMBLE_MAX >> 16;
+		ns2->rumble.hd.lo_amp = effect->u.rumble.strong_magnitude * RUMBLE_MAX >> 16;
+	}
+
+	if (ns2->hdev)
+		schedule_delayed_work(&ns2->rumble_work, 0);
+
+	return 0;
+}
+
+static void switch2_rumble_work(struct work_struct *work)
+{
+	struct switch2_controller *ns2 = container_of(to_delayed_work(work),
+						      struct switch2_controller, rumble_work);
+	unsigned long current_ms = jiffies_to_msecs(get_jiffies_64());
+	unsigned long flags;
+	bool active;
+	int ret;

[ ... ]

drivers/hid/hid-nintendo.c static inline bool switch2_ctlr_is_joycon(enum switch2_ctlr_type type) · line 3241

Quoted 9 lines
drivers/hid/hid-nintendo.c +9 -0
+	if (active) {
+		unsigned long interval = msecs_to_jiffies(4);
+
+		if (!ns2->last_rumble_work)
+			ns2->last_rumble_work = current_ms;
+		else
+			ns2->last_rumble_work += interval;
+		schedule_delayed_work(&ns2->rumble_work,
+			ns2->last_rumble_work + interval - current_ms);

Does this code underflow the scheduled delay?

The current_ms variable is retrieved in milliseconds using jiffies_to_msecs(), but interval is calculated in jiffies using msecs_to_jiffies(). Since ns2->last_rumble_work is initialized to milliseconds but incremented by jiffies, subtracting current_ms could yield a negative value.

On systems where HZ is not 1000, could this underflow the delay parameter and schedule the next work millions of jiffies in the future?

Quoted 4 lines
drivers/hid/hid-nintendo.c +4 -0
+	} else {
+		ns2->last_rumble_work = 0;
+	}
+	spin_unlock_irqrestore(&ns2->rumble_lock, flags);

drivers/hid/hid-nintendo.c static inline bool switch2_ctlr_is_joycon(enum switch2_ctlr_type type) · line 3255

Quoted 10 lines
drivers/hid/hid-nintendo.c +10 -0
+
+	if (!ns2->hdev) {
+		cancel_delayed_work(&ns2->rumble_work);
+		ret = -ENODEV;
+	} else {
+		ret = hid_hw_output_report(ns2->hdev, ns2->rumble_buffer, 64);
+	}
+
+	if (ret < 0)
+		hid_dbg(ns2->hdev, "Failed to send output report ret=%d\n", ret);

If ns2->hdev is NULL, we set ret = -ENODEV and then reach this hid_dbg call. Could this result in a NULL pointer dereference, as the hid_dbg macro passes ns2->hdev to dev_dbg()?

Also, since ns2->hdev is accessed locklessly and without READ_ONCE(), is there a time-of-check to time-of-use race where the compiler might reload the pointer between the check and hid_hw_output_report()?

[ ... ]

drivers/hid/hid-nintendo.c static int switch2_probe(struct hid_device *hdev, const struct hid_device_id *id · line 4055

Quoted 9 lines
drivers/hid/hid-nintendo.c +5 -0
@@ -3881,6 +4048,14 @@ static int switch2_probe(struct hid_device *hdev, const struct hid_device_id *id
 
 	switch2_leds_create(ns2);
 
+#if IS_ENABLED(CONFIG_NINTENDO_FF)
+	if (ns2->ctlr_type != NS2_CTLR_TYPE_GC) {
+		ns2->rumble.hd.hi_freq = RUMBLE_HI_FREQ;
+		ns2->rumble.hd.lo_freq = RUMBLE_LO_FREQ;
+	}

Are we checking ns2->ctlr_type before it is dynamically initialized?

During probe, ns2->ctlr_type appears to be 0 since the struct is freshly allocated and the type is discovered later during the init state machine. This would cause the high and low frequencies to always be written.

Since hd and sd share a union, could this corrupt the sd.error and sd.amplitude fields for actual GameCube controllers?

Quoted 4 lines
drivers/hid/hid-nintendo.c +3 -0
+	spin_lock_init(&ns2->rumble_lock);
+	INIT_DELAYED_WORK(&ns2->rumble_work, switch2_rumble_work);
+#endif
 	hid_set_drvdata(hdev, ns2);
Signature · 1 line
Sashiko AI review · https://sashiko.dev/#/patchset/[email protected]?part=2
#9 lore raw

Re: [PATCH v5 2/3] HID: nintendo: Add rumble support for Switch 2 controllers

Quoted 29 lines · [email protected] wrote:

Thank you for your contribution! Sashiko AI review found 5 potential issue(s) to consider:
- [High] Mixing jiffies and milliseconds in delayed work scheduling logic causes an integer underflow, permanently stalling the haptic rumble.
- [High] A guaranteed NULL pointer dereference occurs in the error logging path of switch2_rumble_work, as well as a TOCTOU race condition.
- [Medium] ns2->ctlr_type is checked before it is dynamically initialized, causing union corruption for GameCube controllers.
- [Low] Implicit type promotion in the max() macro causes a compiler warning.
- [Low] Kconfig documentation claims individual motor control for the Pro Controller, but the implementation identicalizes the rumble data for both motors. --

commit d72f0f8e0dadde2cacc27050b92088c2e5cbea0e
Author: Vicki Pfau <[email protected]>
Subject: HID: nintendo: Add rumble support for Switch 2 controllers

This commit introduces rumble support for Switch 2 controllers, handling both HD Rumble linear resonant actuators and eccentric rotating mass haptics. It adds state machines for rumble scheduling and output reporting via hdev.

Quoted 13 lines
drivers/hid/Kconfig +1 -4
index 26ef589f652c6..a7fe74d982475 100644
@@ -850,10 +850,10 @@ config NINTENDO_FF
 	depends on HID_NINTENDO
 	select INPUT_FF_MEMLESS
 	help
-	Say Y here if you have a Nintendo Switch controller and want to enable
-	force feedback support for it. This works for both joy-cons, the pro
-	controller, and the NSO N64 controller. For the pro controller, both
-	rumble motors can be controlled individually.
+	Say Y here if you have a Nintendo Switch or Switch 2 controller and want
Quoted 8 lines

drivers/hid/Kconfig config NINTENDO_FF · line 865

Quoted 3 lines
drivers/hid/Kconfig +3 -0
+	to enable force feedback support for it. This works for Joy-Cons, the Pro
+	Controllers, and the NSO N64 and GameCube controller. For the Pro
+	Controller, both rumble motors can be controlled individually.

This isn't a bug, but is the help text accurate regarding individual motor control? The implementation in switch2_rumble_work() appears to encode the exact same rumble configuration to both the left and right motor offsets, which would make both motors vibrate identically.

As pointers to different offsets into the rumble_buffer are passed, I assume that the vibrations will actually not be identical.

Quoted 29 lines
Quoted 5 lines
drivers/hid/hid-nintendo.c +0 -0
index 2193afcdefabc..9801799a90758 100644

[ ... ]

Quoted 21 lines
drivers/hid/hid-nintendo.c +17 -0
@@ -3103,6 +3155,107 @@ static inline bool switch2_ctlr_is_joycon(enum switch2_ctlr_type type)
 	return type == NS2_CTLR_TYPE_JCL || type == NS2_CTLR_TYPE_JCR;
 }
 
+#if IS_ENABLED(CONFIG_NINTENDO_FF)
+static void switch2_encode_rumble(struct switch2_hd_rumble *rumble, uint8_t buffer[5])
+{
+	buffer[0] = rumble->hi_freq;
+	buffer[1] = (rumble->hi_freq >> 8) | (rumble->hi_amp << 2);
+	buffer[2] = (rumble->hi_amp >> 6) | (rumble->lo_freq << 4);
+	buffer[3] = (rumble->lo_freq >> 4) | (rumble->lo_amp << 6);
+	buffer[4] = rumble->lo_amp >> 2;
+}
+
+static int switch2_play_effect(struct input_dev *dev, void *data, struct ff_effect *effect)
+{
+	struct switch2_controller *ns2 = input_get_drvdata(dev);
+
+	if (effect->type != FF_RUMBLE)
+		return 0;
+
Quoted 8 lines

drivers/hid/hid-nintendo.c static inline bool switch2_ctlr_is_joycon(enum switch2_ctlr_type type) · line 3178

Quoted 4 lines
drivers/hid/hid-nintendo.c +4 -0
+	guard(spinlock_irqsave)(&ns2->rumble_lock);
+	if (ns2->ctlr_type == NS2_CTLR_TYPE_GC) {
+		ns2->rumble.sd.amplitude = max(effect->u.rumble.strong_magnitude,
+			effect->u.rumble.weak_magnitude >> 1);

Will this max calculation trigger a compiler warning? The strict type checking in the max macro might fail because effect->u.rumble.weak_magnitude >> 1 promotes to an int, while strong_magnitude is a u16.

Apparently it is not strict enough to emit a warning on my machine at least ...

Quoted 35 lines
Quoted 20 lines
drivers/hid/hid-nintendo.c +19 -0
+	} else {
+		ns2->rumble.hd.hi_amp = effect->u.rumble.weak_magnitude * RUMBLE_MAX >> 16;
+		ns2->rumble.hd.lo_amp = effect->u.rumble.strong_magnitude * RUMBLE_MAX >> 16;
+	}
+
+	if (ns2->hdev)
+		schedule_delayed_work(&ns2->rumble_work, 0);
+
+	return 0;
+}
+
+static void switch2_rumble_work(struct work_struct *work)
+{
+	struct switch2_controller *ns2 = container_of(to_delayed_work(work),
+						      struct switch2_controller, rumble_work);
+	unsigned long current_ms = jiffies_to_msecs(get_jiffies_64());
+	unsigned long flags;
+	bool active;
+	int ret;

[ ... ]

drivers/hid/hid-nintendo.c static inline bool switch2_ctlr_is_joycon(enum switch2_ctlr_type type) · line 3241

Quoted 9 lines
drivers/hid/hid-nintendo.c +9 -0
+	if (active) {
+		unsigned long interval = msecs_to_jiffies(4);
+
+		if (!ns2->last_rumble_work)
+			ns2->last_rumble_work = current_ms;
+		else
+			ns2->last_rumble_work += interval;
+		schedule_delayed_work(&ns2->rumble_work,
+			ns2->last_rumble_work + interval - current_ms);

Does this code underflow the scheduled delay?

Quoted 7 lines

The current_ms variable is retrieved in milliseconds using jiffies_to_msecs(), but interval is calculated in jiffies using msecs_to_jiffies(). Since ns2->last_rumble_work is initialized to milliseconds but incremented by jiffies, subtracting current_ms could yield a negative value.

On systems where HZ is not 1000, could this underflow the delay parameter and schedule the next work millions of jiffies in the future?

It's definitely suspicious that we are adding jiffies to milliseconds. We should probably fix that.

Quoted 11 lines
Quoted 11 lines
drivers/hid/hid-nintendo.c +10 -0
+	} else {
+		ns2->last_rumble_work = 0;
+	}
+	spin_unlock_irqrestore(&ns2->rumble_lock, flags);
+
+	if (!ns2->hdev) {
+		cancel_delayed_work(&ns2->rumble_work);
+		ret = -ENODEV;
+	} else {
+		ret = hid_hw_output_report(ns2->hdev, ns2->rumble_buffer, 64);
Quoted 8 lines

drivers/hid/hid-nintendo.c static inline bool switch2_ctlr_is_joycon(enum switch2_ctlr_type type) · line 3255

Quoted 4 lines
drivers/hid/hid-nintendo.c +4 -0
+	}
+
+	if (ret < 0)
+		hid_dbg(ns2->hdev, "Failed to send output report ret=%d\n", ret);

If ns2->hdev is NULL, we set ret = -ENODEV and then reach this hid_dbg call. Could this result in a NULL pointer dereference, as the hid_dbg macro passes ns2->hdev to dev_dbg()?

Since we set ret = -ENODEV when ns2->hdev is NULL we would indeed end up calling hid_dbg() with a NULL argument.

Quoted 19 lines

Also, since ns2->hdev is accessed locklessly and without READ_ONCE(), is there a time-of-check to time-of-use race where the compiler might reload the pointer between the check and hid_hw_output_report()?

[ ... ]

drivers/hid/hid-nintendo.c static int switch2_probe(struct hid_device *hdev, const struct hid_device_id *id · line 4055

Quoted 9 lines
drivers/hid/hid-nintendo.c +5 -0
@@ -3881,6 +4048,14 @@ static int switch2_probe(struct hid_device *hdev, const struct hid_device_id *id
 
 	switch2_leds_create(ns2);
 
+#if IS_ENABLED(CONFIG_NINTENDO_FF)
+	if (ns2->ctlr_type != NS2_CTLR_TYPE_GC) {
+		ns2->rumble.hd.hi_freq = RUMBLE_HI_FREQ;
+		ns2->rumble.hd.lo_freq = RUMBLE_LO_FREQ;
+	}

Are we checking ns2->ctlr_type before it is dynamically initialized?

Quoted 3 lines

During probe, ns2->ctlr_type appears to be 0 since the struct is freshly allocated and the type is discovered later during the init state machine. This would cause the high and low frequencies to always be written.

As switch2_init_controller is called from the other driver's probe function, this might be fine. To avoid writing these values if the initialisation has not happened yet, we should probably guard against a 0 value.

Cheers,
Silvan

Quoted 8 lines

Since hd and sd share a union, could this corrupt the sd.error and sd.amplitude fields for actual GameCube controllers?

Quoted 4 lines
drivers/hid/hid-nintendo.c +3 -0
+	spin_lock_init(&ns2->rumble_lock);
+	INIT_DELAYED_WORK(&ns2->rumble_work, switch2_rumble_work);
+#endif
 	hid_set_drvdata(hdev, ns2);
author maintainer
#10 lore raw

Re: [PATCH v5 2/3] HID: nintendo: Add rumble support for Switch 2 controllers

In reply to [email protected]

Let's tackle some more hallucinations.

Quoted 29 lines · On 5/13/26 5:53 PM, [email protected] wrote:

Thank you for your contribution! Sashiko AI review found 5 potential issue(s) to consider:
- [High] Mixing jiffies and milliseconds in delayed work scheduling logic causes an integer underflow, permanently stalling the haptic rumble.
- [High] A guaranteed NULL pointer dereference occurs in the error logging path of switch2_rumble_work, as well as a TOCTOU race condition.
- [Medium] ns2->ctlr_type is checked before it is dynamically initialized, causing union corruption for GameCube controllers.
- [Low] Implicit type promotion in the max() macro causes a compiler warning.
- [Low] Kconfig documentation claims individual motor control for the Pro Controller, but the implementation identicalizes the rumble data for both motors. --

commit d72f0f8e0dadde2cacc27050b92088c2e5cbea0e
Author: Vicki Pfau <[email protected]>
Subject: HID: nintendo: Add rumble support for Switch 2 controllers

This commit introduces rumble support for Switch 2 controllers, handling both HD Rumble linear resonant actuators and eccentric rotating mass haptics. It adds state machines for rumble scheduling and output reporting via hdev.

Quoted 13 lines
drivers/hid/Kconfig +1 -4
index 26ef589f652c6..a7fe74d982475 100644
@@ -850,10 +850,10 @@ config NINTENDO_FF
 	depends on HID_NINTENDO
 	select INPUT_FF_MEMLESS
 	help
-	Say Y here if you have a Nintendo Switch controller and want to enable
-	force feedback support for it. This works for both joy-cons, the pro
-	controller, and the NSO N64 controller. For the pro controller, both
-	rumble motors can be controlled individually.
+	Say Y here if you have a Nintendo Switch or Switch 2 controller and want
Quoted 8 lines

drivers/hid/Kconfig config NINTENDO_FF · line 865

Quoted 3 lines
drivers/hid/Kconfig +3 -0
+	to enable force feedback support for it. This works for Joy-Cons, the Pro
+	Controllers, and the NSO N64 and GameCube controller. For the Pro
+	Controller, both rumble motors can be controlled individually.

This isn't a bug, but is the help text accurate regarding individual motor control? The implementation in switch2_rumble_work() appears to encode the exact same rumble configuration to both the left and right motor offsets, which would make both motors vibrate identically.

I didn't add this text, so I'm ingoring this.

Quoted 29 lines
Quoted 5 lines
drivers/hid/hid-nintendo.c +0 -0
index 2193afcdefabc..9801799a90758 100644

[ ... ]

Quoted 21 lines
drivers/hid/hid-nintendo.c +17 -0
@@ -3103,6 +3155,107 @@ static inline bool switch2_ctlr_is_joycon(enum switch2_ctlr_type type)
 	return type == NS2_CTLR_TYPE_JCL || type == NS2_CTLR_TYPE_JCR;
 }
 
+#if IS_ENABLED(CONFIG_NINTENDO_FF)
+static void switch2_encode_rumble(struct switch2_hd_rumble *rumble, uint8_t buffer[5])
+{
+	buffer[0] = rumble->hi_freq;
+	buffer[1] = (rumble->hi_freq >> 8) | (rumble->hi_amp << 2);
+	buffer[2] = (rumble->hi_amp >> 6) | (rumble->lo_freq << 4);
+	buffer[3] = (rumble->lo_freq >> 4) | (rumble->lo_amp << 6);
+	buffer[4] = rumble->lo_amp >> 2;
+}
+
+static int switch2_play_effect(struct input_dev *dev, void *data, struct ff_effect *effect)
+{
+	struct switch2_controller *ns2 = input_get_drvdata(dev);
+
+	if (effect->type != FF_RUMBLE)
+		return 0;
+
Quoted 8 lines

drivers/hid/hid-nintendo.c static inline bool switch2_ctlr_is_joycon(enum switch2_ctlr_type type) · line 3178

Quoted 4 lines
drivers/hid/hid-nintendo.c +4 -0
+	guard(spinlock_irqsave)(&ns2->rumble_lock);
+	if (ns2->ctlr_type == NS2_CTLR_TYPE_GC) {
+		ns2->rumble.sd.amplitude = max(effect->u.rumble.strong_magnitude,
+			effect->u.rumble.weak_magnitude >> 1);

Will this max calculation trigger a compiler warning? The strict type checking in the max macro might fail because effect->u.rumble.weak_magnitude >> 1 promotes to an int, while strong_magnitude is a u16.

It doesn't cause a warning locally, but my warnings seem lax for some reason. I'm not sure how to fix that, but I guess adding a cast here is harmless.

Quoted 35 lines
Quoted 20 lines
drivers/hid/hid-nintendo.c +19 -0
+	} else {
+		ns2->rumble.hd.hi_amp = effect->u.rumble.weak_magnitude * RUMBLE_MAX >> 16;
+		ns2->rumble.hd.lo_amp = effect->u.rumble.strong_magnitude * RUMBLE_MAX >> 16;
+	}
+
+	if (ns2->hdev)
+		schedule_delayed_work(&ns2->rumble_work, 0);
+
+	return 0;
+}
+
+static void switch2_rumble_work(struct work_struct *work)
+{
+	struct switch2_controller *ns2 = container_of(to_delayed_work(work),
+						      struct switch2_controller, rumble_work);
+	unsigned long current_ms = jiffies_to_msecs(get_jiffies_64());
+	unsigned long flags;
+	bool active;
+	int ret;

[ ... ]

drivers/hid/hid-nintendo.c static inline bool switch2_ctlr_is_joycon(enum switch2_ctlr_type type) · line 3241

Quoted 9 lines
drivers/hid/hid-nintendo.c +9 -0
+	if (active) {
+		unsigned long interval = msecs_to_jiffies(4);
+
+		if (!ns2->last_rumble_work)
+			ns2->last_rumble_work = current_ms;
+		else
+			ns2->last_rumble_work += interval;
+		schedule_delayed_work(&ns2->rumble_work,
+			ns2->last_rumble_work + interval - current_ms);

Does this code underflow the scheduled delay?

Quoted 7 lines

The current_ms variable is retrieved in milliseconds using jiffies_to_msecs(), but interval is calculated in jiffies using msecs_to_jiffies(). Since ns2->last_rumble_work is initialized to milliseconds but incremented by jiffies, subtracting current_ms could yield a negative value.

On systems where HZ is not 1000, could this underflow the delay parameter and schedule the next work millions of jiffies in the future?

Fixing in v6.

Quoted 11 lines
Quoted 11 lines
drivers/hid/hid-nintendo.c +10 -0
+	} else {
+		ns2->last_rumble_work = 0;
+	}
+	spin_unlock_irqrestore(&ns2->rumble_lock, flags);
+
+	if (!ns2->hdev) {
+		cancel_delayed_work(&ns2->rumble_work);
+		ret = -ENODEV;
+	} else {
+		ret = hid_hw_output_report(ns2->hdev, ns2->rumble_buffer, 64);
Quoted 8 lines

drivers/hid/hid-nintendo.c static inline bool switch2_ctlr_is_joycon(enum switch2_ctlr_type type) · line 3255

Quoted 4 lines
drivers/hid/hid-nintendo.c +4 -0
+	}
+
+	if (ret < 0)
+		hid_dbg(ns2->hdev, "Failed to send output report ret=%d\n", ret);

If ns2->hdev is NULL, we set ret = -ENODEV and then reach this hid_dbg call. Could this result in a NULL pointer dereference, as the hid_dbg macro passes ns2->hdev to dev_dbg()?

Fixing in v6

Quoted 4 lines

Also, since ns2->hdev is accessed locklessly and without READ_ONCE(), is there a time-of-check to time-of-use race where the compiler might reload the pointer between the check and hid_hw_output_report()?

This can only get changed in switch2_remove, which is missing a cancel_delayed_work_sync. Fixing in v6.

Quoted 15 lines

[ ... ]

drivers/hid/hid-nintendo.c static int switch2_probe(struct hid_device *hdev, const struct hid_device_id *id · line 4055

Quoted 9 lines
drivers/hid/hid-nintendo.c +5 -0
@@ -3881,6 +4048,14 @@ static int switch2_probe(struct hid_device *hdev, const struct hid_device_id *id
 
 	switch2_leds_create(ns2);
 
+#if IS_ENABLED(CONFIG_NINTENDO_FF)
+	if (ns2->ctlr_type != NS2_CTLR_TYPE_GC) {
+		ns2->rumble.hd.hi_freq = RUMBLE_HI_FREQ;
+		ns2->rumble.hd.lo_freq = RUMBLE_LO_FREQ;
+	}

Are we checking ns2->ctlr_type before it is dynamically initialized?

Quoted 6 lines

During probe, ns2->ctlr_type appears to be 0 since the struct is freshly allocated and the type is discovered later during the init state machine. This would cause the high and low frequencies to always be written.

Since hd and sd share a union, could this corrupt the sd.error and sd.amplitude fields for actual GameCube controllers?

Really, really unimportant and technically "works by accident" territory but I'll move the logic anyway.

Quoted 6 lines
Quoted 5 lines
drivers/hid/hid-nintendo.c +3 -0
+	spin_lock_init(&ns2->rumble_lock);
+	INIT_DELAYED_WORK(&ns2->rumble_work, switch2_rumble_work);
+#endif
 	hid_set_drvdata(hdev, ns2);
author maintainer
#11 lore raw

PATCH v5 3/3 HID: nintendo: Add unified report format support

In reply to Vicki Pfau

This adds support for the "unified" report format that all controllers also support, which has overlapping fields for like buttons and axes between them.

---
 drivers/hid/hid-nintendo.c | 148 +++++++++++++++++++++++++++++++++++--
 1 file changed, 143 insertions(+), 5 deletions(-)
drivers/hid/hid-nintendo.c +143 -5
index 9801799a9075..9e7eda7b70bf 100644
@@ -2824,6 +2824,36 @@ static int joycon_suspend(struct hid_device *hdev, pm_message_t message)
 #define NS2_BTN3_SR	BIT(6)
 #define NS2_BTN3_SL	BIT(7)
 
+#define NS2_BTN_U1_Y	BIT(0)
+#define NS2_BTN_U1_X	BIT(1)
+#define NS2_BTN_U1_B	BIT(2)
+#define NS2_BTN_U1_A	BIT(3)
+#define NS2_BTN_U1_SR	BIT(4)
+#define NS2_BTN_U1_SL	BIT(5)
+#define NS2_BTN_U1_R	BIT(6)
+#define NS2_BTN_U1_ZR	BIT(7)
+
+#define NS2_BTN_U2_MINUS	BIT(0)
+#define NS2_BTN_U2_PLUS		BIT(1)
+#define NS2_BTN_U2_RS		BIT(2)
+#define NS2_BTN_U2_LS		BIT(3)
+#define NS2_BTN_U2_HOME		BIT(4)
+#define NS2_BTN_U2_CAPTURE	BIT(5)
+#define NS2_BTN_U2_C		BIT(6)
+
+#define NS2_BTN_U3_DOWN		BIT(0)
+#define NS2_BTN_U3_UP		BIT(1)
+#define NS2_BTN_U3_RIGHT	BIT(2)
+#define NS2_BTN_U3_LEFT		BIT(3)
+#define NS2_BTN_U3_SR		BIT(4)
+#define NS2_BTN_U3_SL		BIT(5)
+#define NS2_BTN_U3_L		BIT(6)
+#define NS2_BTN_U3_ZL		BIT(7)
+
+#define NS2_BTN_U4_GR		BIT(0)
+#define NS2_BTN_U4_GL		BIT(1)
+#define NS2_BTN_U4_HEADSET	BIT(5)
+
 #define NS2_BTN_JCR_HOME	BIT(0)
 #define NS2_BTN_JCR_GR		BIT(2)
 #define NS2_BTN_JCR_C		NS2_BTN3_C
@@ -3069,6 +3099,22 @@ static const struct switch2_ctlr_button_mapping ns2_left_joycon_button_mappings[
 	{ /* sentinel */ },
 };
 
+static const struct switch2_ctlr_button_mapping ns2_left_joycon_button_unified_mappings[] = {
+	{ BTN_DPAD_LEFT,	2, NS2_BTN_U3_LEFT,	},
+	{ BTN_DPAD_UP,		2, NS2_BTN_U3_UP,	},
+	{ BTN_DPAD_DOWN,	2, NS2_BTN_U3_DOWN,	},
+	{ BTN_DPAD_RIGHT,	2, NS2_BTN_U3_RIGHT,	},
+	{ BTN_TL,		2, NS2_BTN_U3_L,	},
+	{ BTN_TL2,		2, NS2_BTN_U3_ZL,	},
+	{ BTN_SELECT,		1, NS2_BTN_U2_MINUS,	},
+	{ BTN_THUMBL,		1, NS2_BTN_U2_LS,	},
+	{ KEY_RECORD,		1, NS2_BTN_U2_CAPTURE,	},
+	{ BTN_GRIPR,		2, NS2_BTN_U3_SL,	},
+	{ BTN_GRIPR2,		2, NS2_BTN_U3_SR,	},
+	{ BTN_GRIPL,		3, NS2_BTN_U4_GL,	},
+	{ /* sentinel */ },
+};
+
 static const struct switch2_ctlr_button_mapping ns2_right_joycon_button_mappings[] = {
 	{ BTN_SOUTH,	0, NS2_BTNR_A,		},
 	{ BTN_EAST,	0, NS2_BTNR_B,		},
@@ -3086,6 +3132,23 @@ static const struct switch2_ctlr_button_mapping ns2_right_joycon_button_mappings
 	{ /* sentinel */ },
 };
 
+static const struct switch2_ctlr_button_mapping ns2_right_joycon_button_unified_mappings[] = {
+	{ BTN_SOUTH,	0, NS2_BTN_U1_A,	},
+	{ BTN_EAST,	0, NS2_BTN_U1_B,	},
+	{ BTN_NORTH,	0, NS2_BTN_U1_X,	},
+	{ BTN_WEST,	0, NS2_BTN_U1_Y,	},
+	{ BTN_TR,	0, NS2_BTN_U1_R,	},
+	{ BTN_TR2,	0, NS2_BTN_U1_ZR	},
+	{ BTN_START,	1, NS2_BTN_U2_PLUS,	},
+	{ BTN_THUMBR,	1, NS2_BTN_U2_RS,	},
+	{ BTN_C,	1, NS2_BTN_U2_C,	},
+	{ BTN_MODE,	1, NS2_BTN_U2_HOME,	},
+	{ BTN_GRIPL2,	0, NS2_BTN_U1_SL,	},
+	{ BTN_GRIPL,	0, NS2_BTN_U1_SR,	},
+	{ BTN_GRIPR,	3, NS2_BTN_U4_GR,	},
+	{ /* sentinel */ },
+};
+
 static const struct switch2_ctlr_button_mapping ns2_procon_mappings[] = {
 	{ BTN_SOUTH,	0, NS2_BTNR_A,		},
 	{ BTN_EAST,	0, NS2_BTNR_B,		},
@@ -3107,6 +3170,27 @@ static const struct switch2_ctlr_button_mapping ns2_procon_mappings[] = {
 	{ /* sentinel */ },
 };
 
+static const struct switch2_ctlr_button_mapping ns2_procon_unified_mappings[] = {
+	{ BTN_SOUTH,	0, NS2_BTN_U1_A,	},
+	{ BTN_EAST,	0, NS2_BTN_U1_B,	},
+	{ BTN_NORTH,	0, NS2_BTN_U1_X,	},
+	{ BTN_WEST,	0, NS2_BTN_U1_Y,	},
+	{ BTN_TL,	2, NS2_BTN_U3_L,	},
+	{ BTN_TR,	0, NS2_BTN_U1_R,	},
+	{ BTN_TL2,	2, NS2_BTN_U3_ZL,	},
+	{ BTN_TR2,	0, NS2_BTN_U1_ZR,	},
+	{ BTN_SELECT,	1, NS2_BTN_U2_MINUS,	},
+	{ BTN_START,	1, NS2_BTN_U2_PLUS,	},
+	{ BTN_THUMBL,	1, NS2_BTN_U2_LS,	},
+	{ BTN_THUMBR,	1, NS2_BTN_U2_RS,	},
+	{ BTN_MODE,	1, NS2_BTN_U2_HOME	},
+	{ KEY_RECORD,	1, NS2_BTN_U2_CAPTURE	},
+	{ BTN_GRIPR,	3, NS2_BTN_U4_GR	},
+	{ BTN_GRIPL,	3, NS2_BTN_U4_GL	},
+	{ BTN_C,	1, NS2_BTN_U2_C		},
+	{ /* sentinel */ },
+};
+
 static const struct switch2_ctlr_button_mapping ns2_gccon_mappings[] = {
 	{ BTN_SOUTH,	0, NS2_BTNR_A,		},
 	{ BTN_EAST,	0, NS2_BTNR_B,		},
@@ -3124,6 +3208,23 @@ static const struct switch2_ctlr_button_mapping ns2_gccon_mappings[] = {
 	{ /* sentinel */ },
 };
 
+static const struct switch2_ctlr_button_mapping ns2_gccon_unified_mappings[] = {
+	{ BTN_SOUTH,	0, NS2_BTN_U1_A,	},
+	{ BTN_EAST,	0, NS2_BTN_U1_B,	},
+	{ BTN_NORTH,	0, NS2_BTN_U1_X,	},
+	{ BTN_WEST,	0, NS2_BTN_U1_Y,	},
+	{ BTN_TL2,	2, NS2_BTN_U3_L,	},
+	{ BTN_TR2,	0, NS2_BTN_U1_R,	},
+	{ BTN_TL,	2, NS2_BTN_U3_ZL	},
+	{ BTN_TR,	0, NS2_BTN_U1_ZR	},
+	{ BTN_SELECT,	1, NS2_BTN_U2_MINUS,	},
+	{ BTN_START,	1, NS2_BTN_U2_PLUS,	},
+	{ BTN_MODE,	1, NS2_BTN_U2_HOME	},
+	{ KEY_RECORD,	1, NS2_BTN_U2_CAPTURE	},
+	{ BTN_C,	1, NS2_BTN_U2_C		},
+	{ /* sentinel */ },
+};
+
 static const uint8_t switch2_init_cmd_data[] = {
 	/*
 	 * The last 6 bytes of this packet are the MAC address of
@@ -3690,11 +3791,48 @@ static int switch2_event(struct hid_device *hdev, struct hid_report *report, uin
 
 	switch (report->id) {
 	case NS2_REPORT_UNIFIED:
-		/*
-		 * TODO
-		 * This won't be sent unless the report type gets changed via command
-		 * 03-0A, but we should support it at some point regardless.
-		 */
+		switch (ns2->ctlr_type) {
+		case NS2_CTLR_TYPE_JCL:
+			switch2_report_stick(input, &ns2->stick_calib[0],
+				ABS_X, false, ABS_Y, true, &raw_data[11]);
+			switch2_report_buttons(input, &raw_data[5],
+				ns2_left_joycon_button_unified_mappings);
+			break;
+		case NS2_CTLR_TYPE_JCR:
+			switch2_report_stick(input, &ns2->stick_calib[0],
+				ABS_X, false, ABS_Y, true, &raw_data[14]);
+			switch2_report_buttons(input, &raw_data[5],
+				ns2_right_joycon_button_unified_mappings);
+			break;
+		case NS2_CTLR_TYPE_GC:
+			input_report_abs(input, ABS_HAT0X,
+				!!(raw_data[7] & NS2_BTN_U3_RIGHT) -
+				!!(raw_data[7] & NS2_BTN_U3_LEFT));
+			input_report_abs(input, ABS_HAT0Y,
+				!!(raw_data[7] & NS2_BTN_U3_DOWN) -
+				!!(raw_data[7] & NS2_BTN_U3_UP));
+			switch2_report_buttons(input, &raw_data[5], ns2_gccon_unified_mappings);
+			switch2_report_stick(input, &ns2->stick_calib[0],
+				ABS_X, false, ABS_Y, true, &raw_data[11]);
+			switch2_report_stick(input, &ns2->stick_calib[1],
+				ABS_RX, false, ABS_RY, true, &raw_data[14]);
+			switch2_report_trigger(input, ns2->lt_zero, ABS_Z, raw_data[0x3d]);
+			switch2_report_trigger(input, ns2->rt_zero, ABS_RZ, raw_data[0x3e]);
+			break;
+		case NS2_CTLR_TYPE_PRO:
+			input_report_abs(input, ABS_HAT0X,
+				!!(raw_data[7] & NS2_BTN_U3_RIGHT) -
+				!!(raw_data[7] & NS2_BTN_U3_LEFT));
+			input_report_abs(input, ABS_HAT0Y,
+				!!(raw_data[7] & NS2_BTN_U3_DOWN) -
+				!!(raw_data[7] & NS2_BTN_U3_UP));
+			switch2_report_buttons(input, &raw_data[5], ns2_procon_unified_mappings);
+			switch2_report_stick(input, &ns2->stick_calib[0],
+				ABS_X, false, ABS_Y, true, &raw_data[11]);
+			switch2_report_stick(input, &ns2->stick_calib[1],
+				ABS_RX, false, ABS_RY, true, &raw_data[14]);
+			break;
+		}
 		break;
 	case NS2_REPORT_JCL:
 		switch2_report_stick(input, &ns2->stick_calib[0], ABS_X, false,
Signature · 1 line
2.54.0
#12 lore raw

Re: [PATCH v5 3/3] HID: nintendo: Add unified report format support

Thank you for your contribution! Sashiko AI review found 2 potential issue(s) to consider: - [High] Insufficient bounds checking for NS2_REPORT_UNIFIED payload leads to out-of-bounds reads. - [High]…

Thank you for your contribution! Sashiko AI review found 2 potential issue(s) to consider:
- [High] Insufficient bounds checking for NS2_REPORT_UNIFIED payload leads to out-of-bounds reads.
- [High] Divide-by-zero kernel panic in switch2_report_trigger() caused by attacker-controlled calibration data. --

commit 6e45b564fde8014fc6ade8452a1b793097b3bee4
Author: Vicki Pfau <[email protected]>
Subject: HID: nintendo: Add unified report format support

This commit adds support for the "unified" report format to the nintendo hid driver. It introduces new mappings for overlapping button and axis fields across different controller types.

Quoted 4 lines
drivers/hid/hid-nintendo.c +0 -0
index 9801799a90758..9e7eda7b70bfb 100644

[ ... ]

Quoted 9 lines
drivers/hid/hid-nintendo.c +0 -5
@@ -3690,11 +3791,48 @@ static int switch2_event(struct hid_device *hdev, struct hid_report *report, uin
 
 	switch (report->id) {
 	case NS2_REPORT_UNIFIED:
-		/*
-		 * TODO
-		 * This won't be sent unless the report type gets changed via command
-		 * 03-0A, but we should support it at some point regardless.
-		 */

drivers/hid/hid-nintendo.c static int switch2_event(struct hid_device *hdev, struct hid_report *report, uin · line 3803

Quoted 10 lines
drivers/hid/hid-nintendo.c +10 -0
+		switch (ns2->ctlr_type) {
+		case NS2_CTLR_TYPE_JCL:
+			switch2_report_stick(input, &ns2->stick_calib[0],
+				ABS_X, false, ABS_Y, true, &raw_data[11]);
+			switch2_report_buttons(input, &raw_data[5],
+				ns2_left_joycon_button_unified_mappings);
+			break;
+		case NS2_CTLR_TYPE_JCR:
+			switch2_report_stick(input, &ns2->stick_calib[0],
+				ABS_X, false, ABS_Y, true, &raw_data[14]);

Is this safe if the report size is exactly 15 bytes?

Earlier in switch2_event(), there is a global size check:

    if (size < 15)
        return -EINVAL;

However, switch2_report_stick() reads up to 3 bytes from the provided data pointer (indices 0, 1, and 2). Passing &raw_data[14] means it will read raw_data[14], raw_data[15], and raw_data[16].

Could this cause an out-of-bounds read if an undersized report is sent?

Quoted 7 lines
drivers/hid/hid-nintendo.c +7 -0
+			switch2_report_buttons(input, &raw_data[5],
+				ns2_right_joycon_button_unified_mappings);
+			break;
+		case NS2_CTLR_TYPE_GC:
+			input_report_abs(input, ABS_HAT0X,
+				!!(raw_data[7] & NS2_BTN_U3_RIGHT) -
+				!!(raw_data[7] & NS2_BTN_U3_LEFT));

drivers/hid/hid-nintendo.c static int switch2_event(struct hid_device *hdev, struct hid_report *report, uin · line 3820

Quoted 10 lines
drivers/hid/hid-nintendo.c +10 -0
+			input_report_abs(input, ABS_HAT0Y,
+				!!(raw_data[7] & NS2_BTN_U3_DOWN) -
+				!!(raw_data[7] & NS2_BTN_U3_UP));
+			switch2_report_buttons(input, &raw_data[5], ns2_gccon_unified_mappings);
+			switch2_report_stick(input, &ns2->stick_calib[0],
+				ABS_X, false, ABS_Y, true, &raw_data[11]);
+			switch2_report_stick(input, &ns2->stick_calib[1],
+				ABS_RX, false, ABS_RY, true, &raw_data[14]);
+			switch2_report_trigger(input, ns2->lt_zero, ABS_Z, raw_data[0x3d]);
+			switch2_report_trigger(input, ns2->rt_zero, ABS_RZ, raw_data[0x3e]);

Similarly, does this code read past the end of the buffer?

Accessing raw_data[0x3d] (index 61) and raw_data[0x3e] (index 62) requires the report size to be at least 63 bytes, but the global check only guarantees 15 bytes.

Additionally, could this new path trigger a divide-by-zero panic?

While this issue pre-dates this commit, switch2_report_trigger() contains the following calculation:

    int value = (NS2_TRIGGER_RANGE + 1) * (data - zero) / (232 - zero);

The zero parameter here comes from the device's factory calibration data (ns2->lt_zero and ns2->rt_zero). If a device advertises 232 as its zero calibration point, this division evaluates to 0.

Since this patch exposes a new way to call switch2_report_trigger(), might it be worth adding validation to prevent a zero divisor?

Quoted 2 lines
drivers/hid/hid-nintendo.c +2 -0
+			break;
+		case NS2_CTLR_TYPE_PRO:

[ ... ]

Signature · 1 line
Sashiko AI review · https://sashiko.dev/#/patchset/[email protected]?part=3
#13 lore raw

Re: [PATCH v5 3/3] HID: nintendo: Add unified report format support

Quoted 42 lines · [email protected] wrote:

Thank you for your contribution! Sashiko AI review found 2 potential issue(s) to consider:
- [High] Insufficient bounds checking for NS2_REPORT_UNIFIED payload leads to out-of-bounds reads.
- [High] Divide-by-zero kernel panic in switch2_report_trigger() caused by attacker-controlled calibration data. --

commit 6e45b564fde8014fc6ade8452a1b793097b3bee4
Author: Vicki Pfau <[email protected]>
Subject: HID: nintendo: Add unified report format support

This commit adds support for the "unified" report format to the nintendo hid driver. It introduces new mappings for overlapping button and axis fields across different controller types.

Quoted 4 lines
drivers/hid/hid-nintendo.c +0 -0
index 9801799a90758..9e7eda7b70bfb 100644

[ ... ]

Quoted 9 lines
drivers/hid/hid-nintendo.c +0 -5
@@ -3690,11 +3791,48 @@ static int switch2_event(struct hid_device *hdev, struct hid_report *report, uin
 
 	switch (report->id) {
 	case NS2_REPORT_UNIFIED:
-		/*
-		 * TODO
-		 * This won't be sent unless the report type gets changed via command
-		 * 03-0A, but we should support it at some point regardless.
-		 */

drivers/hid/hid-nintendo.c static int switch2_event(struct hid_device *hdev, struct hid_report *report, uin · line 3803

Quoted 10 lines
drivers/hid/hid-nintendo.c +10 -0
+		switch (ns2->ctlr_type) {
+		case NS2_CTLR_TYPE_JCL:
+			switch2_report_stick(input, &ns2->stick_calib[0],
+				ABS_X, false, ABS_Y, true, &raw_data[11]);
+			switch2_report_buttons(input, &raw_data[5],
+				ns2_left_joycon_button_unified_mappings);
+			break;
+		case NS2_CTLR_TYPE_JCR:
+			switch2_report_stick(input, &ns2->stick_calib[0],
+				ABS_X, false, ABS_Y, true, &raw_data[14]);

Is this safe if the report size is exactly 15 bytes?

Earlier in switch2_event(), there is a global size check:

Quoted 8 lines
    if (size < 15)
        return -EINVAL;

However, switch2_report_stick() reads up to 3 bytes from the provided data pointer (indices 0, 1, and 2). Passing &raw_data[14] means it will read raw_data[14], raw_data[15], and raw_data[16].

Could this cause an out-of-bounds read if an undersized report is sent?

Yes, I assume we would have to check for a bigger size here.

Quoted 21 lines
Quoted 8 lines
drivers/hid/hid-nintendo.c +7 -0
+			switch2_report_buttons(input, &raw_data[5],
+				ns2_right_joycon_button_unified_mappings);
+			break;
+		case NS2_CTLR_TYPE_GC:
+			input_report_abs(input, ABS_HAT0X,
+				!!(raw_data[7] & NS2_BTN_U3_RIGHT) -
+				!!(raw_data[7] & NS2_BTN_U3_LEFT));

drivers/hid/hid-nintendo.c static int switch2_event(struct hid_device *hdev, struct hid_report *report, uin · line 3820

Quoted 10 lines
drivers/hid/hid-nintendo.c +10 -0
+			input_report_abs(input, ABS_HAT0Y,
+				!!(raw_data[7] & NS2_BTN_U3_DOWN) -
+				!!(raw_data[7] & NS2_BTN_U3_UP));
+			switch2_report_buttons(input, &raw_data[5], ns2_gccon_unified_mappings);
+			switch2_report_stick(input, &ns2->stick_calib[0],
+				ABS_X, false, ABS_Y, true, &raw_data[11]);
+			switch2_report_stick(input, &ns2->stick_calib[1],
+				ABS_RX, false, ABS_RY, true, &raw_data[14]);
+			switch2_report_trigger(input, ns2->lt_zero, ABS_Z, raw_data[0x3d]);
+			switch2_report_trigger(input, ns2->rt_zero, ABS_RZ, raw_data[0x3e]);

Similarly, does this code read past the end of the buffer?

Quoted 3 lines

Accessing raw_data[0x3d] (index 61) and raw_data[0x3e] (index 62) requires the report size to be at least 63 bytes, but the global check only guarantees 15 bytes.

Same here.

Quoted 6 lines

Additionally, could this new path trigger a divide-by-zero panic?

While this issue pre-dates this commit, switch2_report_trigger() contains the following calculation:

Quoted 8 lines
    int value = (NS2_TRIGGER_RANGE + 1) * (data - zero) / (232 - zero);

The zero parameter here comes from the device's factory calibration data (ns2->lt_zero and ns2->rt_zero). If a device advertises 232 as its zero calibration point, this division evaluates to 0.

Since this patch exposes a new way to call switch2_report_trigger(), might it be worth adding validation to prevent a zero divisor?

This one has been reported already in one of the earlier emails.

Cheers,
Silvan

Quoted 4 lines
Quoted 3 lines
drivers/hid/hid-nintendo.c +2 -0
+			break;
+		case NS2_CTLR_TYPE_PRO:

[ ... ]

maintainer
#14 lore raw
In reply to Vicki Pfau
Quoted 10 lines · Vicki Pfau

This series adds preliminary support for Switch 2 controllers using the same split-driver model as previous versions. This is a minor iteration on v4, fixing some leaked input_dev objects in error paths as well as fixing a merge/rebase error when CONFIG_PM ifdefs got removed in 7.0.

Vicki Pfau (3):
HID: nintendo: Add preliminary Switch 2 controller driver
HID: nintendo: Add rumble support for Switch 2 controllers
HID: nintendo: Add unified report format support

Quoted 11 lines
 MAINTAINERS                                   |    1 +
 drivers/hid/Kconfig                           |   19 +-
 drivers/hid/hid-ids.h                         |    4 +
 drivers/hid/hid-nintendo.c                    | 1504 ++++++++++++++++-
 drivers/hid/hid-nintendo.h                    |   72 +
 drivers/input/joystick/Kconfig                |   11 +
 drivers/input/joystick/Makefile               |    1 +
 drivers/input/joystick/nintendo-switch2-usb.c |  353 ++++
 8 files changed, 1951 insertions(+), 14 deletions(-)
 create mode 100644 drivers/hid/hid-nintendo.h
 create mode 100644 drivers/input/joystick/nintendo-switch2-usb.c

Dmitry, do you have any review comments to the changes under drivers/input/joystick, please?

Thanks,

Signature · 2 lines
Jiri Kosina
SUSE Labs