Skip to contentSkip to conversation

HID: asus: fix touchpad multitouch initialization regression

PATCH In review Panz Dev 3 participants 4 messages Aug 18

Subsystems (2)
lore
Share & tools
lore

Panz Dev

author
#0 lore raw

PATCH 1/1 HID: asus: fix touchpad multitouch initialization regression

In Linux 7.1, commit 7253091766de ("HID: asus: do not abort probe when not necessary") and commit 0919db9f3583 ("HID: asus: always fully initialize devices") introduced regressions in the probe sequence for ASUS I2C/HID touchpads (such as on the ASUS E200HA):

1. asus_start_multitouch() was placed inside the claimed input check block:

     if (drvdata->input && (hdev->claimed & HID_CLAIMED_INPUT))

On ASUS touchpads, (hdev->claimed & HID_CLAIMED_INPUT) evaluates to false during asus_probe(), skipping asus_start_multitouch().

2. The asus_report_id_init loop in asus_probe() was executed unconditionally for all devices, which sends keyboard initialization feature requests (asus_kbd_init) to touchpad endpoints, corrupting touchpad probe state.

This patch fixes both issues by:
- Skipping keyboard report initialization for touchpad devices (!drvdata->tp).
- Moving asus_start_multitouch() outside the claimed input check block so multitouch initialization is always executed for touchpads.

Tested on ASUS E200HA (where touchpad functionality is fully restored) and ASUS VivoBook Flip 14 TP401MA (confirming zero regressions).

Fixes 7253091766de ("HID: asus: do not abort probe when not necessary")
Fixes 0919db9f3583 ("HID: asus: always fully initialize devices")
---
 drivers/hid/hid-asus.c | 22 ++++++++++++----------
 1 file changed, 12 insertions(+), 10 deletions(-)
drivers/hid/hid-asus.c +13 -11
index 3f5e96900b67..7f19ca1e5a1b 100644
@@ -1294,12 +1294,14 @@ static int asus_probe(struct hid_device *hdev, const struct hid_device_id *id)
 		return ret;
 	}
 
-	for (int r = 0; r < ARRAY_SIZE(asus_report_id_init); r++) {
-		if (asus_has_report_id(hdev, asus_report_id_init[r])) {
-			ret = asus_kbd_init(hdev, asus_report_id_init[r]);
-			if (ret < 0)
-				hid_warn(hdev, "Failed to initialize 0x%x: %d.\n",
-					 asus_report_id_init[r], ret);
+	if (!drvdata->tp) {
+		for (int r = 0; r < ARRAY_SIZE(asus_report_id_init); r++) {
+			if (asus_has_report_id(hdev, asus_report_id_init[r])) {
+				ret = asus_kbd_init(hdev, asus_report_id_init[r]);
+				if (ret < 0)
+					hid_warn(hdev, "Failed to initialize 0x%x: %d.\n",
+						 asus_report_id_init[r], ret);
+			}
 		}
 	}
 
@@ -1327,12 +1329,12 @@ static int asus_probe(struct hid_device *hdev, const struct hid_device_id *id)
 			drvdata->input->name = "Asus TouchPad";
 		else
 			drvdata->input->name = "Asus Keyboard";
+	}
 
-		if (drvdata->tp) {
-			ret = asus_start_multitouch(hdev);
-			if (ret)
-				goto err_stop_hw;
-		}
+	if (drvdata->tp) {
+		ret = asus_start_multitouch(hdev);
+		if (ret)
+			goto err_stop_hw;
 	}
 
 	return 0;
Signature · 1 line
2.47.0
#1 lore raw
Quoted 14 lines · Panz Dev

In Linux 7.1, commit 7253091766de ("HID: asus: do not abort probe when not necessary") and commit 0919db9f3583 ("HID: asus: always fully initialize devices") introduced regressions in the probe sequence for ASUS I2C/HID touchpads (such as on the ASUS E200HA):

1. asus_start_multitouch() was placed inside the claimed input check block:

     if (drvdata->input && (hdev->claimed & HID_CLAIMED_INPUT))

On ASUS touchpads, (hdev->claimed & HID_CLAIMED_INPUT) evaluates to false during asus_probe(), skipping asus_start_multitouch().

2. The asus_report_id_init loop in asus_probe() was executed unconditionally for all devices, which sends keyboard initialization feature requests (asus_kbd_init) to touchpad endpoints, corrupting touchpad probe state.

Quoted 7 lines

This patch fixes both issues by:
- Skipping keyboard report initialization for touchpad devices (!drvdata->tp).
- Moving asus_start_multitouch() outside the claimed input check block so multitouch initialization is always executed for touchpads.

Tested on ASUS E200HA (where touchpad functionality is fully restored) and ASUS VivoBook Flip 14 TP401MA (confirming zero regressions).

Hi,

thanks for this! I forgot to exclude touchpads from that init sequence.

Quoted 39 lines
Fixes 7253091766de ("HID: asus: do not abort probe when not necessary")
Fixes 0919db9f3583 ("HID: asus: always fully initialize devices")
---
 drivers/hid/hid-asus.c | 22 ++++++++++++----------
 1 file changed, 12 insertions(+), 10 deletions(-)
drivers/hid/hid-asus.c +9 -6
index 3f5e96900b67..7f19ca1e5a1b 100644
@@ -1294,12 +1294,14 @@ static int asus_probe(struct hid_device *hdev, const struct hid_device_id *id)
 		return ret;
 	}
 
-	for (int r = 0; r < ARRAY_SIZE(asus_report_id_init); r++) {
-		if (asus_has_report_id(hdev, asus_report_id_init[r])) {
-			ret = asus_kbd_init(hdev, asus_report_id_init[r]);
-			if (ret < 0)
-				hid_warn(hdev, "Failed to initialize 0x%x: %d.\n",
-					 asus_report_id_init[r], ret);
+	if (!drvdata->tp) {
+		for (int r = 0; r < ARRAY_SIZE(asus_report_id_init); r++) {
+			if (asus_has_report_id(hdev, asus_report_id_init[r])) {
+				ret = asus_kbd_init(hdev, asus_report_id_init[r]);
+				if (ret < 0)
+					hid_warn(hdev, "Failed to initialize 0x%x: %d.\n",
+						 asus_report_id_init[r], ret);
+			}
 		}
 	}
 
@@ -1327,12 +1329,12 @@ static int asus_probe(struct hid_device *hdev, const struct hid_device_id *id)
 			drvdata->input->name = "Asus TouchPad";
 		else
 			drvdata->input->name = "Asus Keyboard";
+	}
 

drivers/hid/hid-asus.c static int asus_probe(struct hid_device *hdev, const struct hid_device_id *id) · line 1338

Quoted 10 lines
drivers/hid/hid-asus.c +4 -5
-		if (drvdata->tp) {
-			ret = asus_start_multitouch(hdev);
-			if (ret)
-				goto err_stop_hw;
-		}
+	if (drvdata->tp) {
+		ret = asus_start_multitouch(hdev);
+		if (ret)
+			goto err_stop_hw;
 	}

I think this will re-introduce the security vulnerability fixed in the commit that led to me creating the "HID: asus: do not abort probe when unnecessary"...

Furthermore if that drvdata->input && (hdev->claumed & HID_CLAIMED_INPUT) doesn't evaluate to true it would be useful to know what part is false and that code that prevents null/invalid dereference is wrong, and since I tried touching it the least possible it means there is also the need to change the && back into ||, but that would undo the "keep the proper name of the device" for keyboards...

With these in mind would you want to try fixing them all at once?

Quoted 1 line
drivers/hid/hid-asus.c +0 -0
 	return 0;
#2 lore raw
In reply to Panz Dev
Thank you for your contribution! Sashiko AI review found 4 potential issue(s) to consider:

Thank you for your contribution! Sashiko AI review found 4 potential issue(s) to consider:

New issues:
- [High] Unconditionally calling asus_start_multitouch() outside the HID_CLAIMED_INPUT check introduces a Use-After-Free if input registration fails.

Pre-existing issues:
- [High] Incorrect cleanup order in asus_remove() creates a Use-After-Free race condition by cancelling the workqueue before disabling hardware events.
- [High] Missing hid_is_usb() check in asus_kbd_register_leds() allows an invalid cast to usb_interface, causing out-of-bounds reads or crashes.
- [Medium] Modifying input->name after device registration creates a data race and violates input subsystem API requirements. --

--- Patch [1]: [PATCH] HID: asus: fix touchpad multitouch initialization regression --- commit 53becd94719f665ca15fc8c3f4b7876a402da45f
Author: Panz Dev <[email protected]>

HID: asus: fix touchpad multitouch initialization regression

This patch fixes probe regressions for ASUS I2C/HID touchpads by skipping keyboard report initialization for touchpads and moving asus_start_multitouch() outside the claimed input check block.

Quoted 4 lines
drivers/hid/hid-asus.c +0 -0
index 3f5e96900b67a..72b93bdeb4e8f 100644

[ ... ]

drivers/hid/hid-asus.c static int asus_probe(struct hid_device *hdev, const struct hid_device_id *id) · line 1331

Quoted 4 lines
drivers/hid/hid-asus.c +0 -0
@@ -1327,12 +1329,12 @@ static int asus_probe(struct hid_device *hdev, const struct hid_device_id *id)
 			drvdata->input->name = "Asus TouchPad";
 		else
 			drvdata->input->name = "Asus Keyboard";

[Severity: Medium]
This is a pre-existing issue, but can modifying drvdata->input->name here after device registration (which occurs earlier during hid_hw_start) create a data race?

If a userspace process like udev reads the name attribute concurrently, might it read the default string or race with the pointer reassignment?

drivers/hid/hid-asus.c static int asus_probe(struct hid_device *hdev, const struct hid_device_id *id) · line 1338

Quoted 12 lines
drivers/hid/hid-asus.c +5 -5
+	}
 
-		if (drvdata->tp) {
-			ret = asus_start_multitouch(hdev);
-			if (ret)
-				goto err_stop_hw;
-		}
+	if (drvdata->tp) {
+		ret = asus_start_multitouch(hdev);
+		if (ret)
+			goto err_stop_hw;
 	}

[Severity: High]
Does moving asus_start_multitouch() outside the HID_CLAIMED_INPUT check introduce a use-after-free regression if input registration fails?

If input_register_device() fails during hid_hw_start(), hidinput_connect() frees the allocated input_dev and returns an error without setting HID_CLAIMED_INPUT.

Since the HID_CLAIMED_INPUT safety guard is now removed, the driver will force the touchpad hardware to start sending reports. When the hardware interrupt arrives, asus_raw_event() intercepts it and routes to asus_report_input():

drivers/hid/hid-asus.c:asus_raw_event() {
    ...
	if (drvdata->tp && data[0] == INPUT_REPORT_ID)
		return asus_report_input(drvdata, data, size);
    ...
}

This blindly dereferences the freed drvdata->input pointer, which was assigned earlier in asus_input_configured():

drivers/hid/hid-asus.c:asus_report_input() {
    ...
		input_mt_slot(drvdat->input, i);
		input_mt_report_slot_state(drvdat->input, toolType, down);
    ...
}

Is it possible for this to cause a use-after-free, for instance if triggered via uhid device spoofing or resource allocation failures during probe?

drivers/hid/hid-asus.c static int asus_probe(struct hid_device *hdev, const struct hid_device_id *id) · line 1340

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

[Severity: High]
This isn't a bug introduced by this patch, but does the cleanup order in asus_remove() create a use-after-free race condition?

In asus_remove(), the driver calls:

drivers/hid/hid-asus.c:asus_remove() {
    ...
	if (drvdata->quirks & QUIRK_HID_FN_LOCK)
		cancel_work_sync(&drvdata->fn_lock_sync_work);

	hid_hw_stop(hdev);
    ...
}

If the device sends a KEY_FN_ESC report via hardware interrupt immediately after cancel_work_sync() returns but before hid_hw_stop() completes, might asus_event() re-queue the work via schedule_work()? Once asus_remove() finishes, the devm-managed drvdata is freed, which could leave the work item scheduled and lead to a use-after-free when the worker executes.

[Severity: High]
This is a pre-existing issue, but could the lack of a hid_is_usb() check in asus_kbd_register_leds() allow an invalid cast?

When handling the QUIRK_ROG_ALLY_XPAD quirk:

drivers/hid/hid-asus.c:asus_kbd_register_leds() {
    ...
	if (drvdata->quirks & QUIRK_ROG_ALLY_XPAD) {
		intf = to_usb_interface(hdev->dev.parent);
		udev = interface_to_usbdev(intf);
    ...
}

If the device is connected via another transport (like Bluetooth, I2C, or spoofed via uhid), the parent device is not a USB interface. Would calling interface_to_usbdev() on this invalid pointer lead to out-of-bounds memory access or a crash?

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

Hi Denis,

Thanks for the fast reply!
Good catch on the UAF risk. It turns out simply adding if (!drvdata->tp) to skip the keyboard init loop was actually all that was needed to fix the touchpad probe on the E200HA.

That way, asus_start_multitouch() stays safely inside the claimed input check as before without touching any of the security logic, and the && condition doesn't need to be changed back to ||.

I've tested this on hardware (ASUS E200HA and VivoBook Flip 14 TP401MA) and just sent v2.

Sincerely,
PANZ

Quoted 131 lines · On Tue, Aug 18, 2026 at 11:15 AM <[email protected]> wrote:

Thank you for your contribution! Sashiko AI review found 4 potential issue(s) to consider:

New issues:
- [High] Unconditionally calling asus_start_multitouch() outside the HID_CLAIMED_INPUT check introduces a Use-After-Free if input registration fails.

Pre-existing issues:
- [High] Incorrect cleanup order in asus_remove() creates a Use-After-Free race condition by cancelling the workqueue before disabling hardware events.
- [High] Missing hid_is_usb() check in asus_kbd_register_leds() allows an invalid cast to usb_interface, causing out-of-bounds reads or crashes.
- [Medium] Modifying input->name after device registration creates a data race and violates input subsystem API requirements. --

--- Patch [1]: [PATCH] HID: asus: fix touchpad multitouch initialization regression --- commit 53becd94719f665ca15fc8c3f4b7876a402da45f
Author: Panz Dev <[email protected]>

HID: asus: fix touchpad multitouch initialization regression

This patch fixes probe regressions for ASUS I2C/HID touchpads by skipping keyboard report initialization for touchpads and moving asus_start_multitouch() outside the claimed input check block.

Quoted 4 lines
drivers/hid/hid-asus.c +0 -0
index 3f5e96900b67a..72b93bdeb4e8f 100644

[ ... ]

drivers/hid/hid-asus.c static int asus_probe(struct hid_device *hdev, const struct hid_device_id *id) · line 1331

Quoted 4 lines
drivers/hid/hid-asus.c +0 -0
@@ -1327,12 +1329,12 @@ static int asus_probe(struct hid_device *hdev, const struct hid_device_id *id)
                      drvdata->input->name = "Asus TouchPad";
              else
                      drvdata->input->name = "Asus Keyboard";

[Severity: Medium]
This is a pre-existing issue, but can modifying drvdata->input->name here after device registration (which occurs earlier during hid_hw_start) create a data race?

If a userspace process like udev reads the name attribute concurrently, might it read the default string or race with the pointer reassignment?

drivers/hid/hid-asus.c static int asus_probe(struct hid_device *hdev, const struct hid_device_id *id) · line 1338

Quoted 12 lines
drivers/hid/hid-asus.c +5 -5
+     }
 
-             if (drvdata->tp) {
-                     ret = asus_start_multitouch(hdev);
-                     if (ret)
-                             goto err_stop_hw;
-             }
+     if (drvdata->tp) {
+             ret = asus_start_multitouch(hdev);
+             if (ret)
+                     goto err_stop_hw;
      }

[Severity: High]
Does moving asus_start_multitouch() outside the HID_CLAIMED_INPUT check introduce a use-after-free regression if input registration fails?

If input_register_device() fails during hid_hw_start(), hidinput_connect() frees the allocated input_dev and returns an error without setting HID_CLAIMED_INPUT.

Since the HID_CLAIMED_INPUT safety guard is now removed, the driver will force the touchpad hardware to start sending reports. When the hardware interrupt arrives, asus_raw_event() intercepts it and routes to asus_report_input():

drivers/hid/hid-asus.c:asus_raw_event() {
    ...
        if (drvdata->tp && data[0] == INPUT_REPORT_ID)
                return asus_report_input(drvdata, data, size);
    ...
}

This blindly dereferences the freed drvdata->input pointer, which was assigned earlier in asus_input_configured():

drivers/hid/hid-asus.c:asus_report_input() {
    ...
                input_mt_slot(drvdat->input, i);
                input_mt_report_slot_state(drvdat->input, toolType, down);
    ...
}

Is it possible for this to cause a use-after-free, for instance if triggered via uhid device spoofing or resource allocation failures during probe?

drivers/hid/hid-asus.c static int asus_probe(struct hid_device *hdev, const struct hid_device_id *id) · line 1340

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

[Severity: High]
This isn't a bug introduced by this patch, but does the cleanup order in asus_remove() create a use-after-free race condition?

In asus_remove(), the driver calls:

drivers/hid/hid-asus.c:asus_remove() {
    ...
        if (drvdata->quirks & QUIRK_HID_FN_LOCK)
                cancel_work_sync(&drvdata->fn_lock_sync_work);

        hid_hw_stop(hdev);
    ...
}

If the device sends a KEY_FN_ESC report via hardware interrupt immediately after cancel_work_sync() returns but before hid_hw_stop() completes, might asus_event() re-queue the work via schedule_work()? Once asus_remove() finishes, the devm-managed drvdata is freed, which could leave the work item scheduled and lead to a use-after-free when the worker executes.

[Severity: High]
This is a pre-existing issue, but could the lack of a hid_is_usb() check in asus_kbd_register_leds() allow an invalid cast?

When handling the QUIRK_ROG_ALLY_XPAD quirk:

drivers/hid/hid-asus.c:asus_kbd_register_leds() {
    ...
        if (drvdata->quirks & QUIRK_ROG_ALLY_XPAD) {
                intf = to_usb_interface(hdev->dev.parent);
                udev = interface_to_usbdev(intf);
    ...
}

If the device is connected via another transport (like Bluetooth, I2C, or spoofed via uhid), the parent device is not a USB interface. Would calling interface_to_usbdev() on this invalid pointer lead to out-of-bounds memory access or a crash?

--
Sashiko AI review · https://sashiko.dev/#/patchset/[email protected]?part=1