Surface Pro 6 camera not working — ipu_bridge creates no CSI-2 links #2101

Activity

1. mfa-nw commented on May 4, 2026

   

   mfa-nw

   on May 4, 2026

   More actions

   The i2c device table for the ipu3 cameras was removed from the upstream kernel 6.19 so libcamera cannot link to the cameras. I have reported the regression upstream and am waiting to see if a patch to add back the i2c device table gets accepted. I think this may be the cause of the issue you are experiencing.

   🚀React with 🚀2Reacted by artseluiko and Jan Esser
2. toor11 commented on May 11, 2026

   

   toor11

   on May 11, 2026

   More actions

   This has been fixed in PR #2123.

   Root cause: The upstream patch "media: i2c: dw9719: Remove unused i2c device id table" (kernel 6.19) incorrectly removed the i2c_device_id table from the dw9719 VCM driver. On Surface devices, the dw9719 VCM is instantiated by the IPU bridge via I2C, so it depends on I2C device matching — without the table, the driver has no i2c:dw9719 alias and never binds.

   Without the VCM bound, cio2_notifier_complete() doesn't fire, so the camera sensor entities never get V4L2 subdevice nodes, and libcamera reports no cameras.

   The fix adds back the i2c_device_id table, based on the upstream fix from Sakari Ailus: https://lore.kernel.org/all/[email protected]/
3. fbd-ss commented on May 25, 2026

   

   fbd-ss

   on May 25, 2026

   More actions

   I have the same issue with 6.19.8-surface-3 on linux mint 22.3, surface pro 6.
   It did work originally ( although only getting green output ), using the apt packages in the instructions.
   I tried building camera packages from source to no avail. The links were not being connected as described by @Mattttttttttttttttttttttttttttttt
4. toor11 commented on May 28, 2026

   

   toor11

   on May 28, 2026

   More actions

   Confirming the same regression — and that PR #2123 is the correct fix — on a Surface Book 2 (13"), kernel 6.19.8-arch1-3-surface, libcamera v0.7.1. Independently reproduced the full chain:

   • ipu3-cio2 0000:00:14.3: Connected 3 cameras; all three sensors bind to the cio2 notifier — ov5693 (INT33BE), ov8865 (INT347A), ov7251 (INT347E) all bound (ret 0).
   • dw9719 exposes only OF aliases (of:N*T*Cdongwoon,dw9719…) and no i2c:dw9719. The IPU-bridge-created VCM is a software_node (dw9719-0) with no of_node and no ACPI companion, so i2c_device_match() (OF → ACPI → i2c_match_id) has nothing to match. Forcing the bind returns -ENODEV.
   • With module ipu3_cio2 +p / file *v4l2-async* +p, after ov8865 binds it registers a lens sub-notifier and loops: need …I2C3.CAMR, trying dw9719-0 → direct match not found. Because v4l2-async completes hierarchically, cio2_notifier_complete() never fires.

   Net effect: the missing i2c_device_id blocks the rear VCM, which stalls the entire notifier — so all three sensors end up with 0 link to their CSI-2 receivers (front OV5693 and IR OV7251 too, even though they have no VCM), and cam --list is empty.

   - entity 41: ov5693 2-0036 (1 pad, 0 link, 0 routes)  Sensor
   - entity 37: ov8865 3-0010 (1 pad, 0 link, 0 routes)  Sensor
   - entity 39: ov7251 3-0060 (1 pad, 0 link, 0 routes)  Sensor
   # sensor->csi2 link count: 0
   

   Please add Surface Book 2 13" to the affected-device list. The i2c_device_id re-add in #2123 (mirroring Sakari Ailus's upstream fix) addresses this exactly.