[loisBreant]

The front camera of the IPU6 Surface devices (Pro 8, Pro 9, Go 4) still does not stream on the 7.2 kernels. The ov5693 driver starts the sensor, but the IPU6 CSI-2 receiver never gets a frame.

This sets MIPI_CTRL00 (0x4800) bit 5 (clock lane gate) before stream-on when the link is non-continuous. Because this kernel's ipu-bridge does not set clock-noncontinuous yet, it is also enabled for the OVTI5693 HID, which only appears on IPU6 Surfaces. Only that bit is changed, and a clock_ncont module parameter (-1 auto, 0, 1) can force it either way.

Relation to other work

• It uses the same approach as the upstream series "media: Enable the OV5693 front camera on IPU6 Surface devices" (Fernando Rimoli, in media next), which also gates bit 5 only. Once that series reaches a release, this commit can be dropped in favour of a backport, which moves the decision into ipu-bridge per IPU PCI ID. I can prepare that backport if you prefer it now.
• cameras: ov5693: make the IPU6 Surface front cameras stream (Pro 8/9, Go 4) linux-surface#2171 fixes the same problem for 6.19 by writing the full vendor value (0x2d). This PR targets 7.2, where nothing fixes it yet.

Testing: Surface Pro 9 (i7-1255U, IPU6 8086:465d), 7.2.11-surface-1, Ubuntu 24.04, libcamera 0.7.2 software ISP:

	Without	With
Frames received (2592×1944)	none	28.6 fps
CSI-2 errors while streaming	n/a	none
Chrome via v4l2loopback (MediaStreamTrack.stats)	n/a	28.6 fps delivered, 0 discarded

On this branch's ov5693.c (identical to 7.2.11-surface-1's), the result builds to exactly the same machine code as the module I have been running (same srcversion, identical disassembly).

The full userspace setup for the Surface Pro 9 (libcamera, v4l2loopback, browser support) is documented in https://github.com/loisBreant/surface-pro-9-camera.

Transparency: I developed this with the help of an AI assistant (Claude). It was built, run and verified on real hardware, and the kernel change was reviewed separately for safety before I used it.

[fildunsky]

Surface Pro 8 (IPU6 Tiger Lake, 8086:9a19) supports the bit 5 only approach.

I tested Fernando Rimoli's series at v5 on this machine. It gates only bit 5 of 0x4800 (through clock-noncontinuous from ipu-bridge), and the front camera streamed. That series is now in media next (0e12587945f6), so this PR matches what upstream will ship. Dropping it later in favour of the backport, as you suggest, makes sense.

One thing to watch when testing on the Pro 8: here, the first capture after a cold boot often succeeds even without any 0x4800 write. Only the captures after that fail. So capture at least twice before concluding the fix is present or absent.