Skip to contentSkip to conversation

media: ipu-bridge: the Surface Pro 11 rear sensor is mounted upside down

PATCH In review Sergey Lebedev 3 participants 6 messages Sep 13 – Sep 17 Media (V4L/DVB)

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 media: ipu-bridge: the Surface Pro 11 rear sensor is mounted upside down

The Microsoft Surface Pro for Business 11th Edition with Intel mounts its OV13858 rear sensor rotated 180 degrees and declares it in neither the SSDB nor a matching _PLD, so the rotation reads as 0 and every frame arrives inverted with nothing to say so.

Add it to upside_down_sensor_dmi_ids, the table that exists for this class of firmware. Matched on DMI_PRODUCT_NAME as the rest of that table is; the entry only takes effect when the ACPI HID matches too, so a machine of the same model that shipped a different rear sensor is left alone.

Measured on the machine. Before:

camera_sensor_rotation min=0 max=0 value=0
libcamera: Property: Rotation = 0

After:

camera_sensor_rotation min=180 max=180 value=180
libcamera: Property: Rotation = 180

and the front sensor is untouched at 0.

This makes the kernel report the mounting truthfully, and that is all it does. The pixels are unchanged: capturing one static scene through libcamera with the property at 0 and at 180 and correlating the vertical brightness profile gives +0.995 the same way up and -0.781 flipped, so libcamera delivers the same buffer either way. The sensor has no flip controls, so nothing is corrected in hardware either.

An application that reads the property can now rotate; one that ignores it still shows the image inverted. That is a userspace matter and is not addressed here.

---
Written 2026-08-31 against a variant of upside_down_sensor_dmi_ids that takes a struct with a match type and an id list. No such variant exists in media/next, so the patch could never have applied and was never sent; rewritten today in the form the table actually uses, and measured.

Applies to media/next on its own and on top of

https://lore.kernel.org/linux-media/[email protected]/

which adds the OVTID858 bridge entry the same machine needs.

A second machine reporting the same DMI product name independently reports the same inverted image, in three different applications:

https://lore.kernel.org/linux-media/[email protected]/
https://lore.kernel.org/linux-media/[email protected]/

I have asked its owner to confirm the string and to say whether any of those three applications turns the picture the right way up once the property is set. It would surprise me if they did.

 drivers/media/pci/intel/ipu-bridge.c | 8 ++++++++
 1 file changed, 8 insertions(+)
drivers/media/pci/intel/ipu-bridge.c +8 -0
index 7e65491faf..a5de450284 100644
@@ -192,6 +192,14 @@ static const struct dmi_system_id upside_down_sensor_dmi_ids[] = {
 		},
 		.driver_data = "OVTI02E1",
 	},
+	{
+		.matches = {
+			DMI_EXACT_MATCH(DMI_SYS_VENDOR, "Microsoft Corporation"),
+			DMI_EXACT_MATCH(DMI_PRODUCT_NAME,
+					"Surface Pro for Business 11th Edition with Intel"),
+		},
+		.driver_data = "OVTID858",
+	},
 	{} /* Terminating entry */
 };
 
#2 lore raw

Re: Test [PATCH] media: ipu-bridge: the Surface Pro 11 rear sensor is mounted upside down

Thank you. I will carry your tag to the next version.

I want to be sure I understand your result. Please answer yes or no:

1. Is the image now correct on screen, not upside down?
2. Did you check this in Gnome Snapshot?
3. Did you check this in qcam?
4. Did you check this in Firefox?
5. Was it correct in the live preview, the moving picture?
6. Was it correct in a photo file you saved?

Questions 5 and 6 matter most. A program can turn the picture on the screen and still save the file the old way. I only tested saved files here.

If the answer to 1 is no, that is fine. It does not make the patch wrong. The patch makes the kernel report the correct rotation. That is all it claims to do.

I ask because I tested it here. I captured the same scene twice: once with the patch, once without it. Both images are the same way up. The patch did not turn the picture. It only changed the number libcamera reports, from 0 to 180.

So if your image is now correct, your program is turning it, using that number. I could not see that here, and I would like to repeat what you did. I have the same tablet. Please send:

uname -r
pacman -Q libcamera (or your distribution's command)

and the version of whichever program showed the correct image.

If I can match your setup, I can find which program does the rotation. Then it is a question for that program, not for the kernel.

Sergey

#3 lore raw

More details for Test [PATCH] media: ipu-bridge: the Surface Pro 11 rear sensor is mounted upside down

The image is correctly flipped in Gnome and Firefox (the two i originally tested on) but in my original test i forgot qcam in which the image was not flipped ¿maybe it only flips Snapshot and Firefox because they use pipewire?

The preview in the three apps was fluent and the image saved corresponded to the image shown (rigth orientation for Snapshot, wrong for qcam)

#4 lore raw

Re: More details for Test [PATCH] media: ipu-bridge: the Surface Pro 11 rear sensor is mounted upside down

Thank you. That is the answer, and it settles the question.

You have given me three things today that I could not have got on my own: a test on a second machine, the cause of the colour error, and now this. I am grateful for all three, and the last one I had asked for because I could not test it here.

Your result and my measurement agree, and each explains the other. libcamera does not rotate the image, and your qcam result is the evidence of that on 0.7.2. I measured the same here on 0.5.0: with the patch loaded I asked for rot0, and it answered "Camera configuration adjusted" and returned the same frame, the same way up. The sensor has no flip control, so libcamera tells the application it cannot do it and hands the picture over unturned.

So qcam is right to show it upside down. qcam uses libcamera directly. And your idea about pipewire is very likely the answer for the other two:
Snapshot and Firefox receive the frame through pipewire, which reads the rotation property and applies it.

That is a better outcome for the patch than I expected. The kernel now reports the mounting truthfully, and the two programs most people use act on it:

  Gnome Snapshot    correct, in the preview and in the saved file
  Firefox           correct
  qcam              not flipped, which is the expected behaviour

For the patch itself, no change is needed. Its commit message already says that an application reading the property can rotate and one ignoring it will not, and your three programs have now confirmed both halves of that on a second machine. Your Tested-by applies to it unchanged.

A v2 would only make that paragraph concrete by naming the programs, and I do not think that earns a respin on its own. If a version is needed for any other reason, it goes in then and I carry your tag into it by hand. So there is nothing outstanding from me here.

Sergey

#5 lore raw
In reply to Sergey Lebedev

Hi Sergey,

Quoted 29 lines · Sergey Lebedev

The Microsoft Surface Pro for Business 11th Edition with Intel mounts its OV13858 rear sensor rotated 180 degrees and declares it in neither the SSDB nor a matching _PLD, so the rotation reads as 0 and every frame arrives inverted with nothing to say so.

Add it to upside_down_sensor_dmi_ids, the table that exists for this class of firmware. Matched on DMI_PRODUCT_NAME as the rest of that table is; the entry only takes effect when the ACPI HID matches too, so a machine of the same model that shipped a different rear sensor is left alone.

Measured on the machine. Before:

camera_sensor_rotation min=0 max=0 value=0
libcamera: Property: Rotation = 0

After:

camera_sensor_rotation min=180 max=180 value=180
libcamera: Property: Rotation = 180

and the front sensor is untouched at 0.

This makes the kernel report the mounting truthfully, and that is all it does. The pixels are unchanged: capturing one static scene through libcamera with the property at 0 and at 180 and correlating the vertical brightness profile gives +0.995 the same way up and -0.781 flipped, so libcamera delivers the same buffer either way. The sensor has no flip controls, so nothing is corrected in hardware either.

Quoted 3 lines

An application that reads the property can now rotate; one that ignores it still shows the image inverted. That is a userspace matter and is not addressed here.

This commit message of a trivial patch is exceedingly long, please shorten it. See examples of similar patches, for instance.

Quoted 32 lines

---
Written 2026-08-31 against a variant of upside_down_sensor_dmi_ids that takes a struct with a match type and an id list. No such variant exists in media/next, so the patch could never have applied and was never sent; rewritten today in the form the table actually uses, and measured.

Applies to media/next on its own and on top of

https://lore.kernel.org/linux-media/[email protected]/

which adds the OVTID858 bridge entry the same machine needs.

A second machine reporting the same DMI product name independently reports the same inverted image, in three different applications:

https://lore.kernel.org/linux-media/[email protected]/
https://lore.kernel.org/linux-media/[email protected]/

I have asked its owner to confirm the string and to say whether any of those three applications turns the picture the right way up once the property is set. It would surprise me if they did.

 drivers/media/pci/intel/ipu-bridge.c | 8 ++++++++
 1 file changed, 8 insertions(+)
drivers/media/pci/intel/ipu-bridge.c +0 -0
index 7e65491faf..a5de450284 100644

drivers/media/pci/intel/ipu-bridge.c static const struct dmi_system_id upside_down_sensor_dmi_ids[] = { · line 197

Quoted 6 lines
drivers/media/pci/intel/ipu-bridge.c +3 -0
@@ -192,6 +192,14 @@ static const struct dmi_system_id upside_down_sensor_dmi_ids[] = {
 		},
 		.driver_data = "OVTI02E1",
 	},
+	{
+		.matches = {
+			DMI_EXACT_MATCH(DMI_SYS_VENDOR, "Microsoft Corporation"),

Let's aim to keep these alphabetically ordered.

Quoted 9 lines
drivers/media/pci/intel/ipu-bridge.c +5 -0
+			DMI_EXACT_MATCH(DMI_PRODUCT_NAME,
+					"Surface Pro for Business 11th Edition with Intel"),
+		},
+		.driver_data = "OVTID858",
+	},
 	{} /* Terminating entry */
 };
 
Signature · 3 lines
Regards,

Sakari Ailus