Re: [PATCH v2 0/8] Add AMD ISP4 driver

From: Du, Bin
Date: Fri Aug 08 2025 - 05:12:01 EST

Next message: Valentin Schneider: "Re: [PATCH v3 3/5] sched/fair: Switch to task based throttle model"
Previous message: Bryan O'Donoghue: "Re: [PATCH v9 5/8] media: venus: core: Sort dt_match alphabetically."
Next in thread: Sultan Alsawaf: "Re: [PATCH v2 0/8] Add AMD ISP4 driver"
Messages sorted by: [ date ] [ thread ] [ subject ] [ author ]

Thanks Sultan.

On 8/4/2025 12:25 PM, Sultan Alsawaf wrote:

On Mon, Aug 04, 2025 at 11:32:11AM +0800, Du, Bin wrote:

On 7/31/2025 6:04 PM, Du, Bin wrote:

Thanks Sultan for your test

On 7/31/2025 8:30 AM, Sultan Alsawaf wrote:

On Wed, Jul 30, 2025 at 05:53:58PM +0800, Du, Bin wrote:

On 7/30/2025 1:38 PM, Sultan Alsawaf wrote:

On Tue, Jul 29, 2025 at 06:13:50PM +0800, Du, Bin wrote:

On 7/29/2025 3:45 PM, Sultan Alsawaf wrote:

On Tue, Jul 29, 2025 at 12:42:16AM -0700, Sultan Alsawaf wrote:

On Tue, Jul 29, 2025 at 11:32:23AM +0800, Du, Bin wrote:

Thanks Sultan, please see my comments

On 7/27/2025 6:31 AM, Sultan Alsawaf wrote:

On Fri, Jul 25, 2025 at 06:22:03PM +0800, Du, Bin wrote:

I have the Ryzen AI MAX+ 395 SKU of the HP ZBook Ultra G1a 14.

I cannot for the life of me get
the webcam working under Linux.
The webcam works
under Windows so it's not a hardware issue.

With this patchset and all of
the patches you link here
applied to 6.15, I get
the following errors:
       [   11.970038]
amd_isp_i2c_designware
amd_isp_i2c_designware: Unknown
Synopsys component type:
0xffffffff
       [   11.973162]
amd_isp_i2c_designware
amd_isp_i2c_designware: error
-19: i2c_dw_probe failed

With the old ispkernel code from
February [1] applied on 6.15,
the webcam
indicator LED lights up but
there's no image. I see these
messages at boot:
       [    9.449005]
amd_isp_capture
amd_isp_capture.1.auto: amdgpu:
AMD ISP v4l2 device registered
       [    9.489005]
amd_isp_i2c_designware
amd_isp_i2c_designware.2.auto:
The OV05 sensor device is added
to the ISP I2C bus
       [    9.529012]
amd_isp_i2c_designware
amd_isp_i2c_designware.2.auto:
timeout while trying to abort
current transfer
       [    9.554046]
amd_isp_i2c_designware
amd_isp_i2c_designware.2.auto:
timeout in disabling adapter
       [    9.554174]
amd_isp_i2c_designware
amd_isp_i2c_designware.2.auto:
timeout while trying to abort
current transfer
       [    9.580022]
amd_isp_i2c_designware
amd_isp_i2c_designware.2.auto:
timeout in disabling adapter

And then the kernel crashes due
to the same use-after-free
issues I pointed out
in my other email [2].

Any idea what's going on?

[1]
https://github.com/amd/Linux_ISP_Kernel/commit/
c6d42584fbd0aa42cc91ecf16dc5c4f3dfea0bb4
[2] https://lore.kernel.org/r/aIEiJL83pOYO8lUJ@sultan-box

Hi Sultan,

[1] is for kernel 6.8, believe it
can't be applied to 6.15. We didn't
verify
on 6.15 but we are really glad to
help, would you please provide some
info,
1. Suppose you are using Ubuntu, right? What's the version?
2. 6.15, do you mean
https://github.com/torvalds/linux/tree/
v6.15 ?

After your confirmation, we'll see
what we can do to enable your camera
quickly and easily

Regards,
Bin

Thank you, Bin!

1. I'm using Arch Linux with the ISP4-patched libcamera [1].
2. Yes, here is my kernel source [2].

I have some more findings:

Currently, the first blocking issue is
that the I2C adapter fails to
initialize.
This is because the ISP tile isn't powered on.

I noticed that in the old version of
amd_isp_i2c_designware [3], there were
calls to isp_power_set(), which is
available in the old ISP4 sources [4].
Without isp_power_set(), the I2C adapter
always fails to initialize for me.

How is the ISP tile supposed to get
powered on in the current ISP4 code?

You are correct, yes, i believe the I2C
adapter failure is caused by ISP not
being powered up. Currently in latest code,
isp_power_set is no longer
available, instead, we implemented genPD for ISP in amdgpu
https://lore.kernel.org/all/20250618221923.3944751-1-
pratap.nirujogi@xxxxxxx/
Both amd_isp_i2c and amd_isp_capture are in
the power domain and use the
standard runtime PM API to do the power control

Thanks for that link, I found it along with
another patch on the list to make
the fwnode work ("drm/amd/amdgpu: Initialize
swnode for ISP MFD device").

Also, I noticed that the driver init
ordering matters between all of the
drivers
needed for the ISP4 camera. In
particular, amd_isp_i2c_designware and
amd_isp4
must be initialized before amd_capture,
otherwise amd_capture will fail to find
the fwnode properties for the OV05C10
device attached to the I2C bus.

But there is no driver init ordering
enforced, which also caused some issues
for
me until I figured it out. Maybe probe
deferral (-EPROBE_DEFER) should be used
to ensure each driver waits for its dependencies to init first?

amd_isp_capture only has dependency on
amd_isp4 which is the ACPI platform
driver, it is init before amd_isp_catpure.
Do you see in your side the amd_capture
probe failure caused by failing to
read fwnode properties? If that's the case
please help to check if amd_isp4
is loaded successfully

I got much further now: there aren't any driver
initialization errors, but when
I open the camera, there's no image. The camera
LED turns on so it's active.

And then shortly afterwards, amdgpu dies and the
entire system freezes.

I've attached my full dmesg, please let me know
what you think. Thanks!

I almost forgot, here is my current kernel tree:
https://github.com/kerneltoast/kernel_x86_laptop/tree/v6.16-
sultan-isp4

Sultan

Thanks Sultan, yes, seems much close to the final
success. Will have some
internal discussion.

I got the webcam working. The same bug happened when I tried
Ubuntu's linux-oem
kernel, which made me think that the issue was firmware.

And indeed, the culprit was a firmware update from February. I bisected
linux-firmware and found the commit which broke the webcam for me:

    commit 1cc8c1bfa11251ce8bfcc97d1f15e312f7fe4df0 (HEAD)
    Author: Pratap Nirujogi <pratap.nirujogi@xxxxxxx>
    Date:   Wed Feb 19 12:16:51 2025 -0500

        amdgpu: Update ISP FW for isp v4.1.1

        From internal git commit:
        5058202443e08a673b6772ea6339efb50853be28

        Signed-off-by: Pratap Nirujogi <pratap.nirujogi@xxxxxxx>

     amdgpu/isp_4_1_1.bin | Bin 4543184 -> 6083536 bytes
     1 file changed, 0 insertions(+), 0 deletions(-)

Downgrading firmware to before that commit fixes the webcam.
Any idea why?

Thanks,
Sultan

So, can i say the working firmware binary is this one?

Commit 8f070131
amdgpu: Update ISP FW for isp v4.1.1

From internal git commit:
39b007366cc76ef8c65e3bc6220ccb213f4861fb

Signed-off-by: Pratap Nirujogi <pratap.nirujogi@xxxxxxx>

Correct.

There are too many changes between them, so i can't tell exactly which
change caused this. So, from my side
1. Will try these two firmware to see if we have the same issue.
2. It has been quite a long time since last release, will see if need to
release a latest one.

Thanks. It was a quick bisect for me, so I'm happy to help test if a
bisect
between those two internal git commits is needed.

Really appreciate your test.

In case it makes a difference, I have the laptop with the 2.8K OLED
display. I'm
aware there is one other display variant on other SKUs, which is a
WUXGA IPS.

Good to know, I believe it won't make any difference for ISP

Also, with that old firmware, my camera only works with the old isp4
driver from
that Linux_ISP_Kernel repo (which is the same isp4 driver used in
Ubuntu's
linux-oem kernel). Does the new isp4 driver you've submitted here
require newer
firmware than the old driver located in Linux_ISP_Kernel?

Sultan

We had a try, yes, both of the old FW can't work on the new ISP4 driver,
as you know, for the last months, we did lots of driver modifications
for upstream and cause it incompatible with old FW.
Now, under internal discussion to upstream a new FW to support the new
ISP driver

Regards,
Bin

Hi Sultan,

This is the conclusion of your test,
Driver: https://github.com/amd/Linux_ISP_Kernel/commit/c6d42584fbd0aa42cc91ecf16dc5c4f3dfea0bb4
[1] It works on FW 8f070131(ext):39b00736(int)
[2] It can't work on FW 1cc8c1bf(ext):50582024(int)

Correct.

Would you please help to check if CONFIG_VIDEO_OV05C is defined not in the
.config file when building the kernel? Our assumption is to make [1] work,
CONFIG_VIDEO_OV05C shouldn't be defined. to make [2] work,
CONFIG_VIDEO_OV05C should be defined.

Yes, it is enabled and I have only tested with it enabled:

$ rg CONFIG_VIDEO_OV05C linux-oem-6.14/.config
CONFIG_VIDEO_OV05C=m

That's the Ubuntu linux-oem-6.14 kernel. You can get the full source and .config
I tested by running the following commands:

git clone https://git.launchpad.net/ubuntu/+source/linux-oem-6.14 -b applied/6.14.0-1006.6
cd linux-oem-6.14
python debian/scripts/misc/annotations -e --arch amd64 > .config
make olddefconfig

Let me know if that works.

Sultan

Thanks Sultan for the details, yes, we can reproduce the same issue on old isp driver 4.0 release on FW 1cc8c1bf(ext):50582024(int), after debug, the cause is
- ov05c sensor device is added by amd i2c driver
- When ov05c sensor driver probes, it will try to get gpio description but it will fail because the amd-pinctl driver which creates the gpio resource hasn't been loaded yet.
- the ov05c sensor driver probe failure will finally make sensor not able to work when start streaming

Add following patch is supposed to fix this issue to make it work on FW 1cc8c1bf(ext):50582024(int) when CONFIG_VIDEO_OV05C is defined.

@@ -1121,6 +1129,7 @@ static struct i2c_driver ov05_i2c_driver = {

module_i2c_driver(ov05_i2c_driver);

+MODULE_SOFTDEP("pre: pinctrl-amdisp");
MODULE_ALIAS("ov05");
MODULE_DESCRIPTION("OmniVision OV05 sensor driver");
MODULE_LICENSE("GPL and additional rights");

Please help to see if it works if you get time.

Regards,
Bin

Next message: Valentin Schneider: "Re: [PATCH v3 3/5] sched/fair: Switch to task based throttle model"
Previous message: Bryan O'Donoghue: "Re: [PATCH v9 5/8] media: venus: core: Sort dt_match alphabetically."
Next in thread: Sultan Alsawaf: "Re: [PATCH v2 0/8] Add AMD ISP4 driver"
Messages sorted by: [ date ] [ thread ] [ subject ] [ author ]