drm/vc4: module dysfunctional on Raspberry Pi 3B as of 5.18.0

drm/vc4: module dysfunctional on Raspberry Pi 3B as of 5.18.0

[Peter Mattern]

Hello.

As of Linux 5.18.0, module vc4 apparently isn't working on Raspberry Pi 
3B any more.

If a monitor is attached to the device, the boot messages show up as 
usual, but right when KMS starts, the screen turns black. Similarly, the 
screen also turns black when the module is blacklisted at boot time and 
loaded from the running system.
The problem looks quite similar to the one posted some months ago in [1].

Unfortunately, looking through systemd's journal didn't seem to yield 
any real hint. Nevertheless, the results from grepping vc4 are
→ 5.17.1
 > kernel: vc4-drm soc:gpu: bound 3f400000.hvs (ops vc4_hvs_ops [vc4])
 > kernel: rc rc0: vc4 as /devices/platform/soc/3f902000.hdmi/rc/rc0
 > kernel: input: vc4 as /devices/platform/soc/3f902000.hdmi/rc/rc0/input0
 > kernel: vc4-drm soc:gpu: bound 3f902000.hdmi (ops vc4_hdmi_ops [vc4])
 > kernel: vc4-drm soc:gpu: bound 3f806000.vec (ops vc4_vec_ops [vc4])
 > kernel: vc4-drm soc:gpu: bound 3f004000.txp (ops vc4_txp_ops [vc4])
 > kernel: vc4-drm soc:gpu: bound 3f206000.pixelvalve (ops vc4_crtc_ops 
[vc4])
 > kernel: vc4-drm soc:gpu: bound 3f207000.pixelvalve (ops vc4_crtc_ops 
[vc4])
 > kernel: vc4-drm soc:gpu: bound 3f807000.pixelvalve (ops vc4_crtc_ops 
[vc4])
 > kernel: vc4-drm soc:gpu: bound 3fc00000.v3d (ops vc4_v3d_ops [vc4])
 > kernel: fb0: switching to vc4 from simple
 > kernel: [drm] Initialized vc4 0.0.0 20140616 for soc:gpu on minor 0
 > kernel: vc4-drm soc:gpu: [drm] fb0: vc4drmfb frame buffer device
 > systemd-logind[338]: Watching system buttons on /dev/input/event0 (vc4)
→ 5.18.0
 > kernel: fb0: switching to vc4 from simple
 > kernel: vc4-drm soc:gpu: bound 3f400000.hvs (ops vc4_hvs_ops [vc4])
 > kernel: rc rc0: vc4 as /devices/platform/soc/3f902000.hdmi/rc/rc0
 > kernel: input: vc4 as /devices/platform/soc/3f902000.hdmi/rc/rc0/input0
 > kernel: vc4-drm soc:gpu: bound 3f902000.hdmi (ops vc4_hdmi_ops [vc4])
 > kernel: vc4-drm soc:gpu: bound 3f806000.vec (ops vc4_vec_ops [vc4])
 > kernel: vc4-drm soc:gpu: bound 3f004000.txp (ops vc4_txp_ops [vc4])
 > kernel: vc4-drm soc:gpu: bound 3f206000.pixelvalve (ops vc4_crtc_ops 
[vc4])
 > kernel: vc4-drm soc:gpu: bound 3f207000.pixelvalve (ops vc4_crtc_ops 
[vc4])
 > kernel: vc4-drm soc:gpu: bound 3f807000.pixelvalve (ops vc4_crtc_ops 
[vc4])
 > kernel: vc4-drm soc:gpu: bound 3fc00000.v3d (ops vc4_v3d_ops [vc4])
 > kernel: [drm] Initialized vc4 0.0.0 20140616 for soc:gpu on minor 0
 > kernel: vc4-drm soc:gpu: [drm] fb0: vc4drmfb frame buffer device
 > systemd-logind[337]: Watching system buttons on /dev/input/event0 (vc4)

Regards,

Peter Mattern

[1] 
https://lists.freedesktop.org/archives/dri-devel/2022-January/339458.html

Re: drm/vc4: module dysfunctional on Raspberry Pi 3B as of 5.18.0

[Peter Robinson]

Hi Peter,

Adding Stefan and Maxime

> As of Linux 5.18.0, module vc4 apparently isn't working on Raspberry Pi
> 3B any more.
>
> If a monitor is attached to the device, the boot messages show up as
> usual, but right when KMS starts, the screen turns black. Similarly, the
> screen also turns black when the module is blacklisted at boot time and
> loaded from the running system.
> The problem looks quite similar to the one posted some months ago in [1].
>
> Unfortunately, looking through systemd's journal didn't seem to yield
> any real hint. Nevertheless, the results from grepping vc4 are

I'm seeing the same issue with vc4 on a RPi3 on 5.18.1 on Fedora so
can confirm the regression. Maxime would know what might be up here?

> → 5.17.1
>  > kernel: vc4-drm soc:gpu: bound 3f400000.hvs (ops vc4_hvs_ops [vc4])
>  > kernel: rc rc0: vc4 as /devices/platform/soc/3f902000.hdmi/rc/rc0
>  > kernel: input: vc4 as /devices/platform/soc/3f902000.hdmi/rc/rc0/input0
>  > kernel: vc4-drm soc:gpu: bound 3f902000.hdmi (ops vc4_hdmi_ops [vc4])
>  > kernel: vc4-drm soc:gpu: bound 3f806000.vec (ops vc4_vec_ops [vc4])
>  > kernel: vc4-drm soc:gpu: bound 3f004000.txp (ops vc4_txp_ops [vc4])
>  > kernel: vc4-drm soc:gpu: bound 3f206000.pixelvalve (ops vc4_crtc_ops
> [vc4])
>  > kernel: vc4-drm soc:gpu: bound 3f207000.pixelvalve (ops vc4_crtc_ops
> [vc4])
>  > kernel: vc4-drm soc:gpu: bound 3f807000.pixelvalve (ops vc4_crtc_ops
> [vc4])
>  > kernel: vc4-drm soc:gpu: bound 3fc00000.v3d (ops vc4_v3d_ops [vc4])
>  > kernel: fb0: switching to vc4 from simple
>  > kernel: [drm] Initialized vc4 0.0.0 20140616 for soc:gpu on minor 0
>  > kernel: vc4-drm soc:gpu: [drm] fb0: vc4drmfb frame buffer device
>  > systemd-logind[338]: Watching system buttons on /dev/input/event0 (vc4)
> → 5.18.0
>  > kernel: fb0: switching to vc4 from simple
>  > kernel: vc4-drm soc:gpu: bound 3f400000.hvs (ops vc4_hvs_ops [vc4])
>  > kernel: rc rc0: vc4 as /devices/platform/soc/3f902000.hdmi/rc/rc0
>  > kernel: input: vc4 as /devices/platform/soc/3f902000.hdmi/rc/rc0/input0
>  > kernel: vc4-drm soc:gpu: bound 3f902000.hdmi (ops vc4_hdmi_ops [vc4])
>  > kernel: vc4-drm soc:gpu: bound 3f806000.vec (ops vc4_vec_ops [vc4])
>  > kernel: vc4-drm soc:gpu: bound 3f004000.txp (ops vc4_txp_ops [vc4])
>  > kernel: vc4-drm soc:gpu: bound 3f206000.pixelvalve (ops vc4_crtc_ops
> [vc4])
>  > kernel: vc4-drm soc:gpu: bound 3f207000.pixelvalve (ops vc4_crtc_ops
> [vc4])
>  > kernel: vc4-drm soc:gpu: bound 3f807000.pixelvalve (ops vc4_crtc_ops
> [vc4])
>  > kernel: vc4-drm soc:gpu: bound 3fc00000.v3d (ops vc4_v3d_ops [vc4])
>  > kernel: [drm] Initialized vc4 0.0.0 20140616 for soc:gpu on minor 0
>  > kernel: vc4-drm soc:gpu: [drm] fb0: vc4drmfb frame buffer device
>  > systemd-logind[337]: Watching system buttons on /dev/input/event0 (vc4)
>
> Regards,
>
> Peter Mattern
>
> [1]
> https://lists.freedesktop.org/archives/dri-devel/2022-January/339458.html
>

Re: drm/vc4: module dysfunctional on Raspberry Pi 3B as of 5.18.0

[Maxime Ripard]

[-- Attachment #1: Type: text/plain, Size: 3520 bytes --]

Hi Peter(s)

On Wed, Jun 08, 2022 at 02:10:19PM +0100, Peter Robinson wrote:
> Hi Peter,
> 
> Adding Stefan and Maxime
> 
> > As of Linux 5.18.0, module vc4 apparently isn't working on Raspberry Pi
> > 3B any more.
> >
> > If a monitor is attached to the device, the boot messages show up as
> > usual, but right when KMS starts, the screen turns black. Similarly, the
> > screen also turns black when the module is blacklisted at boot time and
> > loaded from the running system.
> > The problem looks quite similar to the one posted some months ago in [1].

If I understand you properly, it results in a blank screen if the
monitor is connected, but the system is still responsive?

If so, it's a very different problem than the link you provided, since
it was occurring when no monitor was connected and resulted in a total
system hang.

> > Unfortunately, looking through systemd's journal didn't seem to yield
> > any real hint. Nevertheless, the results from grepping vc4 are
> 
> I'm seeing the same issue with vc4 on a RPi3 on 5.18.1 on Fedora so
> can confirm the regression. Maxime would know what might be up here?

I tested on 5.18 on my 3B and it works well. Could you paste your kernel
configuration and config.txt somewhere?

> > → 5.17.1
> >  > kernel: vc4-drm soc:gpu: bound 3f400000.hvs (ops vc4_hvs_ops [vc4])
> >  > kernel: rc rc0: vc4 as /devices/platform/soc/3f902000.hdmi/rc/rc0
> >  > kernel: input: vc4 as /devices/platform/soc/3f902000.hdmi/rc/rc0/input0
> >  > kernel: vc4-drm soc:gpu: bound 3f902000.hdmi (ops vc4_hdmi_ops [vc4])
> >  > kernel: vc4-drm soc:gpu: bound 3f806000.vec (ops vc4_vec_ops [vc4])
> >  > kernel: vc4-drm soc:gpu: bound 3f004000.txp (ops vc4_txp_ops [vc4])
> >  > kernel: vc4-drm soc:gpu: bound 3f206000.pixelvalve (ops vc4_crtc_ops
> > [vc4])
> >  > kernel: vc4-drm soc:gpu: bound 3f207000.pixelvalve (ops vc4_crtc_ops
> > [vc4])
> >  > kernel: vc4-drm soc:gpu: bound 3f807000.pixelvalve (ops vc4_crtc_ops
> > [vc4])
> >  > kernel: vc4-drm soc:gpu: bound 3fc00000.v3d (ops vc4_v3d_ops [vc4])
> >  > kernel: fb0: switching to vc4 from simple
> >  > kernel: [drm] Initialized vc4 0.0.0 20140616 for soc:gpu on minor 0
> >  > kernel: vc4-drm soc:gpu: [drm] fb0: vc4drmfb frame buffer device
> >  > systemd-logind[338]: Watching system buttons on /dev/input/event0 (vc4)
> > → 5.18.0
> >  > kernel: fb0: switching to vc4 from simple
> >  > kernel: vc4-drm soc:gpu: bound 3f400000.hvs (ops vc4_hvs_ops [vc4])
> >  > kernel: rc rc0: vc4 as /devices/platform/soc/3f902000.hdmi/rc/rc0
> >  > kernel: input: vc4 as /devices/platform/soc/3f902000.hdmi/rc/rc0/input0
> >  > kernel: vc4-drm soc:gpu: bound 3f902000.hdmi (ops vc4_hdmi_ops [vc4])
> >  > kernel: vc4-drm soc:gpu: bound 3f806000.vec (ops vc4_vec_ops [vc4])
> >  > kernel: vc4-drm soc:gpu: bound 3f004000.txp (ops vc4_txp_ops [vc4])
> >  > kernel: vc4-drm soc:gpu: bound 3f206000.pixelvalve (ops vc4_crtc_ops
> > [vc4])
> >  > kernel: vc4-drm soc:gpu: bound 3f207000.pixelvalve (ops vc4_crtc_ops
> > [vc4])
> >  > kernel: vc4-drm soc:gpu: bound 3f807000.pixelvalve (ops vc4_crtc_ops
> > [vc4])
> >  > kernel: vc4-drm soc:gpu: bound 3fc00000.v3d (ops vc4_v3d_ops [vc4])
> >  > kernel: [drm] Initialized vc4 0.0.0 20140616 for soc:gpu on minor 0
> >  > kernel: vc4-drm soc:gpu: [drm] fb0: vc4drmfb frame buffer device
> >  > systemd-logind[337]: Watching system buttons on /dev/input/event0 (vc4)

Yeah, it doesn't look that different.

Maxime

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 228 bytes --]

Re: drm/vc4: module dysfunctional on Raspberry Pi 3B as of 5.18.0

[Peter Robinson]

Hi Maxime,

> > > As of Linux 5.18.0, module vc4 apparently isn't working on Raspberry Pi
> > > 3B any more.
> > >
> > > If a monitor is attached to the device, the boot messages show up as
> > > usual, but right when KMS starts, the screen turns black. Similarly, the
> > > screen also turns black when the module is blacklisted at boot time and
> > > loaded from the running system.
> > > The problem looks quite similar to the one posted some months ago in [1].
>
> If I understand you properly, it results in a blank screen if the
> monitor is connected, but the system is still responsive?

Yes, it boots fine, I see all serial console output and the module
loads etc, without a screen I didn't notice the issue

> If so, it's a very different problem than the link you provided, since
> it was occurring when no monitor was connected and resulted in a total
> system hang.
>
> > > Unfortunately, looking through systemd's journal didn't seem to yield
> > > any real hint. Nevertheless, the results from grepping vc4 are
> >
> > I'm seeing the same issue with vc4 on a RPi3 on 5.18.1 on Fedora so
> > can confirm the regression. Maxime would know what might be up here?
>
> I tested on 5.18 on my 3B and it works well. Could you paste your kernel
> configuration and config.txt somewhere?

It boots with output on 5.17.13, and not on 5.18. The config is at the
following link (Fedora config). The config.txt is below.
https://pbrobinson.fedorapeople.org/config-5.18.2-201.fc36.armv7hl

===
# Raspberry Pi 2
[pi2]
kernel=rpi2-u-boot.bin
dtoverlay=upstream

# Raspberry Pi 3
[pi3]
kernel=rpi3-u-boot.bin
dtoverlay=upstream

# Raspberry Pi 4
[pi4]
kernel=rpi4-u-boot.bin
dtoverlay=upstream-pi4

# Default Fedora configs for all Raspberry Pi Revisions
[all]
# Enable UART
# Only enable UART if you're going to use it as it has speed implications
# Serial console is ttyS0 on RPi3 and ttyAMA0 on all other variants
# u-boot will auto detect serial and pass corrent options to kernel if enabled
# Speed details: https://www.raspberrypi.org/forums/viewtopic.php?f=28&t=141195

# Terrible hack to work around U-Boot issues on most devices
enable_uart=1

# Early boot delay in the hope monitors are initialised enough to provide EDID
bootcode_delay=1

# We need this to be 32Mb to support VCHI services and drivers which use them
# but this isn't used by mainline VC4 driver so reduce to lowest supported value
# You need to set this to at least 80 for using the camera
gpu_mem=32

# Use eXtended firmware by default
start_x=1

# Stop the RPi turning on HDMI monitors on reboot
hdmi_ignore_cec_init=1

# New option to allow the firmware to load upstream dtb
# Will allow things like camera, touchscreen etc to work OOTB
upstream_kernel=1
===

Re: drm/vc4: module dysfunctional on Raspberry Pi 3B as of 5.18.0

[Maxime Ripard]

[-- Attachment #1: Type: text/plain, Size: 1688 bytes --]

On Wed, Jun 08, 2022 at 04:14:22PM +0100, Peter Robinson wrote:
> > > > As of Linux 5.18.0, module vc4 apparently isn't working on Raspberry Pi
> > > > 3B any more.
> > > >
> > > > If a monitor is attached to the device, the boot messages show up as
> > > > usual, but right when KMS starts, the screen turns black. Similarly, the
> > > > screen also turns black when the module is blacklisted at boot time and
> > > > loaded from the running system.
> > > > The problem looks quite similar to the one posted some months ago in [1].
> >
> > If I understand you properly, it results in a blank screen if the
> > monitor is connected, but the system is still responsive?
> 
> Yes, it boots fine, I see all serial console output and the module
> loads etc, without a screen I didn't notice the issue
> 
> > If so, it's a very different problem than the link you provided, since
> > it was occurring when no monitor was connected and resulted in a total
> > system hang.
> >
> > > > Unfortunately, looking through systemd's journal didn't seem to yield
> > > > any real hint. Nevertheless, the results from grepping vc4 are
> > >
> > > I'm seeing the same issue with vc4 on a RPi3 on 5.18.1 on Fedora so
> > > can confirm the regression. Maxime would know what might be up here?
> >
> > I tested on 5.18 on my 3B and it works well. Could you paste your kernel
> > configuration and config.txt somewhere?
> 
> It boots with output on 5.17.13, and not on 5.18.

Interestingly, it works for 5.18 in my case but doesn't for the current
drm-misc-next branch with similar symptoms.

I'll look into that one and see if the two might be related.

Thanks!
Maxime

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 228 bytes --]

Re: drm/vc4: module dysfunctional on Raspberry Pi 3B as of 5.18.0

[Maxime Ripard]

[-- Attachment #1: Type: text/plain, Size: 2029 bytes --]

On Wed, Jun 08, 2022 at 05:36:08PM +0200, Maxime Ripard wrote:
> On Wed, Jun 08, 2022 at 04:14:22PM +0100, Peter Robinson wrote:
> > > > > As of Linux 5.18.0, module vc4 apparently isn't working on Raspberry Pi
> > > > > 3B any more.
> > > > >
> > > > > If a monitor is attached to the device, the boot messages show up as
> > > > > usual, but right when KMS starts, the screen turns black. Similarly, the
> > > > > screen also turns black when the module is blacklisted at boot time and
> > > > > loaded from the running system.
> > > > > The problem looks quite similar to the one posted some months ago in [1].
> > >
> > > If I understand you properly, it results in a blank screen if the
> > > monitor is connected, but the system is still responsive?
> > 
> > Yes, it boots fine, I see all serial console output and the module
> > loads etc, without a screen I didn't notice the issue
> > 
> > > If so, it's a very different problem than the link you provided, since
> > > it was occurring when no monitor was connected and resulted in a total
> > > system hang.
> > >
> > > > > Unfortunately, looking through systemd's journal didn't seem to yield
> > > > > any real hint. Nevertheless, the results from grepping vc4 are
> > > >
> > > > I'm seeing the same issue with vc4 on a RPi3 on 5.18.1 on Fedora so
> > > > can confirm the regression. Maxime would know what might be up here?
> > >
> > > I tested on 5.18 on my 3B and it works well. Could you paste your kernel
> > > configuration and config.txt somewhere?
> > 
> > It boots with output on 5.17.13, and not on 5.18.
> 
> Interestingly, it works for 5.18 in my case but doesn't for the current
> drm-misc-next branch with similar symptoms.
> 
> I'll look into that one and see if the two might be related.

Actually, it was because drm-misc-next was missing 88110a9f6209, and
thus the DRM driver wouldn't load.

Once that patch is applied, 5.18 and all -rc work fine on my side.

Could you start a bisection maybe?

Maxime

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 228 bytes --]

Re: drm/vc4: module dysfunctional on Raspberry Pi 3B as of 5.18.0

[Peter Robinson]

> > > > > > As of Linux 5.18.0, module vc4 apparently isn't working on Raspberry Pi
> > > > > > 3B any more.
> > > > > >
> > > > > > If a monitor is attached to the device, the boot messages show up as
> > > > > > usual, but right when KMS starts, the screen turns black. Similarly, the
> > > > > > screen also turns black when the module is blacklisted at boot time and
> > > > > > loaded from the running system.
> > > > > > The problem looks quite similar to the one posted some months ago in [1].
> > > >
> > > > If I understand you properly, it results in a blank screen if the
> > > > monitor is connected, but the system is still responsive?
> > >
> > > Yes, it boots fine, I see all serial console output and the module
> > > loads etc, without a screen I didn't notice the issue
> > >
> > > > If so, it's a very different problem than the link you provided, since
> > > > it was occurring when no monitor was connected and resulted in a total
> > > > system hang.
> > > >
> > > > > > Unfortunately, looking through systemd's journal didn't seem to yield
> > > > > > any real hint. Nevertheless, the results from grepping vc4 are
> > > > >
> > > > > I'm seeing the same issue with vc4 on a RPi3 on 5.18.1 on Fedora so
> > > > > can confirm the regression. Maxime would know what might be up here?
> > > >
> > > > I tested on 5.18 on my 3B and it works well. Could you paste your kernel
> > > > configuration and config.txt somewhere?
> > >
> > > It boots with output on 5.17.13, and not on 5.18.
> >
> > Interestingly, it works for 5.18 in my case but doesn't for the current
> > drm-misc-next branch with similar symptoms.
> >
> > I'll look into that one and see if the two might be related.
>
> Actually, it was because drm-misc-next was missing 88110a9f6209, and
> thus the DRM driver wouldn't load.
>
> Once that patch is applied, 5.18 and all -rc work fine on my side.

So that patch looks like it's already in 5.18.x:
$ git tag --contains 88110a9f6209
v5.18
v5.18.1
v5.18.2
v5.18.3
v5.19-rc1


> Could you start a bisection maybe?
>
> Maxime

Re: drm/vc4: module dysfunctional on Raspberry Pi 3B as of 5.18.0

[Peter Mattern]

 > Could you start a bisection maybe?

I for one am having two issues here.

The harmless one is that I'm lacking a cooler for the RPi and my cross 
compiling skills have become a bit rusty.
Both could be fixed quickly, of course.

The not so harmless one is that kernel 5.18.x is completely unusable due 
to an other regression: the SD card is slow to a degree that makes it 
rather unusable, rather simple tasks are tainting the kernel because of 
this. E. g. a simple kernel update takes half an hour on 5.18.x.
Neither sure what's the culprit here nor how to deal with it.

Nonetheless:
Could the module be installed without updating the kernel as a whole, e. 
g.by DKMS?
Is there a way to narrow down the commits in question, e. g. something 
like 'git log ./drivers/gpu/drm/vc4/' in a Git checkout of the kernel 
sources?

Re: drm/vc4: module dysfunctional on Raspberry Pi 3B as of 5.18.0

[Stefan Wahren]

Hi,

Am 08.06.22 um 16:36 schrieb Maxime Ripard:
> Hi Peter(s)
>
> On Wed, Jun 08, 2022 at 02:10:19PM +0100, Peter Robinson wrote:
>> Hi Peter,
>>
>> Adding Stefan and Maxime
>>
>>> As of Linux 5.18.0, module vc4 apparently isn't working on Raspberry Pi
>>> 3B any more.
>>>
>>> If a monitor is attached to the device, the boot messages show up as
>>> usual, but right when KMS starts, the screen turns black. Similarly, the
>>> screen also turns black when the module is blacklisted at boot time and
>>> loaded from the running system.
>>> The problem looks quite similar to the one posted some months ago in [1].
> If I understand you properly, it results in a blank screen if the
> monitor is connected, but the system is still responsive?
>
> If so, it's a very different problem than the link you provided, since
> it was occurring when no monitor was connected and resulted in a total
> system hang.
>
>>> Unfortunately, looking through systemd's journal didn't seem to yield
>>> any real hint. Nevertheless, the results from grepping vc4 are
>> I'm seeing the same issue with vc4 on a RPi3 on 5.18.1 on Fedora so
>> can confirm the regression. Maxime would know what might be up here?

i assume you are using the downstream DTB?


Please provide the version/date of the GPU firmware?

> I tested on 5.18 on my 3B and it works well. Could you paste your kernel
> configuration and config.txt somewhere?
>
>>> → 5.17.1
>>>   > kernel: vc4-drm soc:gpu: bound 3f400000.hvs (ops vc4_hvs_ops [vc4])
>>>   > kernel: rc rc0: vc4 as /devices/platform/soc/3f902000.hdmi/rc/rc0
>>>   > kernel: input: vc4 as /devices/platform/soc/3f902000.hdmi/rc/rc0/input0
>>>   > kernel: vc4-drm soc:gpu: bound 3f902000.hdmi (ops vc4_hdmi_ops [vc4])
>>>   > kernel: vc4-drm soc:gpu: bound 3f806000.vec (ops vc4_vec_ops [vc4])
>>>   > kernel: vc4-drm soc:gpu: bound 3f004000.txp (ops vc4_txp_ops [vc4])
>>>   > kernel: vc4-drm soc:gpu: bound 3f206000.pixelvalve (ops vc4_crtc_ops
>>> [vc4])
>>>   > kernel: vc4-drm soc:gpu: bound 3f207000.pixelvalve (ops vc4_crtc_ops
>>> [vc4])
>>>   > kernel: vc4-drm soc:gpu: bound 3f807000.pixelvalve (ops vc4_crtc_ops
>>> [vc4])
>>>   > kernel: vc4-drm soc:gpu: bound 3fc00000.v3d (ops vc4_v3d_ops [vc4])
>>>   > kernel: fb0: switching to vc4 from simple
>>>   > kernel: [drm] Initialized vc4 0.0.0 20140616 for soc:gpu on minor 0
>>>   > kernel: vc4-drm soc:gpu: [drm] fb0: vc4drmfb frame buffer device
>>>   > systemd-logind[338]: Watching system buttons on /dev/input/event0 (vc4)
>>> → 5.18.0
>>>   > kernel: fb0: switching to vc4 from simple
>>>   > kernel: vc4-drm soc:gpu: bound 3f400000.hvs (ops vc4_hvs_ops [vc4])
>>>   > kernel: rc rc0: vc4 as /devices/platform/soc/3f902000.hdmi/rc/rc0
>>>   > kernel: input: vc4 as /devices/platform/soc/3f902000.hdmi/rc/rc0/input0
>>>   > kernel: vc4-drm soc:gpu: bound 3f902000.hdmi (ops vc4_hdmi_ops [vc4])
>>>   > kernel: vc4-drm soc:gpu: bound 3f806000.vec (ops vc4_vec_ops [vc4])
>>>   > kernel: vc4-drm soc:gpu: bound 3f004000.txp (ops vc4_txp_ops [vc4])
>>>   > kernel: vc4-drm soc:gpu: bound 3f206000.pixelvalve (ops vc4_crtc_ops
>>> [vc4])
>>>   > kernel: vc4-drm soc:gpu: bound 3f207000.pixelvalve (ops vc4_crtc_ops
>>> [vc4])
>>>   > kernel: vc4-drm soc:gpu: bound 3f807000.pixelvalve (ops vc4_crtc_ops
>>> [vc4])
>>>   > kernel: vc4-drm soc:gpu: bound 3fc00000.v3d (ops vc4_v3d_ops [vc4])
>>>   > kernel: [drm] Initialized vc4 0.0.0 20140616 for soc:gpu on minor 0
>>>   > kernel: vc4-drm soc:gpu: [drm] fb0: vc4drmfb frame buffer device
>>>   > systemd-logind[337]: Watching system buttons on /dev/input/event0 (vc4)
> Yeah, it doesn't look that different.
>
> Maxime

Re: drm/vc4: module dysfunctional on Raspberry Pi 3B as of 5.18.0

[Peter Robinson]

> >>> As of Linux 5.18.0, module vc4 apparently isn't working on Raspberry Pi
> >>> 3B any more.
> >>>
> >>> If a monitor is attached to the device, the boot messages show up as
> >>> usual, but right when KMS starts, the screen turns black. Similarly, the
> >>> screen also turns black when the module is blacklisted at boot time and
> >>> loaded from the running system.
> >>> The problem looks quite similar to the one posted some months ago in [1].
> > If I understand you properly, it results in a blank screen if the
> > monitor is connected, but the system is still responsive?
> >
> > If so, it's a very different problem than the link you provided, since
> > it was occurring when no monitor was connected and resulted in a total
> > system hang.
> >
> >>> Unfortunately, looking through systemd's journal didn't seem to yield
> >>> any real hint. Nevertheless, the results from grepping vc4 are
> >> I'm seeing the same issue with vc4 on a RPi3 on 5.18.1 on Fedora so
> >> can confirm the regression. Maxime would know what might be up here?
>
> i assume you are using the downstream DTB?

In this case it's the one in the Linux kernel (not sure if that's
upstream or downstream in this context) not the one provided by the
RPi firmware.

> Please provide the version/date of the GPU firmware?

[    6.205121] raspberrypi-firmware soc:firmware: Attached to firmware
from 2022-05-24T13:26:52
Which is git f145afc

Re: drm/vc4: module dysfunctional on Raspberry Pi 3B as of 5.18.0

[Stefan Wahren]

Hi Peter,

Am 09.06.22 um 13:52 schrieb Peter Robinson:
>>>>> As of Linux 5.18.0, module vc4 apparently isn't working on Raspberry Pi
>>>>> 3B any more.
>>>>>
>>>>> If a monitor is attached to the device, the boot messages show up as
>>>>> usual, but right when KMS starts, the screen turns black. Similarly, the
>>>>> screen also turns black when the module is blacklisted at boot time and
>>>>> loaded from the running system.
>>>>> The problem looks quite similar to the one posted some months ago in [1].
>>> If I understand you properly, it results in a blank screen if the
>>> monitor is connected, but the system is still responsive?
>>>
>>> If so, it's a very different problem than the link you provided, since
>>> it was occurring when no monitor was connected and resulted in a total
>>> system hang.
>>>
>>>>> Unfortunately, looking through systemd's journal didn't seem to yield
>>>>> any real hint. Nevertheless, the results from grepping vc4 are
>>>> I'm seeing the same issue with vc4 on a RPi3 on 5.18.1 on Fedora so
>>>> can confirm the regression. Maxime would know what might be up here?
>> i assume you are using the downstream DTB?
> In this case it's the one in the Linux kernel (not sure if that's
> upstream or downstream in this context) not the one provided by the
> RPi firmware.

okay thanks. I tried the kernel config but its huge.

Is it correct that the issue occurs with your kernel config, but not 
with multi_v7_defconfig?

>
>> Please provide the version/date of the GPU firmware?
> [    6.205121] raspberrypi-firmware soc:firmware: Attached to firmware
> from 2022-05-24T13:26:52
> Which is git f145afc
This is good. I had issues with older firmware which didn't implement 
RPI_FIRMWARE_NOTIFY_DISPLAY_DONE. So it must be something different.

Re: drm/vc4: module dysfunctional on Raspberry Pi 3B as of 5.18.0

[Stefan Wahren]

Hi,

Am 09.06.22 um 23:33 schrieb Stefan Wahren:
> Hi Peter,
>
> Am 09.06.22 um 13:52 schrieb Peter Robinson:
>>>>>> As of Linux 5.18.0, module vc4 apparently isn't working on 
>>>>>> Raspberry Pi
>>>>>> 3B any more.
>>>>>>
>>>>>> If a monitor is attached to the device, the boot messages show up as
>>>>>> usual, but right when KMS starts, the screen turns black. 
>>>>>> Similarly, the
>>>>>> screen also turns black when the module is blacklisted at boot 
>>>>>> time and
>>>>>> loaded from the running system.
>>>>>> The problem looks quite similar to the one posted some months ago 
>>>>>> in [1].
>>>> If I understand you properly, it results in a blank screen if the
>>>> monitor is connected, but the system is still responsive?
>>>>
>>>> If so, it's a very different problem than the link you provided, since
>>>> it was occurring when no monitor was connected and resulted in a total
>>>> system hang.
>>>>
>>>>>> Unfortunately, looking through systemd's journal didn't seem to 
>>>>>> yield
>>>>>> any real hint. Nevertheless, the results from grepping vc4 are
>>>>> I'm seeing the same issue with vc4 on a RPi3 on 5.18.1 on Fedora so
>>>>> can confirm the regression. Maxime would know what might be up here?
>>> i assume you are using the downstream DTB?
>> In this case it's the one in the Linux kernel (not sure if that's
>> upstream or downstream in this context) not the one provided by the
>> RPi firmware.
>
> okay thanks. I tried the kernel config but its huge.
>
> Is it correct that the issue occurs with your kernel config, but not 
> with multi_v7_defconfig?

FWIW here is what i tried to reproduce the issue:

- enable CONFIG_DRM_VC4_HDMI_CEC
- apply the relevant config.txt options from Peter R.
- update my gcc toolchain to 11.3
- switch compiler optimization from speed to size

But the display always works.

Please tell the size of your bcm2837-rpi-3-b.dtb

Re: drm/vc4: module dysfunctional on Raspberry Pi 3B as of 5.18.0

[Peter Robinson]

On Thu, Jun 9, 2022 at 10:33 PM Stefan Wahren <stefan.wahren@i2se.com> wrote:
>
> Hi Peter,
>
> Am 09.06.22 um 13:52 schrieb Peter Robinson:
> >>>>> As of Linux 5.18.0, module vc4 apparently isn't working on Raspberry Pi
> >>>>> 3B any more.
> >>>>>
> >>>>> If a monitor is attached to the device, the boot messages show up as
> >>>>> usual, but right when KMS starts, the screen turns black. Similarly, the
> >>>>> screen also turns black when the module is blacklisted at boot time and
> >>>>> loaded from the running system.
> >>>>> The problem looks quite similar to the one posted some months ago in [1].
> >>> If I understand you properly, it results in a blank screen if the
> >>> monitor is connected, but the system is still responsive?
> >>>
> >>> If so, it's a very different problem than the link you provided, since
> >>> it was occurring when no monitor was connected and resulted in a total
> >>> system hang.
> >>>
> >>>>> Unfortunately, looking through systemd's journal didn't seem to yield
> >>>>> any real hint. Nevertheless, the results from grepping vc4 are
> >>>> I'm seeing the same issue with vc4 on a RPi3 on 5.18.1 on Fedora so
> >>>> can confirm the regression. Maxime would know what might be up here?
> >> i assume you are using the downstream DTB?
> > In this case it's the one in the Linux kernel (not sure if that's
> > upstream or downstream in this context) not the one provided by the
> > RPi firmware.
>
> okay thanks. I tried the kernel config but its huge.

Welcome to distro configs.

> Is it correct that the issue occurs with your kernel config, but not
> with multi_v7_defconfig?

I've not tried it, ultimately the distro config is where it's used and
the rpi config within that hasn't changed for some time.

> >
> >> Please provide the version/date of the GPU firmware?
> > [    6.205121] raspberrypi-firmware soc:firmware: Attached to firmware
> > from 2022-05-24T13:26:52
> > Which is git f145afc
> This is good. I had issues with older firmware which didn't implement
> RPI_FIRMWARE_NOTIFY_DISPLAY_DONE. So it must be something different.

Re: drm/vc4: module dysfunctional on Raspberry Pi 3B as of 5.18.0

[Peter Mattern]

 > If I understand you properly, it results in a blank screen if the
 > monitor is connected, but the system is still responsive?
Yes. Similar to (the other) Peter's findings, the system is fully 
responsive, it's just that the monitor is displaying a black screen.
Meanwhile I stumbled upon another detail: when the problem is manifest, 
the monitor's LED displaying whether a signal is available keeps shining 
permanently. I does not start flashing as it does when an attached 
computer is switched off.

 > If so, it's a very different problem than the link you provided, since
 > it was occurring when no monitor was connected and resulted in a total
 > system hang.
I didn't follow up things too closely. But I remember very well that I 
stumbled upon that thread "Linux 5.16.3 breaks headless Raspberry Pi 3 
boot, again" while I was facing the exact same problem we're discussing 
right now. Also, some findings in the Arch Linux ARM forums were similar 
then [1].
The problem did disappear in 5.16.7 when some offending commit was reverted.

 > I tested on 5.18 on my 3B and it works well. Could you paste your kernel
 > configuration and config.txt somewhere?
Just done in a Mail which was supposed to respond to Stefan's first one, 
which went wrong. (As Stefan isn't subscribed his mail was lacking the 
usual headers like "In-Reply-to". So I started editing by hitting the 
link on top of [2]. But this didn't work.)

[1] https://archlinuxarm.org/forum/viewtopic.php?f=57&t=15842
[2] https://lists.freedesktop.org/archives/dri-devel/2022-June/357819.html

drm/vc4: module dysfunctional on Raspberry Pi 3B as of 5.18.0

[Stefan Wahren]

Hi Peter,

i didn't subscribe to dri-devel, but i noticed your bug report.

Could you please provide more information:

Which kernel config do you use (is it a defconfig)?
Do you use the mainline device tree blob or the Raspberry Pi DTB?
Please provide the version/date of the GPU firmware?
Do you have any KMS related setting in the config.txt?
Can you provide a full dmesg for the bad case?

Best regards

drm/vc4: module dysfunctional on Raspberry Pi 3B as of 5.18.0

[Peter Mattern]

 > Which kernel config do you use (is it a defconfig)?
These are custom configs provided by the distribution, the origin of 
which I do not know.
You can find them at [1] (5.17.0, used in 5.17.1 as well) and [2] 
(5.18.0, modified in 5.18.1).
The difference between these files and their upstream (kernel.org) 
counterpart as created by running 'make ARCH=arm64 defconfig' is about 
11k lines, the difference between the two files of the distribution 
around 800 lines. Didn't investigate the differences in detail yet.

 > Do you use the mainline device tree blob or the Raspberry Pi DTB?
It's the one from the kernel.org sources.
Note: At times you'll find claims that the DTBs of the Raspberry Pi 
Foundation were used on Arch Linux ARM. These probably are due to the 
fact that those blobs happen to be around. But U-Boot's boot.scr does 
pick the "mainline" ones in /boot/dtbs/.

 > Please provide the version/date of the GPU firmware?
It is provided by the Broadcom bootloader package (which is used to 
chainload U-Boot). It happened to be updated in parallel with kernel 
5.18.0 to commit f145afcf [3] and has since been updated to upstream 
commit 8f91a969. Those latest changes don't seem to have an impact on 
the problem we're discussing here (not all combinations bootloader / 
kernel tested).

 > Do you have any KMS related setting in the config.txt?
No.
My epic config.txt is: enable_uart=1

 > Can you provide a full dmesg for the bad case?
I sure could. But grepping through it yields the exact same results 
included in my first post. And assuming this list scrubs attachments I'm 
not sure whether posting those 500 lines in a mail would make sense?

[1] 
https://github.com/archlinuxarm/PKGBUILDs/blob/ae861a830e6e1d6d3e37c94462c8f48671aa9b79/core/linux-aarch64/config
[2] 
https://github.com/archlinuxarm/PKGBUILDs/blob/9c5b51b25ecc038cd505b26e15bb80464783a713/core/linux-aarch64/config
[3] 
https://github.com/archlinuxarm/PKGBUILDs/blob/722080b0aaa7e158d91b323d27c41839d1ef42f0/alarm/raspberrypi-bootloader/PKGBUILD