Re: [REGRESSION] drm/panel/panel-simple v6.17 WARNING regression

["Luca Ceresoli" <luca.ceresoli@bootlin.com>]

Hello Ernest,

thanks for your detailed analysis and report!

On Tue Oct 14, 2025 at 10:13 AM CEST, Ernest Van Hoecke wrote:
> Hello everyone,
>
> Commit 94d50c1a2ca3 ("drm/bridge: get/put the bridge reference in drm_bridge_attach/detach()")
> introduced a regression with the below warning dure probe with panel
> dpi described in the device tree.
>
> [    9.160074] ------------[ cut here ]------------
> [    9.160092] WARNING: CPU: 0 PID: 66 at /usr/src/kernel/lib/refcount.c:25 drm_bridge_attach+0x2c/0x1dc
> [    9.160138] refcount_t: addition on 0; use-after-free.
> [    9.160147] Modules linked in: coda_vpu(+) snd_soc_fsl_asrc snd_compress v4l2_jpeg pwm_imx27 imx_vdoa spi_imx imx6_media(C) imx_media_common(C) videobuf2_dma_contig etnaviv snd_soc_fsl_ssi v4l2_mem2mem imx_pcm_dma panel_simple gpio_keys gpu_sched pwm_bl loop fuse ipv6 autofs4
> [    9.160423] CPU: 0 UID: 0 PID: 66 Comm: kworker/u10:2 Tainted: G         C          6.17.0-rc5-0.0.0-devel #1 PREEMPT
> [    9.160459] Tainted: [C]=CRAP
> [    9.160476] Hardware name: Freescale i.MX6 Quad/DualLite (Device Tree)
> [    9.160497] Workqueue: events_unbound deferred_probe_work_func
> [    9.160535] Call trace:
> [    9.160546]  unwind_backtrace from show_stack+0x10/0x14
> [    9.160603]  show_stack from dump_stack_lvl+0x54/0x68
> [    9.160637]  dump_stack_lvl from __warn+0x7c/0xd4
> [    9.160672]  __warn from warn_slowpath_fmt+0x130/0x1c4
> [    9.160726]  warn_slowpath_fmt from drm_bridge_attach+0x2c/0x1dc
> [    9.160780]  drm_bridge_attach from imx_pd_bind+0x74/0xa0
> [    9.160836]  imx_pd_bind from component_bind_all+0xfc/0x254
> [    9.160881]  component_bind_all from imx_drm_bind+0xa8/0x154
> [    9.160903]  imx_drm_bind from try_to_bring_up_aggregate_device+0x1f8/0x2b0
> [    9.160959]  try_to_bring_up_aggregate_device from __component_add+0x9c/0x160
> [    9.161003]  __component_add from imx_pd_probe+0xa8/0x160
> [    9.161042]  imx_pd_probe from platform_probe+0x5c/0x90
> [    9.161066]  platform_probe from really_probe+0xd0/0x3a4
> [    9.161093]  really_probe from __driver_probe_device+0x8c/0x1d4
> [    9.161144]  __driver_probe_device from driver_probe_device+0x34/0xd0
> [    9.161195]  driver_probe_device from __device_attach_driver+0xa4/0x134
> [    9.161248]  __device_attach_driver from bus_for_each_drv+0x90/0xe4
> [    9.161299]  bus_for_each_drv from __device_attach+0xa4/0x1cc
> [    9.161328]  __device_attach from bus_probe_device+0x84/0x88
> [    9.161354]  bus_probe_device from deferred_probe_work_func+0x8c/0xcc
> [    9.161395]  deferred_probe_work_func from process_one_work+0x158/0x2e0
> [    9.161434]  process_one_work from worker_thread+0x254/0x3fc
> [    9.161449]  worker_thread from kthread+0x128/0x24c
> [    9.161473]  kthread from ret_from_fork+0x14/0x28
> [    9.161494] Exception stack(0xe0aa1fb0 to 0xe0aa1ff8)
> [    9.161505] 1fa0:                                     00000000 00000000 00000000 00000000
> [    9.161517] 1fc0: 00000000 00000000 00000000 00000000 00000000 00000000 00000000 00000000
> [    9.161529] 1fe0: 00000000 00000000 00000000 00000000 00000013 00000000
> [    9.161539] ---[ end trace 0000000000000000 ]---
>
> Note that this commit was added to branch 'drm-next-2025-07-30',
> which did not contain an already mainlined fix for another regression [1].
> Without that fix [2], this new regression is masked.
>
> Reverting 94d50c1a2ca3 on top of
> commit 260f6f4fda93 ("Merge tag 'drm-next-2025-07-30' of https://gitlab.freedesktop.org/drm/kernel")
> fixes the issue. At that point, [1] was fixed in mainline, and the new
> regression reported here was merged in. v6.17-rc1 to v6.17-rc7 are in
> this state.

This is because the DRM_IMX has not been converted to the new
devm_drm_bridge_alloc() [0] DRM bridge allocation API, which is now
mandatory.

I converted all the bridges, except for a few which I missed because my
search was based on looking for drm_bridge_add() calls, and this driver
does not call it.

I recently sent a series proposing to make drm_bridge_add() mandatory
before drm_bridge_attach() in the docs and warn if that is violated [1]. If
you apply patch 4 of that series you should see the warning.

Let me have a look at the DRM_IMX driver, I'll try to send a series
converting it to the new API within today.

[0] https://elixir.bootlin.com/linux/v6.17.1/source/include/drm/drm_bridge.h#L1282
[1] https://lore.kernel.org/lkml/20251003-b4-drm-bridge-alloc-add-before-attach-v1-0-92fb40d27704@bootlin.com/#t

> However, later on, another regression seems to be introduced by
> commit 8fa5909400f3 ("drm/bridge: get the bridge returned by drm_bridge_chain_get_first_bridge()")
> so reverting 94d50c1a2ca3 on top of drm-misc-next does not solve
> everything. This was tested by rebasing drm-misc-next onto (260f6f4fda93
> plus the revert of 94d50c1a2ca3) and then bisecting.
>
> So in v6.18-rc1, both regressions are present.
>
> There, I get the following additional warnings:
>
> [    9.732278] ------------[ cut here ]------------
> [    9.732336] WARNING: CPU: 0 PID: 38 at lib/refcount.c:22 drm_bridge_get+0x10/0x18
> [    9.744608] refcount_t: saturated; leaking memory.

Not sure here, but it may well be another symptom of the same bug: the
refcount was not initialized correctly, so it is found inconsistent later
when trying to increase it. Let's fix the known issue and then we'll see.

Best regards,
Luca

--
Luca Ceresoli, Bootlin
Embedded Linux and Kernel engineering
https://bootlin.com