On Wed, Jun 18, 2025 at 10:51:58AM +0200, Luca Ceresoli wrote:
> Hello Anusha, Francesco,
>
> On Tue, 17 Jun 2025 11:17:20 -0500
> Anusha Srivatsa <asriv...@redhat.com> wrote:
>
> > On Thu, Jun 12, 2025 at 3:24 AM Francesco Dolcini <france...@dolcini.it>
> > wrote:
> >
> > > Hello all,
> > >
> > > Commit de04bb0089a9 ("drm/panel/panel-simple: Use the new allocation in
> > > place of devm_kzalloc()")
> > > from 6.16-rc1 introduced a regression with this warning during probe
> > > with panel dpi described in the DT.
> > >
> > > A revert solves the issue.
> > >
> > > The issue is that connector_type is set to DRM_MODE_CONNECTOR_DPI in
> > > panel_dpi_probe() that after that change is called after
> > > devm_drm_panel_alloc().
> > >
> > > I am not sure if there are other implication for this change in the call
> > > ordering, apart the one that triggers this warning.
> > >
> > > [ 12.089274] ------------[ cut here ]------------
> > > [ 12.089303] WARNING: CPU: 0 PID: 96 at
> > > drivers/gpu/drm/bridge/panel.c:377 devm_drm_of_get_bridge+0xac/0xb8
> > > [ 12.130808] Modules linked in: v4l2_jpeg pwm_imx27(+) imx_vdoa
> > > gpu_sched panel_simple imx6_media(C) imx_media_common
> > > (C) videobuf2_dma_contig pwm_bl gpio_keys v4l2_mem2mem fuse ipv6 autofs4
> > > [ 12.147774] CPU: 0 UID: 0 PID: 96 Comm: kworker/u8:3 Tainted: G
> > > C 6.16.0-rc1+ #1 PREEMPT
> > > [ 12.157446] Tainted: [C]=CRAP
> > > [ 12.160418] Hardware name: Freescale i.MX6 Quad/DualLite (Device Tree)
> > > [ 12.166953] Workqueue: events_unbound deferred_probe_work_func
> > > [ 12.172805] Call trace:
> > > [ 12.172815] unwind_backtrace from show_stack+0x10/0x14
> > > [ 12.180598] show_stack from dump_stack_lvl+0x68/0x74
> > > [ 12.185674] dump_stack_lvl from __warn+0x7c/0xe0
> > > [ 12.190407] __warn from warn_slowpath_fmt+0x1b8/0x1c0
> > > [ 12.195567] warn_slowpath_fmt from devm_drm_of_get_bridge+0xac/0xb8
> > > [ 12.201949] devm_drm_of_get_bridge from imx_pd_probe+0x58/0x164
> > > [ 12.207976] imx_pd_probe from platform_probe+0x5c/0xb0
> > > [ 12.213220] platform_probe from really_probe+0xd0/0x3a4
> > > [ 12.218551] really_probe from __driver_probe_device+0x8c/0x1d4
> > > [ 12.224486] __driver_probe_device from driver_probe_device+0x30/0xc0
> > > [ 12.230942] driver_probe_device from __device_attach_driver+0x98/0x10c
> > > [ 12.237572] __device_attach_driver from bus_for_each_drv+0x90/0xe4
> > > [ 12.243854] bus_for_each_drv from __device_attach+0xa8/0x1c8
> > > [ 12.249614] __device_attach from bus_probe_device+0x88/0x8c
> > > [ 12.255285] bus_probe_device from deferred_probe_work_func+0x8c/0xcc
> > > [ 12.261739] deferred_probe_work_func from process_one_work+0x154/0x2dc
> > > [ 12.268371] process_one_work from worker_thread+0x250/0x3f0
> > > [ 12.274043] worker_thread from kthread+0x12c/0x24c
> > > [ 12.278940] kthread from ret_from_fork+0x14/0x28
> > > [ 12.283660] Exception stack(0xd0be9fb0 to 0xd0be9ff8)
> > > [ 12.288720] 9fa0: 00000000 00000000
> > > 00000000 00000000
> > > [ 12.296906] 9fc0: 00000000 00000000 00000000 00000000 00000000 00000000
> > > 00000000 00000000
> > > [ 12.305089] 9fe0: 00000000 00000000 00000000 00000000 00000013 00000000
> > > [ 12.312050] ---[ end trace 0000000000000000 ]---
> > >
> > > #regzbot ^introduced: de04bb0089a96cc00d13b12cbf66a088befe3057
> > >
> > > Any advise?
> > >
> > > Hey Francesco!
> >
> > This mail reached my spam and I hadn't realised till today. Thanks for
> > bringing this to attention.
> >
> > Thinking out loud here: If we called dpi_probe() before allocating the
> > panel using devm_drm_panel_alloc()
> > then we would have the connector type. But dpi_probe() needs the panel to
> > be allocated....
>
> Reading the panel-simple.c code, the handling of the panel_dsi
> descriptor feels a bit hacky, and the recent change to
> devm_drm_panel_alloc() breaks it easily. Perhaps it would be cleaner to
> assess the whole descriptor before ding any allocation/init.
>
> You're right tat panel_dpi_probe() needs the panel, but it's only at the
> very end, to assign the descriptor:
>
> panel->desc = desc;
>
> I think a good fix would be to clean it up by having:
>
> * panel_dpi_probe() not take a panel pointer but rather returning a
> filled descriptor
> * panel_simple_probe() call panel_dpi_probe() early [before
> devm_drm_panel_alloc()] and get the filled descriptor
> * call devm_drm_panel_alloc() with that descriptor in the panel-dsi
> case, or with the good old descriptor otherwise
>
> As a good side effect, it would get rid of a case where
> devm_drm_panel_alloc() is called with a Unknown connector type.
>
> Anusha, does it look like a good plan?It is, and I'd even go one step further. Like you said, panel_dpi_probe kind of exists to allocate and initialize the panel descriptor, and is called on the descriptor being equal to the (uninitialized) panel_dpi global variable. We should also get rid of that hack, so do something like creating a function that returns the descriptor, and is indeed called first in panel_simple_probe. It first calls of_device_get_match_data(), and if there's no match, and if the device is compatible with panel-dpi, then it calls panel_dpi_probe (we should probably change that name too). That way, we can get rid of the panel_dpi variable entirely. Maxime
signature.asc
Description: PGP signature
