mirror of https://lore.kernel.org/lkml/
 help / color / mirror / Atom feed
From: Henry Zhang <henryzhang62@gmail.com>
To: mripard@kernel.org, mcanal@igalia.com
Cc: dave.stevenson@raspberrypi.com, kernel-list@raspberrypi.com,
	maarten.lankhorst@linux.intel.com, tzimmermann@suse.de,
	airlied@gmail.com, simona@ffwll.ch,
	dri-devel@lists.freedesktop.org, linux-kernel@vger.kernel.org,
	henryzhang62@gmail.com
Subject: Re: [PATCH] drm/vc4: dsi: Allocate the encoder per DRM device
Date: Wed, 18 Nov 2026 09:48:16 -0800	[thread overview]
Message-ID: <20261118174816.1382711-1-henryzhang62@gmail.com> (raw)
In-Reply-To: <av11cTfEuyFJW5h8@houat>

On Wed, Nov 18, 2026, Maxime Ripard wrote:
> On Tue, Nov 17, 2026 at 12:59:02PM -0800, Henry Zhang wrote:
> > The DSI encoder is embedded in struct vc4_dsi, which lives as long as
> > the DSI platform device
>
> That's not true.
>
> It's allocated with devm_drm_bridge_alloc() that will *not* free it in
> remove but give away its reference there.
>
> We have another reference taken in vc4_dsi_bind, that is only given up
> through a drm-managed action and thus will only be freed when the DRM
> device is destroyed.

You're right, thanks. struct vc4_dsi is refcounted through its bridge,
and the old DRM device keeps it alive. I'll fix that sentence. The
problem is not that struct vc4_dsi is freed too early, but that the
encoder embedded in it is used by two DRM devices at the same time.

> All vc4 components should be tied to the DRM device lifetime so that's
> not generally correct either. Something might have snuck back in though.

The encoder itself is DRM-managed: it is initialised with
drmm_encoder_init() and cleaned up when the DRM device is released.
But it is embedded in struct vc4_dsi, which outlives an unbind and is
shared by the next DRM device that binds the component. That part is
not tied to a single DRM device, since 6f0ca506c30e ("drm/vc4: dsi:
Switch to DRM-managed encoder initialization").

> Also, generally speaking, unloading panel drivers is unsupported at this
> point. It's a work in progress, but requires more changes to be safe.

Understood.

> This is surprising though. I'd like to see a detailed analysis of what's
> actually going on on mainline because it doesn't look right.

Maíra asked the same, so this answers both. It reproduces on
drm-misc-next (fb191ef62) on a Raspberry Pi 4, DSI1 in video mode, with
an out-of-tree DSI panel module and no downstream changes to vc4. The
compositor is stopped, a process holds /dev/dri/card1 open, the panel
module is unloaded and loaded again, then the held fd is closed:

1. Unloading the panel module detaches from the DSI host, which deletes
   the DSI component and unbinds vc4. The held fd keeps the old DRM
   device alive, with dsi->encoder still on its encoder list and the
   DSI bridge still attached to it.
2. Loading it again binds a new DRM device. vc4_dsi_bind() calls
   drmm_encoder_init() on the same dsi->encoder, which links it into
   the new device's encoder list.
3. drm_bridge_attach() returns -EBUSY, as the bridge still belongs to
   the old device, so the bind fails.
4. The failed bind releases the new DRM device. drm_encoder_cleanup()
   unlinks dsi->encoder and clears it, but the old device's encoder
   list still points to it.
5. Closing the held fd releases the old DRM device, which walks that
   list: list corruption, the WARN in drmm_encoder_alloc_release() and
   a NULL pointer dereference in drm_mode_config_cleanup().

Kprobes on drmm_encoder_init(), drm_bridge_attach() and
drm_encoder_cleanup() show it (condensed below). The old DRM device is
ffff000049249000, the new one ffff000041a79000, dsi->encoder is
ffff000044138000 and dsi->bridge ffff0000441380f0:

  223.656483 drmm_encoder_init    drm=ffff000041a79000 enc=ffff000044138000
  223.656520 drm_bridge_attach    enc=ffff000044138000 br=ffff0000441380f0
  223.656529 drm_bridge_attach    returned -16
  223.715147 drm_encoder_cleanup  enc=ffff000044138000 dev=ffff000041a79000
  (held fd closed)
  233.848952 drm_encoder_cleanup  enc=ffff0000409cca58 dev=ffff000049249000

and the first list_del report shows the cleared dsi->encoder.head
(ffff000044138008) still linked in the old device's list:

[  233.849039] list_del corruption. prev->next should be ffff0000409cca60, but was 0000000000000000. (prev=ffff000044138008)
[  233.849099] WARNING: lib/list_debug.c:62 at __list_del_entry_valid_or_report+0x128/0x160, CPU#0: python3/2080
[  233.849394] CPU: 0 UID: 0 PID: 2080 Comm: python3 Tainted: G         C O        7.4.0-rc2-dmn-test+ #1 PREEMPT(full) 
[  233.849424] Hardware name: Raspberry Pi 4 Model B Rev 1.5 (DT)
[  233.849448] pc : __list_del_entry_valid_or_report+0x128/0x160
[  233.849465] lr : __list_del_entry_valid_or_report+0x128/0x160
[  233.849734] Call trace:
[  233.849744]  __list_del_entry_valid_or_report+0x128/0x160 (P)
[  233.849764]  drm_encoder_cleanup+0xcc/0x150 [drm]
[  233.849950]  drmm_encoder_alloc_release+0x24/0x40 [drm]
[  233.850090]  drm_managed_release+0xb0/0x178 [drm]
[  233.850231]  drm_dev_put.part.0+0x98/0xd0 [drm]
[  233.850371]  drm_minor_release+0x20/0x50 [drm]
[  233.850511]  drm_release+0x110/0x148 [drm]
[  233.850651]  __fput+0xd4/0x300
[  233.850672]  fput_close_sync+0x48/0x110
[  233.850690]  filp_close_sync+0x2c/0x48
[  233.850706]  put_files_struct+0xd4/0x148
[  233.850723]  exit_files+0x4c/0x70
[  233.850738]  do_exit+0x25c/0xa20
[  233.850756]  do_group_exit+0x3c/0xa0
[  233.850772]  get_signal+0x920/0xa18
[  233.850791]  arch_do_signal_or_restart+0xa0/0x3f8
[  233.850808]  exit_to_user_mode_loop+0x100/0x188
[  233.850825]  el0_svc+0x1cc/0x1d0
[  233.850845]  el0t_64_sync_handler+0xa0/0xe8
[  233.850864]  el0t_64_sync+0x198/0x1a0
[  233.850879] ---[ end trace 0000000000000000 ]---
[  233.869067] WARNING: drivers/gpu/drm/drm_encoder.c:228 at drmm_encoder_alloc_release+0x30/0x40 [drm], CPU#0: python3/2080
[  233.869478] CPU: 0 UID: 0 PID: 2080 Comm: python3 Tainted: G        WC O        7.4.0-rc2-dmn-test+ #1 PREEMPT(full) 
[  233.869506] Hardware name: Raspberry Pi 4 Model B Rev 1.5 (DT)
[  233.869529] pc : drmm_encoder_alloc_release+0x30/0x40 [drm]
[  233.869668] lr : drm_managed_release+0xb0/0x178 [drm]
[  233.870051] Call trace:
[  233.870059]  drmm_encoder_alloc_release+0x30/0x40 [drm] (P)
[  233.870201]  drm_dev_put.part.0+0x98/0xd0 [drm]
[  233.870340]  drm_minor_release+0x20/0x50 [drm]
[  233.870479]  drm_release+0x110/0x148 [drm]
[  233.870619]  __fput+0xd4/0x300
[  233.870639]  fput_close_sync+0x48/0x110
[  233.870657]  filp_close_sync+0x2c/0x48
[  233.870672]  put_files_struct+0xd4/0x148
[  233.870688]  exit_files+0x4c/0x70
[  233.870703]  do_exit+0x25c/0xa20
[  233.870721]  do_group_exit+0x3c/0xa0
[  233.870737]  get_signal+0x920/0xa18
[  233.870755]  arch_do_signal_or_restart+0xa0/0x3f8
[  233.870772]  exit_to_user_mode_loop+0x100/0x188
[  233.870787]  el0_svc+0x1cc/0x1d0
[  233.870807]  el0t_64_sync_handler+0xa0/0xe8
[  233.870825]  el0t_64_sync+0x198/0x1a0
[  233.870839] ---[ end trace 0000000000000000 ]---
[two more list_del corruption reports with the same call trace cut]
[  233.875117] Unable to handle kernel NULL pointer dereference at virtual address 0000000000000008
[  233.875331] Internal error: Oops: 0000000096000004 [#1]  SMP
[  233.875631] CPU: 0 UID: 0 PID: 2080 Comm: python3 Tainted: G        WC O        7.4.0-rc2-dmn-test+ #1 PREEMPT(full) 
[  233.875679] Hardware name: Raspberry Pi 4 Model B Rev 1.5 (DT)
[  233.875717] pc : drm_mode_config_cleanup+0x58/0x320 [drm]
[  233.875875] lr : drm_mode_config_init_release+0x18/0x30 [drm]
[  233.876371] Call trace:
[  233.876383]  drm_mode_config_cleanup+0x58/0x320 [drm] (P)
[  233.876539]  drm_mode_config_init_release+0x18/0x30 [drm]
[  233.876692]  drm_managed_release+0xb0/0x178 [drm]
[  233.876842]  drm_dev_put.part.0+0x98/0xd0 [drm]
[  233.876998]  drm_minor_release+0x20/0x50 [drm]
[  233.877149]  drm_release+0x110/0x148 [drm]
[  233.877299]  __fput+0xd4/0x300
[  233.877330]  fput_close_sync+0x48/0x110
[  233.877355]  filp_close_sync+0x2c/0x48
[  233.877377]  put_files_struct+0xd4/0x148
[  233.877400]  exit_files+0x4c/0x70
[  233.877422]  do_exit+0x25c/0xa20
[  233.877446]  do_group_exit+0x3c/0xa0
[  233.877469]  get_signal+0x920/0xa18
[  233.877494]  arch_do_signal_or_restart+0xa0/0x3f8
[  233.877519]  exit_to_user_mode_loop+0x100/0x188
[  233.877542]  el0_svc+0x1cc/0x1d0
[  233.877567]  el0t_64_sync_handler+0xa0/0xe8
[  233.877593]  el0t_64_sync+0x198/0x1a0
[  233.877620] Code: eb0102bf 54000160 d1002294 f9404001 (f9400421) 
[  233.877643] ---[ end trace 0000000000000000 ]---

Additional sanitized trace excerpts are available on request.

With the patch, the same test on the same kernel still fails the bind
with -EBUSY, but closing the held fd gives no warning or oops, and
loading the panel module again afterwards brings the display back. This
was one focused run per build, not a stress test, and the patch does
not make the rebind succeed while the old device holds the bridge.

Would you prefer a v2 with the corrected description and the mainline
trace, or should this go with the panel unloading work?

Thanks,
Henry

  reply	other threads:[~2026-11-18 17:48 UTC|newest]

Thread overview: 4+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2026-11-17 20:59 Henry Zhang
2026-11-18 10:09 ` Maxime Ripard
2026-11-18 17:48   ` Henry Zhang [this message]
2026-11-18 14:00 ` Maíra Canal

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=20261118174816.1382711-1-henryzhang62@gmail.com \
    --to=henryzhang62@gmail.com \
    --cc=airlied@gmail.com \
    --cc=dave.stevenson@raspberrypi.com \
    --cc=dri-devel@lists.freedesktop.org \
    --cc=kernel-list@raspberrypi.com \
    --cc=linux-kernel@vger.kernel.org \
    --cc=maarten.lankhorst@linux.intel.com \
    --cc=mcanal@igalia.com \
    --cc=mripard@kernel.org \
    --cc=simona@ffwll.ch \
    --cc=tzimmermann@suse.de \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox

all inboxes | Powered by JetHome®