[PATCH] drm/meson: fix resource cleanup in meson_drv_bind_master() on error
Anand Moon
linux.amoon at gmail.com
Thu Apr 10 13:46:16 UTC 2025
Hi Martin,
On Thu, 10 Apr 2025 at 03:30, Martin Blumenstingl
<martin.blumenstingl at googlemail.com> wrote:
>
> meson_drv_bind_master() does not free resources in the order they are
> allocated. This can lead to crashes such as:
> Unable to handle kernel NULL pointer dereference at virtual address 00000000000000c8
> [...]
> Hardware name: Beelink GT-King Pro (DT)
> pstate: 60000005 (nZCv daif -PAN -UAO -TCO -DIT -SSBS BTYPE=--)
> pc : meson_dw_hdmi_unbind+0x10/0x24 [meson_dw_hdmi]
> lr : component_unbind+0x38/0x60
> [...]
> Call trace:
> meson_dw_hdmi_unbind+0x10/0x24 [meson_dw_hdmi]
> component_unbind+0x38/0x60
> component_unbind_all+0xb8/0xc4
> meson_drv_bind_master+0x1ec/0x514 [meson_drm]
> meson_drv_bind+0x14/0x20 [meson_drm]
> try_to_bring_up_aggregate_device+0xa8/0x160
> __component_add+0xb8/0x1a8
> component_add+0x14/0x20
> meson_dw_hdmi_probe+0x1c/0x28 [meson_dw_hdmi]
> platform_probe+0x68/0xdc
> really_probe+0xc0/0x39c
> __driver_probe_device+0x7c/0x14c
> driver_probe_device+0x3c/0x120
> __driver_attach+0xc4/0x200
> bus_for_each_dev+0x78/0xd8
> driver_attach+0x24/0x30
> bus_add_driver+0x110/0x240
> driver_register+0x68/0x124
> __platform_driver_register+0x24/0x30
> meson_dw_hdmi_platform_driver_init+0x20/0x1000 [meson_dw_hdmi]
> do_one_initcall+0x50/0x1bc
> do_init_module+0x54/0x1fc
> load_module+0x788/0x884
> init_module_from_file+0x88/0xd4
> __arm64_sys_finit_module+0x248/0x340
> invoke_syscall+0x48/0x104
> el0_svc_common.constprop.0+0x40/0xe0
> do_el0_svc+0x1c/0x28
> el0_svc+0x30/0xcc
> el0t_64_sync_handler+0x120/0x12c
> el0t_64_sync+0x190/0x194
>
> Clean up resources in the error path in the same order and under the
> same conditions as they were allocated to fix this.
>
> Reported-by: Furkan Kardame <f.kardame at manjaro.org>
> Signed-off-by: Martin Blumenstingl <martin.blumenstingl at googlemail.com>
> ---
> This issue was reported off-list so I'm unable to provide a link to the
> report.
>
> I'm not sure which Fixes tag fits best. My preference so far is
> Fixes: 6a044642988b ("drm/meson: fix unbind path if HDMI fails to bind")
> but I'll happily take other suggestions as well.
>
Thanks for this Fix.
Reviewed-by: Anand Moon <linux.amoon at gmail.com>
Tested-by: Anand Moon <linux.amoon at gmail.com>
I have tested with the following script on AML-S905X-CC
---------
#! /bin/bash
set +x
cd /sys/bus/platform/drivers/meson-drm/
for i in $(seq 1 10); do
echo "========================" $i
echo d0100000.vpu > unbind
/usr/bin/sleep 1
echo d0100000.vpu > bind
/usr/bin/sleep 1
done
Thanks
-Anand
More information about the dri-devel
mailing list