[PATCH] drm/mediatek: Set private->all_drm_private[i]->drm to NULL if mtk_drm_bind returns err

[Guoqing Jiang]

Hi,

On 12/10/24 17:16, CK Hu (胡俊光) wrote:
> Hi, Guoqing:
>
> On Fri, 2024-12-06 at 21:55 +0800, Guoqing Jiang wrote:
>> External email : Please do not click links or open attachments until you have verified the sender or the content.
>>
>>
>> The pointer need to be set to NULL, otherwise KASAN complains about
>> use-after-free. Because in mtk_drm_bind, all private's drm are set
>> as follows.
>>
>> private->all_drm_private[i]->drm = drm;
>>
>> And drm will be released by drm_dev_put in case mtk_drm_kms_init returns
>> failure. However, the shutdown path still accesses the previous allocated
>> memory in drm_atomic_helper_shutdown.
>>
>> [   84.874820] watchdog: watchdog0: watchdog did not stop!
>> [   86.512054] ==================================================================
>> [   86.513162] BUG: KASAN: use-after-free in drm_atomic_helper_shutdown+0x33c/0x378
>> [   86.514258] Read of size 8 at addr ffff0000d46fc068 by task shutdown/1
>> [   86.515213]
>> [   86.515455] CPU: 1 UID: 0 PID: 1 Comm: shutdown Not tainted 6.13.0-rc1-mtk+gfa1a78e5d24b-dirty #55
>> [   86.516752] Hardware name: Unknown Unknown Product/Unknown Product, BIOS 2022.10 10/01/2022
>> [   86.517960] Call trace:
>> [   86.518333]  show_stack+0x20/0x38 (C)
>> [   86.518891]  dump_stack_lvl+0x90/0xd0
>> [   86.519443]  print_report+0xf8/0x5b0
>> [   86.519985]  kasan_report+0xb4/0x100
>> [   86.520526]  __asan_report_load8_noabort+0x20/0x30
>> [   86.521240]  drm_atomic_helper_shutdown+0x33c/0x378
>> [   86.521966]  mtk_drm_shutdown+0x54/0x80
>> [   86.522546]  platform_shutdown+0x64/0x90
>> [   86.523137]  device_shutdown+0x260/0x5b8
>> [   86.523728]  kernel_restart+0x78/0xf0
>> [   86.524282]  __do_sys_reboot+0x258/0x2f0
>> [   86.524871]  __arm64_sys_reboot+0x90/0xd8
>> [   86.525473]  invoke_syscall+0x74/0x268
>> [   86.526041]  el0_svc_common.constprop.0+0xb0/0x240
>> [   86.526751]  do_el0_svc+0x4c/0x70
>> [   86.527251]  el0_svc+0x4c/0xc0
>> [   86.527719]  el0t_64_sync_handler+0x144/0x168
>> [   86.528367]  el0t_64_sync+0x198/0x1a0
>> [   86.528920]
>> [   86.529157] The buggy address belongs to the physical page:
>> [   86.529972] page: refcount:0 mapcount:0 mapping:0000000000000000 index:0xffff0000d46fd4d0 pfn:0x1146fc
>> [   86.531319] flags: 0xbfffc0000000000(node=0|zone=2|lastcpupid=0xffff)
>> [   86.532267] raw: 0bfffc0000000000 0000000000000000 dead000000000122 0000000000000000
>> [   86.533390] raw: ffff0000d46fd4d0 0000000000000000 00000000ffffffff 0000000000000000
>> [   86.534511] page dumped because: kasan: bad access detected
>> [   86.535323]
>> [   86.535559] Memory state around the buggy address:
>> [   86.536265]  ffff0000d46fbf00: ff ff ff ff ff ff ff ff ff ff ff ff ff ff ff ff
>> [   86.537314]  ffff0000d46fbf80: ff ff ff ff ff ff ff ff ff ff ff ff ff ff ff ff
>> [   86.538363] >ffff0000d46fc000: ff ff ff ff ff ff ff ff ff ff ff ff ff ff ff ff
>> [   86.544733]                                                           ^
>> [   86.551057]  ffff0000d46fc080: ff ff ff ff ff ff ff ff ff ff ff ff ff ff ff ff
>> [   86.557510]  ffff0000d46fc100: ff ff ff ff ff ff ff ff ff ff ff ff ff ff ff ff
>> [   86.563928] ==================================================================
>> [   86.571093] Disabling lock debugging due to kernel taint
>> [   86.577642] Unable to handle kernel paging request at virtual address e0e9c0920000000b
>> [   86.581834] KASAN: maybe wild-memory-access in range [0x0752049000000058-0x075204900000005f]
>> ...
> Refer to [1] to add Fixes tag.
>
> [1] https://www.kernel.org/doc/html/v6.12/process/submitting-patches.html

Thanks for reminder! I think the problem is caused by commit 
1ef7ed48356c, so

Fixes: 1ef7ed48356c ("drm/mediatek: Modify mediatek-drm for mt8195 multi 
mmsys support")

And before send a new version with the tag, I will wait a few days for 
other comments.

Thanks,
Guoqing