Re: [PATCH v7 1/6] drm/amdgpu: Fix init ordering in amdgpu_vram_mgr_init()

On 7/3/2026 6:41 PM, Thomas Hellström wrote:
> On Fri, 2026-07-03 at 15:08 +0200, Christian König wrote:
>> Arun please take a look at this.
>>
>> Thanks,
>> Christian.
> FWIW Sashiko claims there is yet another pre-existing bug WRT ordering
> here, but since the fix wasn't needed for the rest of the series, I
> focused on this one.
Thanks for pointing that out. I have sent a fix for the pre-existing bug 
reported by Sashiko as a separate patch.
https://patchwork.freedesktop.org/patch/741782/

Thanks,
Arun.
>
> Thanks,
> Thomas
>
>
>> On 7/3/26 15:05, Thomas Hellström wrote:
>>> drmm_cgroup_register_region() is called before INIT_LIST_HEAD() and
>>> gpu_buddy_init() in amdgpu_vram_mgr_init(). If it fails, the
>>> function
>>> returns early and bypasses those initializations.
>>>
>>> Since adev->mman.initialized is set to true before
>>> amdgpu_vram_mgr_init()
>>> is called, a failure triggers amdgpu_ttm_fini(), which calls
>>> amdgpu_vram_mgr_fini(), which then:
>>>
>>>   - Calls list_for_each_entry_safe() on reservations_pending and
>>>     reserved_pages, whose list_head::next pointers are zero-
>>> initialized
>>>     (NULL). The loop does not recognize them as empty and
>>> dereferences NULL.
>>>
>>>   - Calls gpu_buddy_fini(), which iterates free_trees[]
>>> unconditionally
>>>     via for_each_free_tree(). Since mm->free_trees is NULL
>>>     (never allocated), this dereferences NULL.
>>>
>>> Both result in a kernel panic on the module load error path.
>>>
>>> Fix by moving drmm_cgroup_register_region() to after the list and
>>> buddy
>>> allocator are fully initialized, so the teardown path is safe to
>>> run.
>>>
>>> Reported-by: Sashiko-bot <[email protected]>
>>> Closes:
>>> https://sashiko.dev/#/patchset/[email protected]?part=4
>>> Fixes: 2b624a2c1865 ("drm/ttm: Handle cgroup based eviction in
>>> TTM")
>>> Cc: Friedrich Vock <[email protected]>
>>> Cc: Maarten Lankhorst <[email protected]>
>>> Cc: Tejun Heo <[email protected]>
>>> Cc: Maxime Ripard <[email protected]>
>>> Cc: Christian König <[email protected]>
>>> Cc: Alex Deucher <[email protected]>
>>> Cc: [email protected]
>>> Cc: [email protected]
>>> Cc: <[email protected]> # v6.14+
>>> Assisted-by: GitHub_Copilot:claude-sonnet-4.6
>>> Signed-off-by: Thomas Hellström <[email protected]>
>>> ---
>>>   drivers/gpu/drm/amd/amdgpu/amdgpu_vram_mgr.c | 7 ++++---
>>>   1 file changed, 4 insertions(+), 3 deletions(-)
>>>
>>> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_vram_mgr.c
>>> b/drivers/gpu/drm/amd/amdgpu/amdgpu_vram_mgr.c
>>> index 2a241a5b12c4..ac3f71d77140 100644
>>> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_vram_mgr.c
>>> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_vram_mgr.c
>>> @@ -918,9 +918,6 @@ int amdgpu_vram_mgr_init(struct amdgpu_device
>>> *adev)
>>>   	struct ttm_resource_manager *man = &mgr->manager;
>>>   	int err;
>>>   
>>> -	man->cg = drmm_cgroup_register_region(adev_to_drm(adev),
>>> "vram", adev->gmc.real_vram_size);
>>> -	if (IS_ERR(man->cg))
>>> -		return PTR_ERR(man->cg);
>>>   	ttm_resource_manager_init(man, &adev->mman.bdev,
>>>   				  adev->gmc.real_vram_size);
>>>   
>>> @@ -935,6 +932,10 @@ int amdgpu_vram_mgr_init(struct amdgpu_device
>>> *adev)
>>>   	if (err)
>>>   		return err;
>>>   
>>> +	man->cg = drmm_cgroup_register_region(adev_to_drm(adev),
>>> "vram", adev->gmc.real_vram_size);
>>> +	if (IS_ERR(man->cg))
>>> +		return PTR_ERR(man->cg);
>>> +
>>>   	ttm_set_driver_manager(&adev->mman.bdev, TTM_PL_VRAM,
>>> &mgr->manager);
>>>   	ttm_resource_manager_set_used(man, true);
>>>   	return 0;