[PATCH] drm/amdgpu: fix an underflow on non-4KB-page systems

[Christian König]

Am 29.03.21 um 20:08 schrieb Xi Ruoyao:
> On 2021-03-29 20:04 +0200, Christian König wrote:
>> Am 29.03.21 um 19:53 schrieb Xℹ Ruoyao:
>>> If the initial value of `num_entires` (calculated at line 1654) is not
>>> an integral multiple of `AMDGPU_GPU_PAGES_IN_CPU_PAGE`, in line 1681 a
>>> value greater than the initial value will be assigned to it.  That causes
>>> `start > last + 1` after line 1708.  Then in the next iteration an
>>> underflow happens at line 1654.  It causes message
>>>
>>>       *ERROR* Couldn't update BO_VA (-12)
>>>
>>> printed in kernel log, and GPU hanging.
>>>
>>> Fortify the criteria of the loop to fix this issue.
>> NAK the value of num_entries must always be a multiple of
>> AMDGPU_GPU_PAGES_IN_CPU_PAGE or otherwise we corrupt the page tables.
>>
>> How do you trigger that?
> Simply run "OpenGL area" from gtk3-demo (which just renders a triangle with GL)
> under Xorg, on MIPS64.  See the BugLink.

You need to identify the root cause of this, most likely start or last 
are not a multiple of AMDGPU_GPU_PAGES_IN_CPU_PAGE.

Christian.

>
>> Christian.
>>
>>> BugLink: https://gitlab.freedesktop.org/drm/amd/-/issues/1549
>>> Fixes: a39f2a8d7066 ("drm/amdgpu: nuke amdgpu_vm_bo_split_mapping v2")
>>> Reported-by: Xi Ruoyao <xry111 at mengyan1223.wang>
>>> Reported-by: Dan Horák <dan at danny.cz>
>>> Cc: stable at vger.kernel.org
>>> Signed-off-by: Xi Ruoyao <xry111 at mengyan1223.wang>
>>> ---
>>>    drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c | 2 +-
>>>    1 file changed, 1 insertion(+), 1 deletion(-)
>>>
>>> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
>>> b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
>>> index ad91c0c3c423..cee0cc9c8085 100644
>>> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
>>> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
>>> @@ -1707,7 +1707,7 @@ static int amdgpu_vm_bo_update_mapping(struct
>>> amdgpu_device *adev,
>>>                  }
>>>                  start = tmp;
>>>    
>>> -       } while (unlikely(start != last + 1));
>>> +       } while (unlikely(start < last + 1));
>>>    
>>>          r = vm->update_funcs->commit(&params, fence);
>>>    
>>>
>>> base-commit: a5e13c6df0e41702d2b2c77c8ad41677ebb065b3