[PATCH] Fix IB va_start+ib_bytes range check on 32Bit systems

[Christian König]

Am 29.03.2017 um 16:54 schrieb Michel Dänzer:
> On 29/03/17 10:22 PM, Christian König wrote:
>> Am 29.03.2017 um 11:18 schrieb Jan Burgmeier:
>>> Signed-off-by: Jan Burgmeier <jan.burgmeier at unicon-software.com>
>>> ---
>>>    drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c | 4 +++-
>>>    1 file changed, 3 insertions(+), 1 deletion(-)
>>>
>>> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
>>> b/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
>>> index 99424cb8020b..583d22974e14 100644
>>> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
>>> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
>>> @@ -908,6 +908,7 @@ static int amdgpu_cs_ib_fill(struct amdgpu_device
>>> *adev,
>>>                struct amdgpu_bo *aobj = NULL;
>>>                uint64_t offset;
>>>                uint8_t *kptr;
>>> +            uint64_t it_last;
>>>                  m = amdgpu_cs_find_mapping(parser, chunk_ib->va_start,
>>>                               &aobj);
>>> @@ -916,8 +917,9 @@ static int amdgpu_cs_ib_fill(struct amdgpu_device
>>> *adev,
>>>                    return -EINVAL;
>>>                }
>>>    +            it_last = m->it.last;
>>>                if ((chunk_ib->va_start + chunk_ib->ib_bytes) >
>>> -                (m->it.last + 1) * AMDGPU_GPU_PAGE_SIZE) {
>>> +                (it_last + 1) * AMDGPU_GPU_PAGE_SIZE) {
>> Nice catch, but just adding a u64 case should do here as well. E.g:
>>
>> if ((chunk_ib->va_start + chunk_ib->ib_bytes) >
>>      (u64)(m->it.last + 1) * AMDGPU_GPU_PAGE_SIZE) {
> That won't work correctly if m->it.last == 0xffffffff ? Or is that not
> possible?
Hui, why? is it.last signed?

And even then m->it.last probably won't ever become 0xffffffff on a 
32bit system.

BTW: We need to fix using the 64bit R/B tree instead of the long sized 
tree for Vega10 here anyway.

Regards,
Christian.