[PATCH 1/3] drm/amdgpu: Add flag for allocating memory for sensitive data

Kuehling, Felix Felix.Kuehling at amd.com
Tue Jul 9 19:00:46 UTC 2019


On 2019-07-09 6:34 a.m., Michel Dänzer wrote:
> On 2019-07-09 7:32 a.m., Kuehling, Felix wrote:
>> This memory allocation flag will be used to indicate BOs containing
>> sensitive data that should not be leaked to other processes.
>>
>> Signed-off-by: Felix Kuehling <Felix.Kuehling at amd.com>
>> ---
>>   include/uapi/drm/amdgpu_drm.h | 4 ++++
>>   1 file changed, 4 insertions(+)
>>
>> diff --git a/include/uapi/drm/amdgpu_drm.h b/include/uapi/drm/amdgpu_drm.h
>> index 61870478bc9c..58659c28c26e 100644
>> --- a/include/uapi/drm/amdgpu_drm.h
>> +++ b/include/uapi/drm/amdgpu_drm.h
>> @@ -131,6 +131,10 @@ extern "C" {
>>    * for the second page onward should be set to NC.
>>    */
>>   #define AMDGPU_GEM_CREATE_MQD_GFX9		(1 << 8)
>> +/* Flag that BO may contain sensitive data that must be cleared before
>> + * releasing the memory
>> + */
>> +#define AMDGPU_GEM_CREATE_VRAM_SENSITIVE	(1 << 9)
>>   
>>   struct drm_amdgpu_gem_create_in  {
>>   	/** the requested memory size */
>>
> This flag essentially means "Please don't leak my BO contents".
> Similarly, AMDGPU_GEM_CREATE_VRAM_CLEARED essentially means "Please
> don't let me see previous memory contents".
>
> I'd argue that neither flag should really be needed; BO contents
> shouldn't be leaked by default.

My conclusion from previous discussions was that CREATE_VRAM_CLEARED has 
no security implications. It's basically completely ineffective as a 
security measure. It's more a convenience feature. Therefore I think it 
still has a place as that.

I'd agree on principle that data shouldn't be leaked by default, but it 
has been the default for a long time. My impression was that graphics 
guys cared more about performance than security. So changing the default 
may be a hard sell. On the compute side we already took a big 
performance hit by clearing all our VRAM, so this change would be an 
improvement for us. Therefore I think it still makes sense to let the 
application choose.

Regards,
   Felix


>
>


More information about the amd-gfx mailing list