[Intel-gfx] [RFC] drm/i915: Android native sync support

[Tvrtko Ursulin]

On 01/22/2015 03:47 PM, Damien Lespiau wrote:
> On Thu, Jan 22, 2015 at 03:28:04PM +0000, Tvrtko Ursulin wrote:
>>
>> On 01/22/2015 02:04 PM, Damien Lespiau wrote:
>>> On Thu, Jan 22, 2015 at 01:41:48PM +0000, Tvrtko Ursulin wrote:
>>>>>> @@ -718,7 +718,7 @@ struct drm_i915_gem_execbuffer2 {
>>>>>>   #define I915_EXEC_CONSTANTS_REL_SURFACE (2<<6) /* gen4/5 only */
>>>>>>   	__u64 flags;
>>>>>>   	__u64 rsvd1; /* now used for context info */
>>>>>> -	__u64 rsvd2;
>>>>>> +	__u64 rsvd2; /* now used for fence fd */
>>>>> If we are going to use this slot for fence fd, may as well make it
>>>>> supply both before/after.
>>>>
>>>> Not sure what you mean by before/after?
>>>>
>>>> In the future it will take in the input fence fd and return the output fence
>>>> if that's what you mean.
>>>
>>> BTW, couldn't we take 32bits here and leave 32bits reserved?
>>
>> I guess so. Any ideas why the same wasn't done with rsvd1 - I see only
>> 32-bits are used for context id there?
>
> Nop, no idea, except may if someone was doing an assert(p->rsvd2 == 0);
> Spliting the field would break ABI.

I don't see it, it wouldn't assert unless the same, presumably old, 
userspace was also setting flags it doesn't know about?

Regards,

Tvrtko