[Intel-gfx] [RFCv3 10/15] drm/i915: update PDPs by condition when submit the LRC context

[Zhi Wang]

Hi Chris:
     Your idea is good. :) We could emit PDP upgrade LRIs like i915 
before emit GVT worload under GVT context, then we can reuse the 
pd_dirty_ring bitmap.

But I have to expose intel_logical_ring_emit_pdps() function for GVT-g. 
Is it acceptable?

Thanks,
Zhi.

On 03/11/16 20:56, Wang, Zhi A wrote:
> OK. I will see. :) Thanks for the comments.
>
> -----Original Message-----
> From: Chris Wilson [mailto:chris at chris-wilson.co.uk]
> Sent: Friday, March 11, 2016 7:28 PM
> To: Wang, Zhi A
> Cc: intel-gfx at lists.freedesktop.org; igvt-g at lists.01.org; Tian, Kevin; Lv, Zhiyuan; Niu, Bing; Song, Jike; daniel.vetter at ffwll.ch; Cowperthwaite, David J; joonas.lahtinen at linux.intel.com
> Subject: Re: [RFCv3 10/15] drm/i915: update PDPs by condition when submit the LRC context
>
> On Fri, Mar 11, 2016 at 06:59:41PM +0800, Zhi Wang wrote:
>> Previously the PDPs inside the ring context will be updated
>>
>> - When populating a LRC context
>> - Before submitting a LRC context (only for 32 bit PPGTT, as the amount
>> of used PDPs may be changed during PPGTT page table grow)
>>
>> Under GVT-g, each VM owns a GVT context used as the shadow context. When
>> guest submits a context, GVT-g will load guest context into the GVT
>> context, and submit this context to i915 GEM submission system.
>>
>> So this GVT context could be used by different guest context, and the
>> PPGTT root pointer in guest contexts might be different as well, if guest
>> is using full PPGTT mode.
>>
>> In current i915, the root pointer in a LRC context will only be updated
>> during the LRC context creation.
>>
>> This patch postpones the root pointer upgrade to the time of submission,
>> which will give GVT-g a chance to reload a new PPGTT page table root
>> pointer into an existing GVT context.
>
> Not explained is why you cannot use the existing dirty flags.
> -Chris
>