Re: [PATCH 4/7] drm/i915: always pin lrc context for vgpu with Intel GVT-g

From: Zhiyuan Lv <zhiyuan.lv@intel.com>
To: Chris Wilson <chris@chris-wilson.co.uk>,
	intel-gfx@lists.freedesktop.org, igvt-g@lists.01.org, "Wang,
	Zhi A" <zhi.a.wang@intel.com>,
	"Tian, Kevin" <kevin.tian@intel.com>
Subject: Re: [PATCH 4/7] drm/i915: always pin lrc context for vgpu with Intel GVT-g
Date: Fri, 21 Aug 2015 14:13:12 +0800	[thread overview]
Message-ID: <20150821061312.GB22270@zlv-hp-dev> (raw)
In-Reply-To: <20150820091654.GD3226@zlv-hp-dev>

Hi Chris,

If we cannot always pin lr context into GGTT, the LRCA cannot be used
as a context identifier for us. Then we have to consider a proper
interface for i915 in VM to notify GVT-g device model.

The context creation might be OK. We can pin context first, then
notify the context creation, after that, unpin the context. In GVT-g,
we can get the context's guest physical addresses from GTT table, and
use that to identify a guest context.

But the destroy can be a problem. It does not make sense to pin
context and unpin that time right?

Do you have any suggestions/comments? Thanks in advance!

Regards,
-Zhiyuan

On Thu, Aug 20, 2015 at 05:16:54PM +0800, Zhiyuan Lv wrote:
> Hi Chris,
> 
> On Thu, Aug 20, 2015 at 09:36:00AM +0100, Chris Wilson wrote:
> > On Thu, Aug 20, 2015 at 03:45:21PM +0800, Zhiyuan Lv wrote:
> > > Intel GVT-g will perform EXECLIST context shadowing and ring buffer
> > > shadowing. The shadow copy is created when guest creates a context.
> > > If a context changes its LRCA address, the hypervisor is hard to know
> > > whether it is a new context or not. We always pin context objects to
> > > global GTT to make life easier.
> > 
> > Nak. Please explain why we need to workaround a bug in the host. We
> > cannot pin the context as that breaks userspace (e.g. synmark) who can
> > and will try to use more contexts than we have room.
> 
> This change is only for i915 running inside a VM. Host i915 driver's
> behavior will not be impacted.
> 
> The reason we want to pin context is that our hypervisor identifies
> guest contexts with their LRCA, and need LRCA unchanged during
> contexts' life cycle so that shadow copy of contexts can easily find
> their counterparts. If we cannot pin them, we have to consider more
> complicated shadow implementation ...
> 
> BTW Chris, are there many apps like synmark that may used up GGTT with
> contexts if they are all pinned? Thanks!
> 
> Regards,
> -Zhiyuan
> 
> > -Chris
> > 
> > -- 
> > Chris Wilson, Intel Open Source Technology Centre
_______________________________________________
Intel-gfx mailing list
Intel-gfx@lists.freedesktop.org
http://lists.freedesktop.org/mailman/listinfo/intel-gfx