[PATCH] drm/i915/gvt: enter failsafe mode when guest requires more resources

[Tian, Kevin]

> From: He, Min
> Sent: Wednesday, February 15, 2017 4:02 PM
> 
> 
> 在 2/15/2017 3:51 PM, Tian, Kevin 写道:
> >> From: Min He
> >> Sent: Wednesday, February 15, 2017 3:34 PM
> >>
> >> Windows guest will notitfy GVTg to request more resources through g2v
> >> interface, when its resources are not enough.
> >> This patch is to handle this case and let vgpu enter failsafe mode to
> >> avoid to many error messages.
> >>
> >> Signed-off-by: Min He <min.he at intel.com>
> >> Signed-off-by: Pei Zhang <pei.zhang at intel.com>
> >> ---
> >>   drivers/gpu/drm/i915/gvt/handlers.c | 15 +++++++++++++++
> >>   drivers/gpu/drm/i915/i915_pvinfo.h  | 17 +++++++++++++++--
> >>   2 files changed, 30 insertions(+), 2 deletions(-)
> >>
> >> diff --git a/drivers/gpu/drm/i915/gvt/handlers.c
> b/drivers/gpu/drm/i915/gvt/handlers.c
> >> index 6089a1e..50a1850 100644
> >> --- a/drivers/gpu/drm/i915/gvt/handlers.c
> >> +++ b/drivers/gpu/drm/i915/gvt/handlers.c
> >> @@ -1090,6 +1090,21 @@ static int pvinfo_mmio_write(struct intel_vgpu *vgpu,
> >> unsigned int offset,
> >>   	case _vgtif_reg(execlist_context_descriptor_lo):
> >>   	case _vgtif_reg(execlist_context_descriptor_hi):
> >>   		break;
> >> +	case _vgtif_reg(min_low_gmadr):
> >> +		gvt_err("Guest requires %d aperture size!\n", data);
> >> +		break;
> >> +	case _vgtif_reg(min_high_gmadr):
> >> +		gvt_err("Guest requires %d hidden gm size!\n", data);
> >> +		break;
> >> +	case _vgtif_reg(min_fence_num):
> >> +		gvt_err("Guest requires %d fence size!\n", data);
> >> +		break;
> > above information is not an obvious error message. at least sth like:
> >
> > Guest requests more fences which is not supported by this type.
> Ok. Will modify these messages in V2.
> 
> >
> >> +	case _vgtif_reg(rs_insufficient):
> >> +		pr_err("***Guest resource is not enough**********\n");
> >> +		pr_err("***Please try another vgpu type**********\n");
> >> +		pr_err("***Entering failsafe mode****************\n");
> >> +		vgpu->failsafe = true;
> >> +		break;
> >>   	default:
> >>   		gvt_err("invalid pvinfo write offset %x bytes %x data %x\n",
> >>   				offset, bytes, data);
> >> diff --git a/drivers/gpu/drm/i915/i915_pvinfo.h
> b/drivers/gpu/drm/i915/i915_pvinfo.h
> >> index c0cb297..98a0f18 100644
> >> --- a/drivers/gpu/drm/i915/i915_pvinfo.h
> >> +++ b/drivers/gpu/drm/i915/i915_pvinfo.h
> >> @@ -86,8 +86,21 @@ struct vgt_if {
> >>   	 */
> >>   	u32 rsv4;
> >>   	u32 display_ready;	/* ready for display owner switch */
> >> -
> >> -	u32 rsv5[4];
> >> +	/*
> >> +	 * driver reported status/error code
> >> +	 *     0: if the avail_rs is sufficient to driver
> >> +	 *  Bit 2,1,0 set indicating
> >> +	 *       Insufficient low_gmadr, high_gmadr, fence resources.
> >> +	 *  Other bits are reserved.
> >> +	 */
> >> +	uint32_t  rs_insufficient;
> >> +	/*
> >> +	 * The driver is required to update the following field with minimal
> >> +	 * required resource size.
> >> +	 */
> >> +	uint32_t  min_low_gmadr;
> >> +	uint32_t  min_high_gmadr;
> >> +	uint32_t  min_fence_num;
> > If such operation is not supported, instead of explicitly defining
> > them and then fail, it's cleaner to keep original definition while
> > simply fallback to failsafe if any reserved offset is written.
> This interface was defined in production branch, but somehow it's
> removed during
> upstream. So I added them back into pvinfo structure.
> 

No need of doing so. If required in the future, you can introduce
them at that time. For now as I commented above, the policy 
is more about accessing to reserved fields...

Thanks
Kevin