[PATCH 3/3] drm/xe: Allow scratch page under fault mode for certain platform

Matthew Brost matthew.brost at intel.com
Wed Jan 29 16:41:07 UTC 2025


On Wed, Jan 29, 2025 at 09:52:54AM +0100, Thomas Hellström wrote:
> On Tue, 2025-01-28 at 17:21 -0500, Oak Zeng wrote:
> > Normally scratch page is not allowed when a vm is operate under page
> > fault mode, i.e., in the existing codes,
> > DRM_XE_VM_CREATE_FLAG_SCRATCH_PAGE
> > and DRM_XE_VM_CREATE_FLAG_FAULT_MODE are mutual exclusive. The reason
> > is fault mode relies on recoverable page to work, while scratch page
> > can mute recoverable page fault.
> > 
> > On xe2 and xe3, HW prefetching can cause page fault interrupt. Due to
> > page fault interrupt overhead (i.e., need Guc and KMD involved to fix
> > the page fault), HW prefetching can be slowed by many orders of
> > magnitude.
> 
> The problem on xe2 and xe3 as I understand it is not the overhead but
> the fact that OOB prefetching causes pagefaults and we can't resolve
> those.
> 
> > 
> > Fix this problem by allowing scratch page under fault mode for xe2
> > and xe3.
> > With scratch page in place, HW prefetching could always hit scratch
> > page
> > instead of causing interrupt.
> > 
> > Signed-off-by: Oak Zeng <oak.zeng at intel.com>
> > ---
> >  drivers/gpu/drm/xe/xe_vm.c | 9 +++++----
> >  1 file changed, 5 insertions(+), 4 deletions(-)
> > 
> > diff --git a/drivers/gpu/drm/xe/xe_vm.c b/drivers/gpu/drm/xe/xe_vm.c
> > index 196d347c6ac0..3346f88f284a 100644
> > --- a/drivers/gpu/drm/xe/xe_vm.c
> > +++ b/drivers/gpu/drm/xe/xe_vm.c
> > @@ -1732,6 +1732,11 @@ int xe_vm_create_ioctl(struct drm_device *dev,
> > void *data,
> >  	if (XE_IOCTL_DBG(xe, args->extensions))
> >  		return -EINVAL;
> >  
> > +	if (XE_IOCTL_DBG(xe, args->flags &
> > DRM_XE_VM_CREATE_FLAG_SCRATCH_PAGE &&
> > +			 args->flags &
> > DRM_XE_VM_CREATE_FLAG_FAULT_MODE &&
> > +			 !(IS_LUNARLAKE(xe) || IS_BATTLEMAGE(xe) ||
> > IS_PANTHERLAKE(xe))))
> > +		return -EINVAL;
> > +
> 
> Could we perhaps add a bit in the xe->info structure?
> 

+1. This seems like a good idea.

Matt

> /Thomas
> 
> 
> >  	if (XE_WA(xe_root_mmio_gt(xe), 14016763929))
> >  		args->flags |= DRM_XE_VM_CREATE_FLAG_SCRATCH_PAGE;
> >  
> > @@ -1745,10 +1750,6 @@ int xe_vm_create_ioctl(struct drm_device *dev,
> > void *data,
> >  	if (XE_IOCTL_DBG(xe, args->flags &
> > ~ALL_DRM_XE_VM_CREATE_FLAGS))
> >  		return -EINVAL;
> >  
> > -	if (XE_IOCTL_DBG(xe, args->flags &
> > DRM_XE_VM_CREATE_FLAG_SCRATCH_PAGE &&
> > -			 args->flags &
> > DRM_XE_VM_CREATE_FLAG_FAULT_MODE))
> > -		return -EINVAL;
> > -
> >  	if (XE_IOCTL_DBG(xe, !(args->flags &
> > DRM_XE_VM_CREATE_FLAG_LR_MODE) &&
> >  			 args->flags &
> > DRM_XE_VM_CREATE_FLAG_FAULT_MODE))
> >  		return -EINVAL;
> 


More information about the Intel-xe mailing list