Re: [PATCH 0/3] SVM compression fixes

[Matthew Brost <matthew.brost@intel.com>]

On Tue, Jul 15, 2025 at 01:59:16PM +0100, Matthew Auld wrote:
> On 14/07/2025 18:33, Matthew Brost wrote:
> > Compression is broken for SVM and for devcoredump when reading out BOs.
> 
> For devcoredump case is this the bo_read() call? Does that mean we would
> still run into trouble for < 16K buffers and then also igpu in general?
> 

Yes, we will need some more follow up patches I think.

Let's start here with this series and then fixup remaining devcoredump cases?

Matt

> > 
> > Marking the first two patches as fixes. One option could be to drop
> > patch 1 and proceed with just the last two patches as fixes, if we
> > decide we're confident in SVM with compression.
> > 
> > Tested with [1]. The append memory copy compression sections fail on
> > drm-tip and pass after this series.
> > 
> > Matt
> > 
> > [1] https://patchwork.freedesktop.org/series/151581/
> > 
> > Matthew Brost (3):
> >    drm/xe: Disable compression on SVM
> >    drm/xe: Update xe_migrate_vram to support compression
> >    drm/xe: Enable compressed memory for SVM
> > 
> >   drivers/gpu/drm/xe/xe_migrate.c | 37 ++++++++++++++++++++++++++-------
> >   1 file changed, 30 insertions(+), 7 deletions(-)
> > 
>