Message ID | 20221212171958.82593-1-matthew.auld@intel.com (mailing list archive) |
---|---|
State | New, archived |
Headers | show |
Series | [1/2] drm/i915/migrate: fix corner case in CCS aux copying | expand |
On 12/12/2022 6:19 PM, Matthew Auld wrote: > In the case of lmem -> lmem transfers, which is currently only possible > with small-bar systems, we need to ensure we copy the CCS aux state > as-is, rather than nuke it. This should fix some nasty display > corruption sometimes seen on DG2 small-bar systems, when also using > DG2_RC_CCS_CC for the surface. > > Fixes: e3afc690188b ("drm/i915/display: consider DG2_RC_CCS_CC when migrating buffers") > Signed-off-by: Matthew Auld <matthew.auld@intel.com> > Cc: Ville Syrjälä <ville.syrjala@linux.intel.com> > Cc: Nirmoy Das <nirmoy.das@intel.com> > Cc: Andrzej Hajda <andrzej.hajda@intel.com> > Cc: Shuicheng Lin <shuicheng.lin@intel.com> Reviewed-by: Nirmoy Das <nirmoy.das@intel.com> > --- > drivers/gpu/drm/i915/gt/intel_migrate.c | 37 +++++++++++++++++++------ > 1 file changed, 29 insertions(+), 8 deletions(-) > > diff --git a/drivers/gpu/drm/i915/gt/intel_migrate.c b/drivers/gpu/drm/i915/gt/intel_migrate.c > index e08a739b7091..3f638f198796 100644 > --- a/drivers/gpu/drm/i915/gt/intel_migrate.c > +++ b/drivers/gpu/drm/i915/gt/intel_migrate.c > @@ -839,14 +839,35 @@ intel_context_migrate_copy(struct intel_context *ce, > if (err) > goto out_rq; > > - /* > - * While we can't always restore/manage the CCS state, > - * we still need to ensure we don't leak the CCS state > - * from the previous user, so make sure we overwrite it > - * with something. > - */ > - err = emit_copy_ccs(rq, dst_offset, INDIRECT_ACCESS, > - dst_offset, DIRECT_ACCESS, len); > + if (src_is_lmem) { > + /* > + * If the src is already in lmem, then we must > + * be doing an lmem -> lmem transfer, and so > + * should be safe to directly copy the CCS > + * state. In this case we have either > + * initialised the CCS aux state when first > + * clearing the pages (since it is already > + * allocated in lmem), or the user has > + * potentially populated it, in which case we > + * need to copy the CCS state as-is. > + */ > + err = emit_copy_ccs(rq, > + dst_offset, INDIRECT_ACCESS, > + src_offset, INDIRECT_ACCESS, > + len); > + } else { > + /* > + * While we can't always restore/manage the CCS > + * state, we still need to ensure we don't leak > + * the CCS state from the previous user, so make > + * sure we overwrite it with something. > + */ > + err = emit_copy_ccs(rq, > + dst_offset, INDIRECT_ACCESS, > + dst_offset, DIRECT_ACCESS, > + len); > + } > + > if (err) > goto out_rq; >
diff --git a/drivers/gpu/drm/i915/gt/intel_migrate.c b/drivers/gpu/drm/i915/gt/intel_migrate.c index e08a739b7091..3f638f198796 100644 --- a/drivers/gpu/drm/i915/gt/intel_migrate.c +++ b/drivers/gpu/drm/i915/gt/intel_migrate.c @@ -839,14 +839,35 @@ intel_context_migrate_copy(struct intel_context *ce, if (err) goto out_rq; - /* - * While we can't always restore/manage the CCS state, - * we still need to ensure we don't leak the CCS state - * from the previous user, so make sure we overwrite it - * with something. - */ - err = emit_copy_ccs(rq, dst_offset, INDIRECT_ACCESS, - dst_offset, DIRECT_ACCESS, len); + if (src_is_lmem) { + /* + * If the src is already in lmem, then we must + * be doing an lmem -> lmem transfer, and so + * should be safe to directly copy the CCS + * state. In this case we have either + * initialised the CCS aux state when first + * clearing the pages (since it is already + * allocated in lmem), or the user has + * potentially populated it, in which case we + * need to copy the CCS state as-is. + */ + err = emit_copy_ccs(rq, + dst_offset, INDIRECT_ACCESS, + src_offset, INDIRECT_ACCESS, + len); + } else { + /* + * While we can't always restore/manage the CCS + * state, we still need to ensure we don't leak + * the CCS state from the previous user, so make + * sure we overwrite it with something. + */ + err = emit_copy_ccs(rq, + dst_offset, INDIRECT_ACCESS, + dst_offset, DIRECT_ACCESS, + len); + } + if (err) goto out_rq;
In the case of lmem -> lmem transfers, which is currently only possible with small-bar systems, we need to ensure we copy the CCS aux state as-is, rather than nuke it. This should fix some nasty display corruption sometimes seen on DG2 small-bar systems, when also using DG2_RC_CCS_CC for the surface. Fixes: e3afc690188b ("drm/i915/display: consider DG2_RC_CCS_CC when migrating buffers") Signed-off-by: Matthew Auld <matthew.auld@intel.com> Cc: Ville Syrjälä <ville.syrjala@linux.intel.com> Cc: Nirmoy Das <nirmoy.das@intel.com> Cc: Andrzej Hajda <andrzej.hajda@intel.com> Cc: Shuicheng Lin <shuicheng.lin@intel.com> --- drivers/gpu/drm/i915/gt/intel_migrate.c | 37 +++++++++++++++++++------ 1 file changed, 29 insertions(+), 8 deletions(-)