On 8/21/19 8:20 AM, Daniel Vetter wrote:
On Wed, Aug 21, 2019 at 3:55 PM Ville Syrjälä
<ville.syrjala@xxxxxxxxxxxxxxx> wrote:
On Tue, Aug 20, 2019 at 01:57:44PM -0700, Daniele Ceraolo Spurio wrote:
On 8/20/19 12:54 PM, Daniel Vetter wrote:
The cpu (de)tiler hw is gone, this stopped being useful. Plus it never
supported any of the fancy new tiling formats, which means userspace
also stopped using the magic side-channel this provides.
This would totally break a lot of the igts, but they're already broken
for the same reasons as userspace on gen12 would be.
v2: Look at ggtt->num_fences instead, that also avoids the need for a
comment (Chris). This also means that gen12 support really needs to
make sure num_fences is set to 0. There is a patch for that, but it
checks for HAS_MAPPABLE_APERTURE, which I'm not sure is the right
thing really. Adding relevant people.
We'd obviously need to make that setting for all gen12+, because TGL
does have mappable aperture.
Apart from the tiling ioctl, the only place I see where we set tiling is
intel_alloc_initial_plane_obj(), can the users of that object handle the
lack of fences gracefully?
Gen4+ display engine has its own tiling controls and doesn't care about
fences. So we should be able to leave the obj tiling set to NONE.
Hmm. Actually I think we should reject tiled framebuffers in the BIOS
fb takeover because fbdev needs a linear view for the memory. No can
do without the fence.
Yeah I think this is just more fallout from "no more fences in the hw".
-Daniel
Is anyone looking at implementing this (fence = 0 and reject tiled FBs)
for TGL? I can pick it up if no one is.
Daniele
When I wrote the num_fences=0 patch I was
expecting display to be unavailable, so I didn't really look at that
part of the code.
It'd also be nice to be more explicit with fencing since we seem to
often call i915_vma_pin_iomap, which implicitly applies a fence if
needed, on objects that can't be tiled or have had a fence assigned a
few lines before. This is more a nice to have tough, possibly together
with a split of the "mappable" and "fenceable" attributes of the vma.
Daniele
Cc: Daniele Ceraolo Spurio <daniele.ceraolospurio@xxxxxxxxx>
Cc: Stuart Summers <stuart.summers@xxxxxxxxx>
Cc: Matthew Auld <matthew.auld@xxxxxxxxx>
Cc: Kenneth Graunke <kenneth@xxxxxxxxxxxxx>
Cc: Jason Ekstrand <jason@xxxxxxxxxxxxxx>
Cc: Chris Wilson <chris@xxxxxxxxxxxxxxxxxx>
Cc: Lucas De Marchi <lucas.demarchi@xxxxxxxxx>
Signed-off-by: Daniel Vetter <daniel.vetter@xxxxxxxxx>
---
drivers/gpu/drm/i915/gem/i915_gem_tiling.c | 7 +++++++
1 file changed, 7 insertions(+)
diff --git a/drivers/gpu/drm/i915/gem/i915_gem_tiling.c b/drivers/gpu/drm/i915/gem/i915_gem_tiling.c
index ca0c2f451742..e5d1ae8d4dba 100644
--- a/drivers/gpu/drm/i915/gem/i915_gem_tiling.c
+++ b/drivers/gpu/drm/i915/gem/i915_gem_tiling.c
@@ -313,10 +313,14 @@ int
i915_gem_set_tiling_ioctl(struct drm_device *dev, void *data,
struct drm_file *file)
{
+ struct drm_i915_private *dev_priv = to_i915(dev);
struct drm_i915_gem_set_tiling *args = data;
struct drm_i915_gem_object *obj;
int err;
+ if (!dev_priv->ggtt.num_fences)
+ return -EOPNOTSUPP;
+
obj = i915_gem_object_lookup(file, args->handle);
if (!obj)
return -ENOENT;
@@ -402,6 +406,9 @@ i915_gem_get_tiling_ioctl(struct drm_device *dev, void *data,
struct drm_i915_gem_object *obj;
int err = -ENOENT;
+ if (!dev_priv->ggtt.num_fences)
+ return -EOPNOTSUPP;
+
rcu_read_lock();
obj = i915_gem_object_lookup_rcu(file, args->handle);
if (obj) {
_______________________________________________
Intel-gfx mailing list
Intel-gfx@xxxxxxxxxxxxxxxxxxxxx
https://lists.freedesktop.org/mailman/listinfo/intel-gfx
--
Ville Syrjälä
Intel
_______________________________________________
Intel-gfx mailing list
Intel-gfx@xxxxxxxxxxxxxxxxxxxxx
https://lists.freedesktop.org/mailman/listinfo/intel-gfx
