On gen4+, we no longer need to setup a fence for tiled operations and so can program the GPU irrespective of the kernel tiled settings. However, this means that we then cannot access the object through a GTT mmapping (as we have no fence to do detiling) and nor can we use a WC mmap as the swizzle is unknown.
14b5ab3 sna: Enable use of GPU tiling even when SET_TILING fails
src/sna/kgem.c | 76 ++++++++++++++++++++++++++++++++++++++++++--------------
src/sna/kgem.h | 1 +
2 files changed, 59 insertions(+), 18 deletions(-)
Upstream: cgit.freedesktop.org