Since the GGTT fault will overlap with the pwrite access, there is no implicit moment at which the kernel will automagically flush the backing store. Userspace has to be explicit in its domain control, or do itself. Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=112037 Signed-off-by: Chris Wilson <chris@xxxxxxxxxxxxxxxxxx> --- tests/i915/gem_persistent_relocs.c | 3 +++ 1 file changed, 3 insertions(+) diff --git a/tests/i915/gem_persistent_relocs.c b/tests/i915/gem_persistent_relocs.c index 2ab7091ad..20ee03027 100644 --- a/tests/i915/gem_persistent_relocs.c +++ b/tests/i915/gem_persistent_relocs.c @@ -224,8 +224,11 @@ static void do_test(int fd, bool faulting_reloc) relocs_bo_handle[i] = gem_create(fd, 4096); gem_write(fd, relocs_bo_handle[i], 0, reloc, sizeof(reloc)); + gtt_relocs_ptr[i] = gem_mmap__gtt(fd, relocs_bo_handle[i], 4096, PROT_READ | PROT_WRITE); + gem_set_domain(fd, relocs_bo_handle[i], + I915_GEM_DOMAIN_GTT, I915_GEM_DOMAIN_GTT); } /* repeat must be smaller than 4096/small_pitch */ -- 2.23.0 _______________________________________________ Intel-gfx mailing list Intel-gfx@xxxxxxxxxxxxxxxxxxxxx https://lists.freedesktop.org/mailman/listinfo/intel-gfx
