On 8/16/19 3:13 PM, Chris Wilson wrote:
Quoting Daniele Ceraolo Spurio (2019-08-16 22:50:43)
On 8/16/19 12:17 PM, Chris Wilson wrote:
+static int live_context_size(void *arg)
+{
+ /*
+ * Hide the old default state -- we lie about the context size
+ * and get confused when the default state is smaller than
+ * expected. For our do nothing request, inheriting the
+ * active state is sufficient, we are only checking that we
+ * don't use more than we planned.
+ */
+ saved.state = fetch_and_zero(&engine->default_state);
+ saved.pinned = fetch_and_zero(&engine->pinned_default_state);
+
+ engine->context_size += I915_GTT_PAGE_SIZE;
if CONFIG_DRM_I915_DEBUG_GEM is set we already bump the size inside the
context_alloc(), do we need to bump it again here?
No, it comes out in the wash as we apply the same redzone twice. At least,
adding and checking a second page after what we believe to be the end of
the context image does not help sensitivity (so makes a worse test imo).
The benefit of this selftest is that we check all submission modes, and
can set up any execution pattern we think might be required (within
reason). So, I think it still has a use even if we need to remind
ourselves of the overlap.
-Chris
Fair enough. With the above written down as a comment in the code:
Reviewed-by: Daniele Ceraolo Spurio <daniele.ceraolospurio@xxxxxxxxx>
Daniele
_______________________________________________
Intel-gfx mailing list
Intel-gfx@xxxxxxxxxxxxxxxxxxxxx
https://lists.freedesktop.org/mailman/listinfo/intel-gfx
