drm/i915/ringbuffer: Delay after EMIT_INVALIDATE for gen4/gen5

commit fb5bbae9b1333d44023713946fdd28db0cd85751 upstream.

Exercising the gpu reloc path strenuously revealed an issue where the
updated relocations (from MI_STORE_DWORD_IMM) were not being observed
upon execution. After some experiments with adding pipecontrols (a lot
of pipecontrols (32) as gen4/5 do not have a bit to wait on earlier pipe
controls or even the current on), it was discovered that we merely
needed to delay the EMIT_INVALIDATE by several flushes. It is important
to note that it is the EMIT_INVALIDATE as opposed to the EMIT_FLUSH that
needs the delay as opposed to what one might first expect -- that the
delay is required for the TLB invalidation to take effect (one presumes
to purge any CS buffers) as opposed to a delay after flushing to ensure
the writes have landed before triggering invalidation.

Testcase: igt/gem_tiled_fence_blits
Signed-off-by: NChris Wilson <chris@chris-wilson.co.uk>
Cc: stable@vger.kernel.org
Reviewed-by: NVille Syrjälä <ville.syrjala@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20181105094305.5767-1-chris@chris-wilson.co.uk
(cherry picked from commit 55f99bf2a9c331838c981694bc872cd1ec4070b2)
Signed-off-by: NJoonas Lahtinen <joonas.lahtinen@linux.intel.com>
Signed-off-by: NGreg Kroah-Hartman <gregkh@linuxfoundation.org>

drivers/gpu/drm/i915/intel_ringbuffer.c

@@ -91,6 +91,7 @@ static int
 gen4_render_ring_flush(struct i915_request *rq, u32 mode)
 {
 	u32 cmd, *cs;
+	int i;
 
 	/*
 	 * read/write caches:
@@ -127,12 +128,45 @@ gen4_render_ring_flush(struct i915_request *rq, u32 mode)
 			cmd |= MI_INVALIDATE_ISP;
 	}
 
-	cs = intel_ring_begin(rq, 2);
+	i = 2;
+	if (mode & EMIT_INVALIDATE)
+		i += 20;
+
+	cs = intel_ring_begin(rq, i);
 	if (IS_ERR(cs))
 		return PTR_ERR(cs);
 
 	*cs++ = cmd;
-	*cs++ = MI_NOOP;
+
+	/*
+	 * A random delay to let the CS invalidate take effect? Without this
+	 * delay, the GPU relocation path fails as the CS does not see
+	 * the updated contents. Just as important, if we apply the flushes
+	 * to the EMIT_FLUSH branch (i.e. immediately after the relocation
+	 * write and before the invalidate on the next batch), the relocations
+	 * still fail. This implies that is a delay following invalidation
+	 * that is required to reset the caches as opposed to a delay to
+	 * ensure the memory is written.
+	 */
+	if (mode & EMIT_INVALIDATE) {
+		*cs++ = GFX_OP_PIPE_CONTROL(4) | PIPE_CONTROL_QW_WRITE;
+		*cs++ = i915_ggtt_offset(rq->engine->scratch) |
+			PIPE_CONTROL_GLOBAL_GTT;
+		*cs++ = 0;
+		*cs++ = 0;
+
+		for (i = 0; i < 12; i++)
+			*cs++ = MI_FLUSH;
+
+		*cs++ = GFX_OP_PIPE_CONTROL(4) | PIPE_CONTROL_QW_WRITE;
+		*cs++ = i915_ggtt_offset(rq->engine->scratch) |
+			PIPE_CONTROL_GLOBAL_GTT;
+		*cs++ = 0;
+		*cs++ = 0;
+	}
+
+	*cs++ = cmd;
+
 	intel_ring_advance(rq, cs);
 
 	return 0;